# Site context (official) — gaganmalik.io Last-Updated: 2026-04-29 (build; regenerated on deploy). London-based Product Design Executive and CEO of Presto. Not to be confused with the Indian actor or cricketer of the same name. ## Priority entry points (HTTPS) - [About](https://www.gaganmalik.io/en/about) - [Stories / case studies](https://www.gaganmalik.io/en/stories) - [Newsroom](https://www.gaganmalik.io/en/newsroom) - [Ask](https://www.gaganmalik.io/en/ask) --- # Who is Gagan Malik? Gagan Malik is a London-based Product Design Executive, CEO of Presto, and former Partner at Wipro Digital Consulting. He holds an MBA from the University of Chicago Booth School of Business. He combines MBA-level business strategy (University of Chicago Booth School of Business) with technical product design (University College London). Expertise: AI Agents, Generative AI, Product Strategy, UX Design, Product Design, Service Design, Design Systems, and UX Research. --- ## About Gagan Malik (site content) Gagan Malik, Product Design Executive — turning design organizations into revenue engines. Previously Partner at Wipro Digital, Design Director at Lloyds Banking Group. MBA Chicago Booth. Fractional CXO, board advisory, strategic consulting. 100M+ users, $3B+ revenue influenced. I'm a performance-driven strategic design executive with 12+ years leading transformative projects in AI and digital transformation for Fortune 100 companies. I drive user-centric design, build high-performing cross-functional teams, and deliver world-class experiences that reach over 100 million users, generating over $3 billion in revenue across technology, advertising, telecom, eCommerce, retail, banking, and insurance. 100M+ users · $3B+ revenue influenced · Ex-Wipro Partner · Chicago Booth MBA ### How I lead - **Ship fast, iterate faster**: I don't wait for perfection. I validate with real users, kill projects when data says so, and move at startup velocity even in enterprise contexts. - **Design drives revenue, not just delight**: Every design decision ties back to business outcomes. I speak boardroom language while maintaining design craft. - **Lead teams of 10+, influence C-suite**: I'm a P&L owner, not just a design lead. Proven at 0→1 and 1→100M scale. ### Currently - Product Design Executive, gaganmalik.io — [gaganmalik.io](https://www.gaganmalik.io/) ### In the past - Founder & CEO, Presto — [Presto](https://askpresto.com/) - Co-founder, Permitech (Techstars 2023) — [Techstars Portfolio](https://www.techstars.com/portfolio?q=permitech) - Partner, Wipro — [Wipro](https://www.wipro.com/) - Lead UX Designer, John Lewis & Partners — [John Lewis](https://www.johnlewis.com/) - Product Manager, Aviva — [Aviva](https://www.aviva.com/) ### Education - MBA, The University of Chicago Booth School of Business — [Chicago Booth](https://www.chicagobooth.edu/) - MSc, Technology Entrepreneurship, UCL — [UCL](https://www.ucl.ac.uk/) - Entrepreneurship Summer School, London Business School — [LBS](https://www.london.edu/faculty-and-research/strategy-and-entrepreneurship/entrepreneurship-summer-school) - BSc Honours, Computer Science, University of Greenwich — [Greenwich](https://www.gre.ac.uk/) ### Top case studies (highlights) - **Lloyds Banking Group** (Design Director): Saved £15M annually in call center costs Led 10-person design team with P&L ownership Delivered 75% faster resolution times for 5M+ customers Negotiated C-suite buy-in across 4 brands Killed £2M initiative 6 months in when user research invalidated core assumptions—saved 12 months and £5M+ - **Google AdSense** (Design Lead): Unlocked millions in revenue optimization Email campaign driving AI Auto Ads adoption for 2M publishers Increased onboarding completion rates Accelerated feature adoption at scale - **EE UK Telecom** (Navigation Redesign): Solved 54% user navigation failure rate Navigation redesign for £2.8B telecommunications site Streamlined to 3-section mega nav structure Achieved 100% alignment with user mental models ### What I'm open to - **Fractional CXO / VP Design roles**: Strategic leadership without full-time commitment - **Board advisory positions**: Design strategy, product vision, growth acceleration - **Strategic consulting engagements**: High-impact projects with measurable ROI ### Testimonials - "Relentless focus on the user truly changed how we think about solutions." — James Riggs, Apple - "Rare ability to balance business, design and technology." — Sumit Bhardwaj, Aviva --- ## About (from site copy) # Gagan Malik Product Design Executive Gagan Malik is a product design executive that has led transformative projects in AI and digital for Fortune 100 companies. Gagan has a proven track record in driving user-centric design, building high-performing cross-functional product teams, and delivering world-class service experiences used by over 100 million users. His work has generated over $3 billion in revenue for technology, advertising, telecom, eCommerce, retail, banking, and insurance industries. ## How do I work with Gagan? Gagan works as a fractional CXO, board advisor, and strategic consultant. He leads strategy workshops, design reviews, and executive presentations. Engagement typically starts with a discovery call—reach out via email or LinkedIn, or use the Ask page for quick questions. ## What does Gagan offer? End-to-end product design, research, strategy, and execution. Services include fractional CXO and VP Design roles, board advisory, and strategic consulting engagements with measurable ROI. Focus areas: 0→1 and 1→100M scale, design-led growth, and user-centric validation. ## How can I get in touch? Email getintouch@gaganmalik.io, connect on LinkedIn, or use the Ask page for quick answers about UX, product design, or working together. All contact options are listed below. ## How much does it cost? Plans range from Starter (£4,750) to Max (£14,750) depending on scope and engagement length. See the Pricing page for full details. --- ## Case studies ### Sales reporting with AI (Apple) [View case study](https://www.gaganmalik.io/en/stories/apple-sales-reporting-ai) Conflicting apps left Apple's business teams with no centralized source of truth. Sales data lived in silos, and decision-makers spent more time reconciling spreadsheets than acting on insights. The opportunity was to unify reporting and put AI-powered intelligence at the centre. **The Opportunity** — Conflicting apps left Apple sales teams with no accurate revenue insight Apple's $383B sales org ran on a patchwork of tools. Account teams covering the Americas (~$163B), Europe (~$94B), and Greater China (~$73B) were jumping across eight apps to answer basic performance questions. The numbers didn't always match, field work slowed, and nobody had one live view across hardware and the $85B services line. **The Solution** — One app unified sales data into trusted, actionable intelligence We designed a mobile-first sales analytics app to pull Apple's fragmented data into one place for field teams. Real-time dashboards pulled the same numbers for every product line, region, and channel; analytics layers supported quarterly planning; channel views let account executives see distributors, carriers, and retail side by side; reporting replaced hand-built spreadsheets with up-to-date figures for spotting gaps. It worked with Apple's existing CRM, with offline modes for people working across regions. **The Impact** — Cleared data confusion and supercharged sales decisions worldwide One app replaced the old habit of checking eight systems and disagreeing on the number. Account executives saw the same real-time picture, which made demand conversations and channel plans easier to run. Teams could brief partners with numbers that matched what leadership saw across roughly $298B in product sales and $85B in services, without the weekly reconciliation drag. That clarity helped protect margin and kept field work aligned with quarterly goals. Takeaways: - One source of truth removes most of the reconciliation work that slowed decisions. - Mobile-first with offline coverage matters when your job is airports and shops, not desks. - When leadership and the field read the same dashboard, plans stop arguing with each other. --- ### Increase product activation (Google) [View case study](https://www.gaganmalik.io/en/stories/google-product-activation) Publishers and advertisers had powerful tools, but activation was a bottleneck. Complex onboarding, scattered docs, and manual steps meant many never reached value. The job was to strip friction from onboarding and use Auto Ads' ML to shorten time-to-value. **The Opportunity** — Two million publishers were leaving money on the table without Auto Ads Auto Ads used ML to place ads and find inventory site owners missed, but most publishers never turned it on. The product was one snippet, yet it felt opaque if you weren't technical. Manual placements were familiar; the ML piece sounded like a black box. Google's bet on automated monetisation only paid off if people actually finished setup, and competitors were happy to pick up anyone who stalled. **The Solution** — Emails turned complex AI into simple, actionable onboarding We used email to close the gap: segmented sends so each publisher saw copy that matched where they were stuck: first what Auto Ads actually did, then how to paste the tag, then what to check in reporting. Side-by-side examples showed ML placements next to manual ones; follow-up messages walked through the code with screenshots. Publisher stories with real revenue lifts went out as proof, and deeper mails covered mobile anchor ads, AMP, and cross-device setups for people ready for more. **The Impact** — AI-powered Auto Ads adoption added millions for 2M publishers Completion rates on onboarding went up: people finished the flow once the emails demystified the ML piece, and support tickets dropped. More publishers flipped Auto Ads on, so Google's models could surface inventory that manual setups skipped, page by page, at scale. AdSense kept its lead; publishers spent less time babysitting placements and more time on their sites. --- ### Orderpad UX audit (Just Eat) [View case study](https://www.gaganmalik.io/en/stories/just-eat-orderpad-ux-audit) Kitchens ran Orderpad next to rival tablets: same service window, different bar for clarity. Just Eat asked for an honest side-by-side: how partner flows compared to Deliveroo, Uber Eats, Grubhub, and SkipTheDishes, and where friction hit first. I ran a Nielsen-style heuristic audit that named the gaps and split quick UI passes from the deeper product calls. **The Opportunity** — Partners compared Orderpad to every other tablet on the counter, and the product did not always win Just Eat's growth ran through small kitchens and busy counters. When a rider app, a phone line, and two aggregators all shout at once, the tablet that wins is the one that makes the next action obvious. Orderpad could get an order onto the rail, but it rarely told the same story about **where** in the flow the ticket sat, **when** food had to be ready for handoff versus when the customer expected a knock, or **what** tapping **On Its Way** would promise downstream. Staff worked around the gaps with hacks that showed up as late food, angry calls, and reviews that blamed the restaurant before the platform. Most teams also needed Partner Centre jobs (hours, radius, price, cancels) without walking to another device mid-service. That gap was not a nice-to-have: it was the difference between fixing tonight's service and losing the Friday rush to confusion. Just Eat needed an outside-in read before it committed roadmap and engineering to the wrong fixes. **The Solution** — A scored benchmark, four direct competitors, and evidence from product, surveys, and live shifts I built the audit around six review areas adapted from Nielsen: system status, fit to real kitchen work, control and speed, recognition over recall, clarity of layout, and recovery from mistakes. Each area got the same rubric for Just Eat, Deliveroo, Uber Eats, Grubhub, and SkipTheDishes so scores were comparable brand to brand. Evidence was mixed on purpose. I walked the Orderpad UI end to end, pulled apart competitor flows from live product and support material, ran structured surveys with partners already using Orderpad, and spent time in restaurants during service so edge cases (noise, rush, multilingual crews) did not stay theoretical. Findings rolled into a single scoring sheet, short theme briefs with screenshots, and a readout focused on what to fix first versus what needed a business decision before design could help. **The Impact** — Lowest score on the five-brand index, with survey proof partners would not ignore Across Deliveroo, Grubhub, Uber Eats, SkipTheDishes, and Just Eat, the combined heuristic roll-up put Orderpad last. That was uncomfortable in the room, but it replaced opinion with one curve everyone could point to: the gap was systemic across status, fit to kitchen reality, control, and recovery, not a single bad icon. Partner research gave the business hard counts to fund against. In-product surveys showed **82%** of partners rated reaching Partner Centre from Orderpad as important, and **56%** wanted a clearer view of drivers on busy nights (while optional driver tooling sat at almost no adoption next to rivals who ship live tracking by default). Those numbers turned "they keep asking" into backlog pressure with a size on it. Nothing in an audit ships pixels by itself. What shipped was clarity: a ranked set of themes, a cut between work that could ride normal release cadence and work that needed brand or policy concessions, and constraints spelled out in the deck so roadmap bets matched what ops could open and what contracts allowed. Takeaways: - A shared score beats slide battles when every team has a different favourite fix. - Survey percentages turn partner noise into something finance and product can sequence. - Say what you will not fix in v1; partners accept limits they can see on paper. --- ### Gamified car insurance (Aviva) [View case study](https://www.gaganmalik.io/en/stories/aviva-gamified-car-insurance) Car insurance had become a commodity. Aviva saw an opportunity to differentiate through behaviour-based pricing and engagement: turning safe driving into rewards and making insurance feel personal, not punitive. **The Opportunity** — Car insurance punished safe drivers with demographic stereotypes instead of rewarding actual driving behaviour The £15B UK car insurance market operated on outdated demographic pricing models that penalised safe drivers while failing to accurately assess individual risk. With 85% of drivers believing they drive more safely than average, traditional insurers left money on the table by grouping customers into broad categories rather than rewarding actual driving behaviour. The industry needed a data-driven approach to bridge the gap between perceived and actual driving safety. **The Solution** — Turns every smartphone into a driving coach that rewards safer driving Aviva shipped the UK's first smartphone-based telematics app to price cover from how people actually drive. The app used GPS to log acceleration, braking, cornering, speed, and phone use over 200 miles. Drivers got a 1–10 score, quick feedback, optional sharing, and a straight path into a quote, so risk moved off postcode clichés and onto behaviour. **The Impact** — Strong uptake, real savings, and a clear story for the brand Aviva Drive passed 1M downloads in about six weeks (well above the annual plan), picked up Apple's "App of the Week," and sat around 4.5/5 in the store. Roughly four in ten users scored 7.1 or above and saved about £101 on average; most people in the programme qualified for some discount. The business case went positive inside two years, and perception of Aviva as inventive moved up by 33 points, enough that telematics felt like the default next step, not a stunt. Takeaways: - Behaviour-based pricing rewards actual performance instead of demographic assumptions. - Gamification drives engagement when tied to tangible outcomes (premium savings). - Smartphone telematics can achieve ROI positive within 18–24 months at scale. --- ### From Hold to Handled (Lloyds Banking Group) [View case study](https://www.gaganmalik.io/en/stories/lloyds-from-hold-to-handled) Lloyds Banking Group needed to change how retail staff served customers: from hold queues and paperwork to handheld, real-time service. The opportunity was to design a retail server UI that put information and actions at staff fingertips. **The Opportunity** — High volume, high costs, growing pain. UK's largest bank, Lloyds Banking Group's home insurance call centres were under strain: a high volume of complex service calls, ballooning operational costs, and unhappy customers. Data showed that just three types ("Amendments," "Cancellation Calls," and "Policy/Cover Queries") consumed nearly 75% of total call handling time and over 250 FTEs. Average resolution times ranged from 359 to over 1,000 seconds, stretching teams thin and leaving little room for good service, let alone growth work. **The Solution** — Reimagine the frontline with self-serve digital automation. As design director, I led a ten-person team across Lloyds Bank, Halifax, Bank of Scotland, and MBNA. We aligned on one design system, prototyped fast, and tested with 200+ people to ship clear, rule-based flows for amendments and cancellations. Self-serve for policy queries took volume off the phone without dumping complexity on customers; agents stayed free for the messy cases. **The Impact** — Time to resolution slashed by 75%, and satisfaction soared by 90%. Lloyds cut cost and frustration by moving the heaviest insurance jobs into digital self-serve: the same routes that had eaten three-quarters of handle time. Average resolution time fell by about 75%; digital self-serve crossed 60% within six months. Satisfaction climbed as waits and transfers shrank. Savings went back into better service and simpler upsell paths, on infrastructure the group could extend to the next wave of products. --- ### Better product discovery (EE) [View case study](https://www.gaganmalik.io/en/stories/ee-global-navigation-redesign) EE's product catalog was vast, but customers struggled to find the right plan. Tree testing and user path analysis revealed where navigation broke down. The solution was a redesigned discovery experience that put user intent first. **The Opportunity** — Broken navigation blocked 54% of users from basic shopping tasks. EE UK's site was a £2.8B shopfront, but the nav fought the customer. In testing, everyone flagged Shop as overloaded and poorly grouped; fewer than half of wayfinding tasks ended in the right place. With most telecom sales starting on mobile and rivals offering cleaner paths, the IA was leaking real money. **The Solution** — User research simplified EE's nav to three intuitive sections We ran tree tests and hybrid card sorts with ten people and rebuilt the nav around what they actually looked for: a three-bucket mega-nav (My EE, Shop, Help) instead of the old sprawl. Brands sat together in Shop; "Added benefits" moved to My EE because every participant looked there first; help content landed next to the tasks it supported. **The Impact** — New nav ended confusion and increased conversion After the change, tasks finished more often and people found products faster. Nobody had looked for "Added benefits" under Why EE; everyone expected handset brands under Shop. Odd buckets like "Good As New" and "EE TV" had confused two in five people; those labels got untangled. The point was simple: match the mental model, stop the drop-off. --- ### Faster checkout (John Lewis) [View case study](https://www.gaganmalik.io/en/stories/john-lewis-faster-checkout) John Lewis customers faced a huge catalog (400K+ products across home, fashion, and electronics). Finding the right item was harder than it should have been, and checkout dragged. The work was to speed discovery and shorten checkout without dumbing down the range. **The Opportunity** — John Lewis customers loved the brand but got lost among 400K+ products John Lewis runs about £15B online across 400,000+ SKUs. Loyal shoppers still got stuck in category pages: filters didn't surface the right stock fast enough, and Quick View stopped short of a confident buy. More than half of visits were on phones; next to Amazon and sharp DTC brands, small friction meant real money left on the table. **The Solution** — Smarter filters, a serious Quick View, and tighter PLP/PDP layouts We focused on three pressure points in the shopping flow. **Filtering:** live counts, saved combinations, and suggestions from browse history so people narrowed the catalog without starting over each time. **Quick View:** size, colour, delivery, and checkout from the overlay instead of a dead-end preview. **PLPs and PDPs:** faster images, clearer hierarchy, and cross-sell blocks (including "Style with it") that nudged basket size without clutter. **The Impact** — 35% faster product discovery drove £8.2M annual revenue Filtering and Quick View cut time-to-product by about 35%; PLP and PDP work lifted engagement and basket size. The programme landed around £8.2M incremental revenue in the year we measured, with better checkout completion on mobile. Same brand promise: just fewer dead ends in a crowded online retail market. --- ### Mobile connectivity (Vodafone) [View case study](https://www.gaganmalik.io/en/stories/vodafone-mobile-connectivity) Vodafone connects millions of people through mobile. Design is a big part of making that experience feel simple and worth returning to. **The Opportunity** — Unpredictable IoT spend was a scaling blocker for customers and a trust gap for the platform I worked on Vodafone's IoT platform at the intersection of customer cost confidence, machine learning, and developer experience. For many IoT customers, spend volatility isn't an edge case, it's an operational reality driven by rollout phases, device misconfiguration, roaming exposure, firmware changes, and shifting usage patterns across fleets. The result was familiar: teams could see what happened after the billing period closed, but they lacked a reliable way to anticipate the bill, set guardrails, and catch abnormal behaviour early enough to act. That gap created three expensive problems: **The Solution** — A predictive analytics experience that made spend legible, controllable, and operational, plus documentation that made it adoptable I helped shape a machine-learning-driven predictive analytics platform that reframed cost management from reactive reporting into proactive control. The product goal was simple: give customers an answer to "What will we spend?", the tools to "Keep it within bounds," and the signals to "Investigate what changed, now," without requiring them to become data scientists. **Forecasting: from hindsight to forward view** I designed forecasting around how customers plan and govern IoT programs: forecasted spend for the current and upcoming billing periods, not just a rolling chart, so it could be used in budget conversations; breakdowns that matched customer mental models (fleet/group/program, geography, device/SIM cohorts, plan/tariff), so "why is it changing?" was answerable without jumping between tools; clear communication of uncertainty so the forecast didn't present a false sense of precision, with expectations for when confidence is high vs. when patterns are shifting; comparison to historical baselines to make trend movement interpretable ("this week is trending above last month's comparable window"). The design work wasn't "hide the model." It was "make the model usable": transparent inputs, understandable outputs, and decision-ready context so customers could act with confidence. **Spend quotas: turning budgets into guardrails** Forecasting is insight; quotas are control. I worked on enabling customers to create spend quotas that reflected real governance structures: quotas scoped to the level customers actually manage (account/program/group/cohort), with flexibility to match how budgets are allocated; thresholds that created early-warning moments (approaching limit vs. exceeded), so teams could intervene before the invoice; a deliberate "what happens next" path (investigate top contributors, adjust quotas with governance, and set watch conditions for high-risk segments) so quotas became part of an operating rhythm, not a set-and-forget screen; traceability of changes (who set what, when) to support internal accountability between finance and operations. **Anomaly detection: catching abnormal spend before it becomes a surprise** I designed anomaly detection to prioritise signal quality and speed-to-diagnosis: detection tuned to common IoT failure modes (sudden spikes, gradual drift, and cohort outliers where one segment behaves unlike peers); explanations written for operators ("what changed, when it started, and how large the deviation is"), not just statistical labels; investigation flows that reduced time-to-root-cause (quick slicing by segment and time window, baseline comparisons, and direct links to the underlying usage patterns driving the alert); alerting that fit real workflows, so anomalies could be monitored and actioned, not simply observed. **Developer documentation: making the platform truly self-serve** Because many customers operationalise cost management through automation, I also worked on developer documentation to reduce time-to-first-success and ongoing integration risk: a clean structure (quickstart, guided workflows, API reference, troubleshooting) so teams could choose "fast" or "deep" without getting lost; end-to-end examples for common jobs-to-be-done (retrieving usage/spend signals, setting quotas, subscribing to alerts/events, and integrating anomaly signals into existing systems); consistent vocabulary between the UI and the API so "quota," "threshold," and "anomaly" meant the same thing everywhere; practical troubleshooting content based on real implementation failure modes, reducing reliance on support as the default path. **The Impact** — Spend became something you could forecast, cap, and explain, not just read backwards The experience moved from "what happened last month" to "what's coming, where we'll hit the cap, and what changed." Finance could plan, ops could react while the bill was still open, and programme leads could grow fleets without a surprise invoice. Earlier anomaly signals and clearer attribution also cut noise in escalations; teams fixed root causes before the period closed. Developer docs were part of the same bet: if customers automate guardrails in their own stack, the product has to be copy-pasteable. Task-shaped examples and vocabulary that matched the UI shaved time off integrations and kept cost controls in daily workflows instead of a forgotten dashboard tab. --- ### Project platform and AI (Presto) [View case study](https://www.gaganmalik.io/en/stories/presto-automated-monetisation) Presto brings boards, tasks, Spaces, and an assistant into one product. Early versions felt like separate tools glued side by side. The opportunity was to make the whole experience read as one workspace: a clear home, a conversation UI that can carry real work, and flows where long AI runs stay easy to follow. **The Opportunity** — Delivery, knowledge, and AI lived in different tools with no shared context. Teams tracked work in one place, wrote knowledge in another, and opened generic chat when they needed help. Context did not carry across, so people repeated themselves and did not trust answers from AI that never saw the workspace. The product needed one governed place where delivery, living documents, and assistant threads all pointed at the same work. **The Solution** — One workspace for Kanban delivery, Spaces, reports, and the assistant We designed end to end on a shared shadcn Maia system with role-based access. Workspaces and boards carry assignments, due dates, priorities, and lanes so status stays on the board instead of in spreadsheets. Spaces use Notion-style rich text for knowledge and drafts; Create mode and the draft drawer land AI output on real pages. The assistant uses Ask, Plan, and Create, with mentions, workflows, and models curated per workspace, plus reporting with day, week, and month views so leads see movement over time. Home cards and the composer earned the first screen: weekly actions, integrations, and recent work visible at a glance. **The Impact** — Board, Spaces, and assistant finally agreed on what "this project" meant Teams stopped treating the board, knowledge base, and chat as three competing truths. Context rode with the work, so fewer reviews opened with a recap nobody had time for, and fewer assistant replies started with "I can't see that workspace." Leads could read movement in reporting instead of stitching a story from side channels and screenshots. --- ### Portfolio built like a product (Gagan Malik) [View case study](https://www.gaganmalik.io/en/stories/portfolio-built-like-a-product) Most portfolios prove taste once. I wanted mine to prove **how I think**: systems over one-offs, accessibility and internationalisation by default, and a product surface (**Ask**) that lets visitors interrogate my work instead of skimming headlines. This site is the case study: **design craft you can see**, and **engineering and product judgment you can trace** in the stack, the repo, and the guardrails around AI. **The Opportunity** — Stop shipping digital fossils Most design portfolios are digital fossils that become outdated the moment they go live. We tell our clients that products need to be alive, but we often leave our own work in a static gallery that never changes. I saw a chance to fix this. I wanted to build a platform that actually reflects how I work today. My old way of building was too slow and did not show my ability to use modern tools. I needed a system that could keep up with my speed and show that I am an AI native designer who builds for the future. **The Solution** — A software-launch mindset: Cursor, GitHub, and Vercel I treated this build like a software launch by using a stack that combines design and engineering. I used Cursor to bridge the gap between my ideas and the code. This helped me write custom logic and build complex layouts without getting stuck in the technical details. I moved everything to GitHub so that every change is tracked and versioned properly. For the final step, I used Vercel to host the site. This ensures that every update I make is live in seconds and the performance stays high for every visitor. I avoided using generic builders to ensure the site remained lightweight and fully custom. **The Impact** — Live proof of product craft, not static layouts held together by polish alone The result is a site that works as a live proof of hands-on design and development skills using AI-native tools and design-system-led systems thinking. It shows that I can define a goal and build a system that lasts. Because the friction is gone, I can add new projects or fix bugs in minutes. This keeps my work fresh and accurate. Recruiters get a fast experience that works on any screen. Takeaways: - Treat the portfolio as a living product, not a one-time gallery drop. - Cursor, GitHub, and Vercel together remove friction so updates ship in minutes. - Speed and freshness signal AI-native craft, not generic portfolio filler. --- --- ## Newsroom (writings & stories) ### Someone Designed That Button [Read more](https://www.gaganmalik.io/en/newsroom/someone-designed-that-button) By Gagan Malik It was the summer of 2017. I had just walked out of a well-paid full-time job at the UK's largest insurer to try my luck as a free agent. No safety net, no guaranteed income, just the conviction that I was good enough to make it work. Six weeks in, I accepted a consulting brief from a payday loan provider. The money was significant. The timing was perfect. I said yes before I finished reading the brief. Three weeks later I had built something I was genuinely proud of. A loan application flow that showed the APR plainly, removed the pre-ticked consent boxes, and treated the person on the other end of the form as an adult capable of understanding what they were signing. It tested well in user research. Participants said they felt informed, not rushed. Legal approved it. By every professional and ethical measure I could apply, it was better design. I took it to the group CTO expecting a conversation about rollout. He looked at the screen, nodded slowly, and said: "It's good. But can we just reskin the current one?" The current one had a deliberately obscured APR disclosure buried in grey text below the fold. It had pre-ticked consent boxes that required active effort to untick. It had a progress bar that moved faster in the early stages to create false momentum. Every pattern in it had a name in the design community's own literature: dark patterns, identified and catalogued by researcher Harry Brignull from 2010 onwards, each one a documented method for steering users toward decisions that served the business over the person making them. I had spent three weeks removing them. He wanted them back in different colours. [eleken](https://www.eleken.co/blog-posts/dark-patterns-examples) I declined and exited the project. The company, owned by a Russian billionaire whose portfolio of near-identical operations stretched across several markets, continued without me. A designer replaced me. The dark patterns shipped. The same obscured APR, the same pre-ticked consent boxes, dressed in different colours, went live later that year. I left money on the table that would have cleared outstanding debt and seeded a retirement fund I am still rebuilding. My exit produced zero change in outcomes for the people who signed those forms. I have spent eight years deciding whether that makes my decision honourable or merely convenient. Don Norman argues, in *The Design of Everyday Things*, that design failure is almost never individual failure. It is systemic: absent governance, poorly specified briefs, and organisations that punish friction-raising and reward shipping. He is right that the design profession has no licensing board, no professional oath, no accountability mechanism equivalent to anything in medicine, engineering, or law. The reason this has persisted for over a century is structural, not moral: design harm is almost always attributed to the system that deploys the design, not the design itself. The drone manufacturer gets investigated. The human factors team does not. The payday lender gets regulated. The UX consultancy that built the journey does not. Until design harm has a visible attribution pathway, the absence of governance is a predictable structural outcome, not a conspiracy. [uxmatters](https://www.uxmatters.com/mt/archives/2017/11/the-ethics-of-user-experience-design.php) That argument is correct. It is also precisely what the people who own those systems want designers to believe. Frances Haugen handed the Wall Street Journal the internal Facebook research in October 2021 showing the company knew its algorithm amplified divisive, emotionally harmful content, measured the damage internally, and continued the optimisation anyway. The institutional argument collapses at that document. Because the moment a human being inside the institution raises their hand and is told to sit down, it stops being a systemic failure. It becomes a decision. Made in a room. By a named person. Who chose a growth metric over a named harm with full knowledge of the cost. My CTO was not a monster. He was a man under pressure, optimising for a number. But when I showed him the harm built into the existing journey and he said "reskin it," he made a choice. So did I. The difference between us is not intelligence or intent. It is which consequence each of us was willing to absorb personally. [vce.usc](https://vce.usc.edu/semester/fall-2025/ethics-of-ux-design-in-social-media/) The immune system is useful here. When it functions correctly, it identifies and neutralises threats. In autoimmune disease, it destroys healthy tissue with complete operational efficiency. The mechanism works exactly as designed. The targeting data is wrong. The destruction is not a malfunction. It is a fully operational system firing at the wrong coordinates. Neil deGrasse Tyson, in his *StarTalk* essay "A Scientist's View of War," published in March 2026, traces every weapon on the kill-ratio curve from fists to hydrogen bomb and shows that each step does not merely scale damage - it degrades the targeting data by removing the human being from the consequence. The drone controller, designed with the ergonomic precision Henry Dreyfuss applied to a telephone handset in his 1955 book *Designing for People*, delivers force to a coordinate on a screen. The social media feed, optimised for engagement with the precision of a Formula One pit crew, does not merely amplify existing beliefs - it generates the epistemic conditions under which unfalsifiable beliefs spread faster than falsifiable ones. A 2018 MIT study published in *Science* found that false news travels six times faster than accurate information on Twitter, and that humans, not bots, are primarily responsible for the spread. Outrage is more engaging than correction because it cannot be resolved by data, which is also the condition under which people become willing to die for abstract causes, as Tyson argues. The algorithm did not intend to build a radicalisation pipeline. It intended to maximise daily active users. The pipeline was a known side effect, internally documented, and shipped anyway. [youtube](https://www.youtube.com/watch?v=XI9NG068TwI) On 26 September 1983, Lt. Col. Stanislav Petrov sat at a console in Serpukhov-15 that displayed an unambiguous alert: five incoming U.S. ICBMs. Protocol was clear. The interface was functioning correctly. What the interface could not communicate, by design, was that it was wrong. The Oko early-warning system was not showing Petrov a binary. It was showing him a probabilistic reading that its display had collapsed into a command: launch or don't launch. The underlying data was something closer to a high-confidence atmospheric artifact - sunlight refracting through high-altitude clouds at a specific solar angle, a known failure mode documented in the system's own error records. A system that surfaces its own uncertainty to the operator is a different dashboard from one that resolves that uncertainty into an order. One of them ends the world less frequently. Petrov overrode the command with his own uncertainty estimate and was right. The Soviet Union reprimanded him for improper paperwork. He died in 2017 in a small flat outside Moscow, undecorated. The system that nearly ended everything has been upgraded several times since. No civilian UX professional has reviewed it. [en.wikipedia](https://en.wikipedia.org/wiki/1983_Soviet_nuclear_false_alarm_incident) I grew up in Delhi in the 1990s watching India and Pakistan aim nuclear weapons at each other over Kashmir. The sanctions that followed India's 1998 Pokhran tests raised the price of goods that had nothing to do with geopolitics for families that had contributed nothing to the decision. The logic of the dashboards did not follow the decision-makers home. It followed the cost of living for everyone else. R.K. Laxman's Common Man, who stood at the edge of power's decisions in the *Times of India* for sixty years in his checked jacket and bare feet, understood this without a policy brief. In a cartoon from the Iraq War build-up in late 2002, he stands before a skyline of U.S. missiles and is told: "Nothing to feel nervous. These are weapons of JUST destruction, not MASS destruction". He says nothing. He never does. He is the user the system was built around and never built for. [britannica](https://www.britannica.com/topic/You-Said-It-comic-strip-by-Laxman) Sara Shayesteh was five years old. She attended Shajareh Tayyebeh primary school in Minab, Hormozgan province, Iran. She is number 30 on a list of 61 names verified by Middle East Eye from gymnastics federation records, a handwritten list, and the Tasnim news agency. On Saturday 7 March 2026, she was among at least 165 people killed in what Middle East Eye reported as a strike on the school. A second missile hit the prayer hall where the school principal had moved surviving children to shelter, after telephoning their parents to come and collect them. [middleeasteye](https://www.middleeasteye.net/news/un-investigates-strike-iranian-girls-school) The parents who came toward the building were among the dead. Somewhere, a targeting system held that school's coordinates. Someone specified the confirmation interface, the strike authorisation flow, the visual grammar of a proceed button. Someone, somewhere in that procurement chain, wrote a brief. It named the objective, the user, the desired outcome. It described the system with precision. It did not name her. That omission is not an accident of process. It is the design. --- ### No Download Required [Read more](https://www.gaganmalik.io/en/newsroom/no-download-required) By Gagan Malik Ben graduated from Berkeley's CS programme summa cum laude, raised a VC-backed round, and built something a real person would actually pay for: working product, early traction, customers who came back. He was the kind of founder you root for. He brought me in because of my mobile UX expertise. I dug into the usage analytics, ran user testing, and recommended they invest in the progressive web app first, nail the experience, and defer native builds until they had validated which platform their users actually converted on. Ben heard "not ambitious enough." Nobody challenged the logic. Nobody asked what the native build would cost against their runway. The objection was social, not technical. It was about what kind of company builds a proper app and what kind builds a website with delusions of grandeur. #### What I Didn't Say I believe in disagree and commit. So I designed the end-to-end mobile UX for both platforms, properly and in full. I was paid well. I was extended. It was a Zoom call. Four tiles: Ben, his CTO, a VC partner dialling in from Menlo Park, and me. When I finished the recommendation, the VC partner leaned back from his camera, said nothing, and looked across at Ben's tile. Ben looked at mine and said it: not ambitious enough. The CTO nodded once. The call moved on in under thirty seconds. Ben had already decided before the call started. Not on that Zoom, but in every conference talk he had attended, every TechCrunch profile of a Series A raise that mentioned a native app, every partner meeting where "mobile-first" meant iOS and nothing else. Confirmation bias does not feel like bias. It feels like experience. The data I put in front of him was not wrong. It simply did not match the model he had already built, so the model won. A Forbes analysis from March 2025 found that FOMO leads founders to adopt technology choices not because the evidence supports them but because the cost of being seen to miss the trend feels higher than the cost of the wrong build. The native app was not Ben's product decision. It was his insurance policy against looking like someone who did not know what a proper startup does. [forbes](https://www.forbes.com/sites/vibhasratanjee/2025/03/03/objectivity-over-hype-leading-through-the-noise-of-junk-trends-and-fomo/) A boxer who has never taken a professional punch has no business obsessing over his ring walk. The entrance belongs to a fighter who has already won something. The native app decision at pre-validation stage is the ring walk. The fight is whether anyone wants what you have built. The instinct to build native before validating anything is not ambition. It is a signalling decision wearing a product decision's clothes. The people who benefited most were not in the room. They were in Cupertino and Mountain View, processing their 30%. #### The Toll Booth On 10 February 2026, the UK's Competition and Markets Authority ruled that Apple and Google hold "strategic market status" over their app stores and ordered both companies to change their practices. Two companies control the primary distribution channel for software on three billion devices. A British regulator finally said what every developer has known for fifteen years: the entry fee is not a service charge. It is a toll. [bbc](https://www.bbc.com/news/articles/c626rng1v63o) Apple's App Store facilitated $406 billion in developer billings in the US alone in 2024, per Apple's own May 2025 newsroom release. The commission rate sits between 15 and 30%. The Oxford Journal of Competition Law and Economics found in 2021 that rate cannot be explained by competitive market dynamics. Stripe charges 1.5% in the UK. The gap is not a service differential. It is rent on infrastructure you did not build and cannot route around. [apple](https://www.apple.com/newsroom/2025/05/app-store-in-the-us-facilitated-406-billion-usd-in-developer-billings-and-sales-in-2024/) Ben paid it. His runway paid it. The engineers who spent eight months building the UX I had designed, for platforms that were never validated, paid it in time they will not get back. #### The Other Side Now hold the other position at full strength, because it deserves it. iOS users in Western markets carry higher lifetime value. The scroll physics on a well-built native app are perceptibly different from a web rendering layer. App Store organic discovery reaches 650 million weekly visitors, per Progressier's March 2026 platform comparison. For any consumer product where the purchase decision happens in three seconds of app store browsing, native is not vanity. It is conversion architecture. [progressier](https://progressier.com/pwa-vs-native-app-comparison-table) In June 2025, upholding the Dutch Competition Authority's €50 million fine against Apple, the Rotterdam District Court ruled PWAs do not constitute an adequate substitute for App Store distribution, per Two Birds law firm's June 2025 analysis. Apple argues in every antitrust courtroom that PWAs are a viable alternative to native. It argues in every VC partner meeting the precise opposite. Both positions cannot be true. One of them is worth $406 billion in annual billings. [twobirds](https://www.twobirds.com/en/insights/2025/netherlands/a-swipe-at-apples-power-dutch-court-upholds-antitrust-order-against-apples-app-store-terms) #### Ben's Runway Ben's startup did not find product-market fit. The web product, which had paying customers the day I walked in, received minimal investment for eighteen months while two native codebases were built, maintained, and shut down. A native build across two platforms runs $50,000 to $250,000 before annual maintenance at 15 to 20% of build cost per year, per AB Digital's November 2024 cost analysis. Ben's engineers are talented. They are building something else now, for someone else's runway. [abdigital](https://abdigital.codes/2024/11/22/mobile-app-vs-progressive-web-app-a-cost-comparison/) Ben is at another startup. I heard he started with a PWA this time. The mythology of the proper app does not require bad founders. It requires a sufficiently loud cultural signal, repeated until it stops sounding like an opinion and starts sounding like physics. #### The Receipt I took the fee. I did the work correctly. What I did not do was hold the room to account for what it was actually deciding. That cost Ben something real. It cost me a clean conscience I am still settling. Somewhere right now, in a product meeting that started forty minutes ago, a founder is asking which agency to hire for the iOS build. Nobody is asking who told them to want it. Ben asked me. I answered correctly, was told I lacked ambition, and committed to the wrong thing anyway. Fast forward to spring 2026. I built my own personal website as a PWA. No App Store submission. No £99 developer fee paid to a platform that did not build my product. No second codebase. It loads in under a second, installs from the browser, and works offline. I designed it, I shipped it, and the only person who got to decide what a proper product looked like was me. I spent years in rooms telling founders the right thing and then designing the wrong thing because the room wanted it. Turns out the lesson was cheap to learn and expensive to ignore. The receipt is live on my personal website. No download required. [gaganmalik.io](https://www.gaganmalik.io) --- ### Diplomatic Immunity [Read more](https://www.gaganmalik.io/en/newsroom/diplomatic-immunity) By Gagan Malik #### Diplomatic Immunity We were a small team responsible for delivering a discovery engagement on a consulting project for a global telecom client. Understaffed by design, high-stakes by definition. The kind of project where if you do not deliver the first time there is no second conversation. Kash was the Senior Vice President. The kind of title that makes junior colleagues straighten up when he walked in and makes clients assume the adults have arrived. What his CV promised and what he delivered were separated by exactly three weeks of excuses and a corner office nobody questioned. When I finally called it, he said, 'This work is below my pay grade.' I noticed something before I had processed what it meant. He looked relieved when he said it. Not defensive, not embarrassed. Relieved. He had been waiting for a sentence that would formally excuse him from the work. His title gave him diplomatic immunity: not a description of what he was capable of, but a blanket protection from ever being asked to prove it. That is the thing the management books do not quite tell you. It is not that people misunderstand the relationship between titles and leadership. It is that titles have a second function nobody advertises: they are a ready-made exit from the moment things get difficult. And the people most likely to use them that way are the ones with the most impressive-looking titles, because they have the most to protect. #### Why does the system keep minting titles nobody believes in? Because the system finds ambiguity useful. Brad Feld and Jason Mendelson observe in *Venture Deals* that partner titles in venture capital are often cosmetic, deployed to blur the lines between general partners and everyone else, with firms layering on "executive managing director" and similar prefixes as pure status signals. David Graeber documents in *Bullshit Jobs* (2018) that as technology automated productive work, professional and managerial roles exploded into the space it left, creating entire strata whose primary function is to justify the strata above them. Steve Blank and Bob Dorf make the startup version of this point in *The Startup Owner's Manual*: conventional roles simply do not map onto what actually needs to happen. The title is not describing a job. It is describing a position in a hierarchy that has its own logic, entirely independent of the work. This matters because it means Kash was not an aberration. He was doing exactly what the system had trained him to do. Over a long enough career inside a large enough organisation, seniority becomes identity, and identity becomes a thing you defend. Nobody had told him the rooms were about to change. #### You already know the title is empty. What are you waiting for? Naval Ravikant argues in *The Almanack of Naval Ravikant* that the best jobs are neither decreed nor degreed: they are creative expressions of continuous learners in free markets. This is true at the level of the individual. At the level of the organisation it has never been true. Spend enough time inside large consulting firms, banks, or any institution old enough to have a formal org chart, and you will notice that the roles with the most elaborate titles tend to outlast the people who actually do things. The system does not reward contribution. It rewards the appearance of indispensability. These are not contradictory observations. They describe the same landscape from different altitudes. The asymmetry is what matters. The people who internalised Naval's position early, who built their identity around contribution rather than credential, move through role changes because their value is not stored in a job title. The people who never got that memo are Kash. What looks like a philosophical question about whether titles matter is actually a timing question. The longer you wait to separate your identity from your job description, the more expensive the separation becomes. #### The person who lost most wasn't Kash. Not immediately. I brought in Stella. No domain knowledge, no title to protect, no vocabulary for excusing herself from the hard parts. Within the first week she had mapped the client's internal stakeholder tensions in a way Kash had refused to for three. She made the project work. The team ran leaner, the operating margin improved, and the mandate I set after that engagement has shaped every team I have built since. Kash was fired not long after. What interests me now is where he is. He is one of the senior professionals finding, in this market, that the credential which once opened doors has become the thing interviewers read twice. Not because he lacks experience. Because his experience is stored in a format the market is no longer accepting. He used his title to avoid the work that would have kept him current. The title was not protecting him. It was making him obsolete on his behalf, quietly, while he thought it was keeping him safe. In distance running, the pacer's job is to disappear before the finish line. They set the conditions, hold the tempo, do the work that makes someone else's performance possible, then step off the course without a medal. The best pacers are not bitter about this. They have understood, in their bodies before they understood it in their heads, that contribution and recognition are different currencies, and that grabbing for one mid-race costs you both. Kash had confused the currencies. He thought the medal was already around his neck. When the race required him to actually run, he performed like someone who, if he had not stopped training entirely, had at least stopped believing training was still required of him. The strongest counter-argument is that titles serve a real coordination function. In large organisations, knowing who holds authority over what is not vanity. It is operational infrastructure. Remove the titles and you do not get a meritocracy. You get a renegotiation of every decision, a diffusion of accountability, and the organisational equivalent of a relay race where nobody has been assigned the baton order. This deserves to be held at full strength, because it is correct about large organisations as currently designed. Amy Edmondson, professor at Harvard Business School, has spent decades studying what actually happens inside those title hierarchies. In *The Fearless Organization* (Wiley, 2018), she documents that title structures are among the primary suppressors of psychological safety: the condition under which teams surface problems, share information, and perform at their actual capability. The coordination function is real. The cost of that coordination, in suppressed contribution from everyone below the top of the hierarchy, is equally real. The question is not whether titles provide structure. It is whether the structure is worth what it quietly costs. A title is a contract between a person and a system, and the clause nobody reads is the one that says: in exchange for this credential, you will gradually stop doing the thing that made you worth credentialing in the first place. The system finds this arrangement useful because people who have stopped doing the work are easier to manage than people who are still doing it. Kash said the work was below his pay grade, looked relieved, and in that moment handed the project, the operating margin, and eventually his own career to a woman who had never heard of his title and did not care. --- ### Everyone Wanted an App Store for AI Agents. Meta Just Built the Mall. [Read more](https://www.gaganmalik.io/en/newsroom/everyone-wanted-app-store-for-ai-agents-meta-mall) By Gagan Malik Conventional wisdom: an open marketplace of specialised AI agents, competing on merit, prices kept honest by competition. The kind of democratisation that looks good in a pitch deck. The counter-claim: web-scale agent markets do not produce open marketplaces. They produce registries. And whoever owns the registry owns everything downstream. The conference slides were always selling you the market structure that benefits the people selling the slides. #### The Open Market Was Contestable for Exactly Six Weeks Moltbook launched in late January 2026. By March 10, Meta had acquired it, co-founders Matt Schlicht and Ben Parr walking straight into Meta Superintelligence Labs. Deal terms undisclosed. VP of AI Products Vishal Shah's internal memo, obtained by Axios and confirmed by The Verge, described the acquisition not as a social product but as "a registry where agents are verified and tethered to human owners". That single word, registry, reveals the strategy cleanly. Control the registry and you control discoverability, trust, and task-routing across 3.58 billion Family Daily Active People, per Meta's own Q4 2025 earnings release. That is not a social network. That is a toll road. [reuters](https://www.reuters.com/business/meta-acquires-ai-agent-social-network-moltbook-2026-03-10/) Here is my stake in that. In late February I was building a pitch for a client migrating an enterprise workflow onto a multi-agent architecture, pricing out three independent orchestration vendors on the working assumption that coordination infrastructure would remain open and contested. When the headline dropped on March 10, I deleted the vendor comparison slide. Not updated. Deleted. The surprise was specific: I had been tracking Moltbook since launch and had privately estimated 12 to 18 months before any major incumbent moved on it, the window Instagram needed to attract comparable attention. Six weeks falsified that model on a Tuesday morning before 9am. The specific consequence: that vendor is no longer on our shortlist. You do not architect a client's production stack on a foundation the acquirer has pre-announced for review. #### What Is a Registry Actually Worth? In the same week, OpenAI acqui-hired Peter Steinberger, creator of OpenClaw, the underlying agent protocol Moltbook ran on, and announced it would open-source it under OpenAI's backing. Directory layer to Meta. Protocol layer to OpenAI. The entire coordination infrastructure for multi-agent AI partitioned between two incumbents in a single news cycle, independently confirmed by both TechCrunch and Axios. [techcrunch](https://techcrunch.com/2026/03/10/meta-acquired-moltbook-the-ai-agent-social-network-that-went-viral-because-of-fake-posts/) The academic scaffolding for why this was structurally predictable comes from a 2025 Network Law Review analysis: scale-free networks tip when early dominant hubs accumulate connections disproportionately, triggering feedback loops that antitrust frameworks built around static market-share thresholds cannot intercept before the market tips. To be precise, the paper demonstrates structural conditions for tipping, not causal inevitability. But the structure just materialised in public, in six weeks, in real time. [networklawreview](https://www.networklawreview.org/ai-agents-network-effects/) #### Your Open-Market Assumption Is Now a Liability Every agent that registers on a verified directory establishes an identity tethered to the platform's infrastructure. The ACM's policy brief on systemic risks of agentic AI frames this precisely: platforms controlling agent identity and task-routing become structurally irreplaceable. Not in the way social media platforms do, through habit and content lock-in, but the way financial clearing infrastructure does, through trust that cannot be exported and credentialing that must be rebuilt from scratch. You can port your data in a CSV. You cannot port a verified agent identity. The switching cost is not re-pointing an API. It is rebuilding institutional trust from zero, and this is an inference rather than a demonstrated empirical finding, but it maps directly to how credentialing lock-in has operated in financial clearing, professional licensing, and DNS infrastructure. [acm](https://www.acm.org/binaries/content/assets/public-policy/europe-tpc/systemic_risks_agentic_ai_policy-brief_final.pdf) If your product roadmap prices agent coordination infrastructure as open and contestable, you are sitting on unacknowledged strategic risk. Not at the margins. At the base. #### The Window For two weeks in August 1940, the Luftwaffe bombed the RAF's Chain Home radar network. Fighter Command was close to collapse. Then Göring stopped, concluded the masts were resilient enough to ignore, and switched to bombing cities. That decision closed a window permanently. What decided the Battle of Britain was not numerical parity. It was that the RAF retained the coordination layer intact and compounding. No volume of subsequent sorties could overcome that structural disadvantage. The dimension this adds is one the evidence alone cannot supply: contestability is a window, not a standing condition. In late January 2026, the agent coordination layer was genuinely open. Moltbook was a six-week-old startup. OpenClaw was an independent protocol. No incumbent owned the registry. Regulators, open-source consortia, and competing platforms all had that window to define the infrastructure before network effects began compounding. The radar stations are now built and hardening. #### The Strongest Case Against This, Properly Stated Benedict Evans argued in late 2025 that LLMs may not produce winner-takes-all dynamics because software is capital-light and genuine lock-in requires switching costs that model capability alone cannot generate. The strongest version of that argument, the one that deserves a full hearing, is historical: coordination layers have also failed to sustain monopolies. XMPP nearly unified messaging. RSS was supposed to decentralise content distribution. Email federation survived despite every structural incentive to consolidate it. For a decade, the open-protocol advocates were right. Big Tech pushed, and the coordination layers held. That is not a trivial record to dismiss. [linkedin](https://www.linkedin.com/posts/benedictevans_software-historically-was-a-capital-light-activity-7384151635964428288-P-n6) Those prior cases involved communication protocols with no identity credentialing requirement. You could switch XMPP clients, carry your contacts, and rebuild in an afternoon. Agent verification removes that escape route entirely, because your verified identity, your trust history, and your routing relationships are all tethered to the registry that issued them. That is a categorically different switching cost, not a larger version of the same one. The European Commission's accelerated abuse-of-dominance investigation into Microsoft's cloud and AI bundling, opened in February 2026, found sufficient evidence of precisely this kind of credentialing lock-in to fast-track the inquiry. That is a formal regulatory finding based on submitted evidence, not another assertion dressed as a counter-example. [techzine](https://www.techzine.eu/news/privacy-compliance/138815/accelerated-investigation-into-microsofts-cloud-and-ai-monopoly/) #### Three Sentences. Then Stop. As of March 14, Meta owns the agent verification layer and OpenAI owns the protocol layer, with both companies under active EU antitrust investigation and no resolution expected before 2028 at the earliest. Every product roadmap that prices agent coordination infrastructure as open and contestable is carrying strategic risk it has not priced, and that risk sits at the foundation. If you are in a client meeting this month pitching a multi-agent architecture and you have not repriced that dependency, your client will find out before you do, and that conversation will be considerably worse than this one. [macfarlanes](https://www.macfarlanes.com/insights/102lya3/ai-tools-under-the-antitrust-spotlight-as-commission-opens-abuse-of-dominance-ca) --- ### Release v1.19.19 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-19-19) Ask voice fullscreen improvements and post-audit security: optional rate limits, response headers, safer API surfaces, billing and push validation. Documented on this page; code ships as Git tags v1.19.19 and v1.19.20. Release 1.19.19 improves the Ask voice fullscreen experience for accessibility and clarity. The dialog uses an explicit title and description for assistive technology, dynamic status text for listening, connecting, TTS, mute, and follow-up prompts, and a short trust line under the sources pill when collapsed. The TTS answer region uses aria-live off during word-level highlighting to reduce screen reader churn; transcript scrolling respects reduced motion. The waveform halo animation respects prefers-reduced-motion, and the particle canvas skips heavy drawing while the tab is hidden. Closing the session returns focus to the voice entry control, and a Type with keyboard instead action exits to the main Ask view. Analytics records voice session end and fullscreen close with a source (button, Escape, or type instead). Hub and hero artwork reflect this release. The same delivery cycle includes a security and quality pass (repo structure can be explored with Graphify when a local graph is built; see project-documentation/GRAPHIFY.md). Optional Upstash Redis rate limits protect /api/chat, /api/ask/transcribe, and /api/feedback when UPSTASH_REDIS_REST_URL and UPSTASH_REDIS_REST_TOKEN are set; without them, behavior is unchanged. Global response headers add X-Frame-Options, X-Content-Type-Options, and Referrer-Policy alongside the existing Permissions-Policy. Chat and transcribe return generic errors to clients while details stay in server logs. Billing validates Stripe customer ids (cus_…) from cookies or body. Web Push subscribe payloads use schema validation with bounded user-agent length. Ask feedback auto-issues use the correct GitHub repository. #### Added - Ask: voice fullscreen sr-only title and description; sources trust cue when the list is collapsed - Ask: Type with keyboard instead link; focus return to the voice mode control after closing - Ask: analytics for ask_voice_session_end and ask_voice_fullscreen_close (source: button, escape, type_instead) - Optional per-IP rate limiting via @upstash/ratelimit (chat, transcribe, feedback) when Redis env is configured - Security headers: X-Frame-Options, X-Content-Type-Options, Referrer-Policy - Newsroom: this combined release note (hub card and article) #### Changed - Ask: voice status line (listening, mic ready, speaking, follow-up hint, prompt) - Ask: TTS scroll container aria-live strategy during highlight; transcript and answer scrolling honor reduced motion - Ask: waveform halo and canvas idle when the document is hidden - Ask: chat and transcribe generic client errors; upstream detail in logs only - Billing portal: validate Stripe customer ids before use - Web Push subscribe: Zod validation and user-agent length cap #### Fixed - Feedback: GitHub repo path for Ask feedback issues (gagan-malik/gagan-malik-website) --- ### Release v1.18.18 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-18-18) Ask transcribe and STT, Presto gallery and portfolio case studies, TOC with listen seek, toolbar and speed sheet polish, newsroom hub updates, custom cover artwork. Release 1.18.18 adds a transcribe API and client speech-to-text hooks with voice UI updates on Ask, a Presto case study gallery with screenshots and a downloadable ZIP, and a refactored Meta portfolio case study with hero metadata, skeletons, and copy refinements. Stories add a mobile share sheet, listen dock bar with fixed toolbar and playback speed sheet, a story hero row with read time and share, and an article table of contents that seeks listen playback to section offsets. Listen mode exposes seekToCharOffset and registerSeekHandler; case studies map TOC ids to character offsets. Newsroom article pages gain matching listen bar and TOC wiring, and this update ships with dedicated cover artwork on the hub card and article hero. Popovers stack above fixed chrome for speed controls; analytics locale prioritisation reports refresh; llms.txt and the media kit ZIP regenerate after production builds. #### Added - Ask: transcribe API route, STT hooks, and voice UI updates - Stories (Presto): case study gallery, screenshots, gallery ZIP download - Stories: Meta portfolio case study, story detail refactor, hero meta row (read time, listen, share), loading skeletons, adjacent nav - Stories / Newsroom: article TOC (mobile sheet and popover); TOC seeks listen playback via seekToCharOffset and case study char offset map - Newsroom hub: bento contrast, transparent hub, media kit nav, article copy - Newsroom: custom v1.18.18 release cover image (hub and hero) #### Changed - Stories: mobile share sheet; listen dock, toolbar spacing, mobile sheet chrome, centered idle/playback rows; portfolio sidebar copy - Listen / speed: preset-only speed sheet; speedSheetDescription i18n; popover z-index so speed UI stacks above fixed chrome - Newsroom: listen bar, share popover, and post wrapper for TOC and seek - Analytics: locale prioritisation report updates #### Fixed - Stories: shared gallery ZIP filenames; Presto hero uses card cover; five gallery screenshots #### Chore - Regenerate llms.txt and media kit ZIP after production build --- ### Release v1.17.17 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-17-17) Media kit and brand assets, Web Push digest and PWA badges, Ask voice and updates UX, navigation and stories polish, new newsroom writing. Release 1.17.17 ships a dedicated media kit with downloadable brand assets and a generated ZIP, Web Push on Android for newsroom digests with updated Privacy and Terms disclosures, and PWA improvements including installed-app badges via the service worker and a clearer Add to Home Screen path on iOS. Ask gains richer retrieval (up to eight source updates per answer with a follow-up control), voice mode fixes across TTS, voices API, and fullscreen layout, and navigation picks up the mobile sheet, skeletons, and tighter menu polish. Stories and archive layouts are refined for small screens, and the newsroom adds new writing alongside this release note. #### Added - Media kit page with brand assets, generated ZIP, navigation entry, and i18n - Web Push on Android for newsroom digests; Privacy and Terms updates for push - PWA: installed-app badges (Service Worker SET_BADGE), iOS Add to Home Screen hint - Ask: up to eight source updates per response with Show more follow-up - Newsroom articles: No Download Required, Diplomatic Immunity, Someone Designed That Button #### Changed - Media kit downloads UX and safer forced-save behavior for asset files - Navigation: mobile sheet, loading skeletons, snappier mobile menu, hamburger-aligned close - Stories: mobile logo sizing, CTA visibility, stats layout - Archive filters: desktop layout and spacing #### Fixed - Ask: voice mode TTS loop, voices API, fullscreen canvas refs, error UX, fullscreen hint alignment - E2E and integration tests for Ask, home, and What is New flows --- ### Release v1.16.16 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-16-16) Knowledge base with Readwise sync, seed embeddings upsert, Ask mobile sheet UX, voice mode fixes, archive filters, PWA toast fixes. Release 1.16.16 adds the knowledge base with Readwise sync (POST /api/knowledge, GitHub Action for incremental sync, OpenAI embed fallback), seed embeddings upsert for faster incremental updates, and Ask/chat improvements including mobile sheet UX and read aloud highlighting. Voice mode gets sphere visibility, layout, and animation fixes. Archive filters are refined (Reset first, hide scrollbar, fix cropped text). PWA toast receives pill buttons, Maia styling, and cancel handling fixes. #### Added - Knowledge base with Readwise sync: POST /api/knowledge, GitHub Action, OpenAI embed fallback - Seed embeddings upsert for faster incremental updates - PB-56: Backlog item for Ask/chat updates #### Changed - Ask: Mobile sheet UX, read aloud highlighting, New conversation fix - Voice mode: Sphere visibility, layout, follow-up, animation fixes - Archive filters: Remove nav border, fix cropped text, move Reset first, hide scrollbar - Listen mode: Primary/secondary font color for word highlight - Newsroom UPDATE pages: Hide Media section - Anthropic model update; Readwise type-safe error access; E2E/unit test updates - GitHub Actions: actions/checkout v6 for Node.js 24 #### Fixed - PWA toast: Pill buttons, preview page, Maia styling; ServiceWorkerState handling; cancel Action type --- ### Release v1.14.14 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-14-14) Listen mode highlight and scroll during TTS playback on newsroom articles; word-level highlighting, scroll-follow, and reduced-motion support. Release 1.14.14 adds word-level text highlighting and scroll-follow during TTS playback on newsroom article pages. When you tap Listen, the current word is highlighted and the view scrolls to keep it in view. Uses the Web Speech API boundary event, React Context for state sharing, and a listen-mode renderer with word spans. Fallback when boundary is unsupported (e.g. Firefox); playback and progress bar continue to work. #### Added - Listen mode highlight and scroll: Word-level text highlighting and scroll-follow during TTS playback on newsroom article pages - ListenModeContext: Shared state for currentCharRange, isPlaying, and boundarySupported between ArticleListenBar and article body - ArticleBodyWithListen: Client component that swaps to plain-text-with-word-spans view when playing; highlights current word; scrolls highlighted span into view with throttle and prefers-reduced-motion support - getArticleCharRanges: Character range mapping in lib/newsroom.ts for listen-mode highlighting; mirrors getArticlePlainText - PB-54: Backlog item and unit tests for getArticleCharRanges, ArticleBodyWithListen, and ArticleListenBar boundary handler #### Changed - ArticleListenBar: Integrates with ListenModeContext; attaches boundary handler to SpeechSynthesisUtterance; syncs currentCharRange and isPlaying; 2.5s timeout to detect unsupported boundary - Newsroom article page: Uses NewsroomPostClientWrapper with ListenModeProvider and ArticleBodyWithListen for post layout --- ### Why We Designed Contempt Into Our Interfaces [Read more](https://www.gaganmalik.io/en/newsroom/why-we-designed-contempt-into-our-interfaces) By Gagan Malik The story goes like this: brilliant engineers wanted to connect humanity. They wanted a digital town square where strangers could talk, share ideas, and build something together. Good intentions. Messy outcomes. Very tragic. Very avoidable. That is not the full story. The full story is that we ran the experiments, read the dashboards, and kept shipping the version that made people angrier and more contemptuous because that version made people stay. We did not stumble into toxic platforms. We optimised into them. #### The Weird One in Ekman's List Paul Ekman spent fifty years mapping human facial expressions across cultures and landed on seven universal emotions. Six of them are things that happen to you: fear, sadness, anger, disgust, happiness, surprise. Contempt is structurally different. It is the only emotion that appears on just one side of the face. A half-sneer. It does not signal what you are feeling. It signals a verdict about someone else: 'I am above you, morally and socially.' [paulekman](https://www.paulekman.com/universal-emotions/) Every major social platform gave that specific emotion a one-click broadcast mechanism. The quote-post. The ratio. The public reply count visible to everyone watching. Whether that was deliberate product strategy or an accidental consequence of building for virality is genuinely debated. What is not debated is the outcome: we took the one emotion in human psychology designed for ranking people and handed it a billion-user megaphone. #### Did the Dashboard Know Before the Designers Did? Ekman's 1992 foundational paper established that contempt is an approach emotion: it drives you toward a target rather than away from it. In plain English, contempt makes you click, reply, and come back tomorrow. I know this personally. In early 2022 I opened Twitter to follow a breaking news story, got pulled into a contempt thread I had not searched for, and looked up forty-five minutes later having produced nothing, decided nothing, and burned roughly $200 (about £160) in productive thinking time. The decision was to 'just check quickly.' The surprise was how fast the algorithm found something to make me sneer at. The consequence was a wasted afternoon and a slightly lower opinion of people I had never met. [gruberpeplab](http://gruberpeplab.com/3131/Ekman_1992_Argumentbasicemotions.pdf) Twitter's own internal Health Metrics research, published in 2021, confirmed the algorithm was disproportionately amplifying politically contemptuous content, not because users searched for it, but because it kept them scrolling. The engineers saw the data. The features stayed live. That is not an accident of scale. That is a decision made on a Tuesday afternoon. #### Your Next Job Interview Is Being Scored on This Data Emotion recognition AI, the technology quietly running inside automated hiring platforms and content moderation pipelines, is built predominantly on Ekman's emotion taxonomy. A 2012 study in 'PNAS' by Jack et al. found that facial expressions are not as universal as Ekman claimed: East Asian participants categorised significantly fewer distinct expressions than Western ones, meaning the scientific foundation these products stand on is genuinely contested. The products shipped anyway, into consequential decisions about employment, credit, and access. [pnas](https://www.pnas.org/doi/10.1073/pnas.1200155109) Here is the inference, flagged as inference: if these systems are trained on datasets scraped from platforms that spent fifteen years algorithmically promoting contempt-coded content, it is reasonable to expect the models will encode contemptuous expressions as markers of normal human engagement. That has not been proven at scale. The hiring platform scanning your face during a video interview is not waiting for the proof either. #### The Part the Struggle Session Gets Right During China's Cultural Revolution, public struggle sessions did something pure punishment never could: they made spectators complicit. You did not just watch someone get denounced. You were required to join, or your silence marked you as a sympathiser. Participation was the mechanism of control, not the violence itself. This is the dimension social media adds that simple 'public shaming' framings miss. Algorithmic amplification does not just show you contempt. It ranks your relevance based on what you engage with, which means staying neutral quietly buries you. The platform does not force you to join the pile-on. It just makes neutrality expensive. #### The Strongest Objection, and Where It Runs Out of Road The honest counter-argument is this: nobody planned it. Engineers built for connection, scale did the rest, and you cannot hold a product team responsible for the aggregate ugliness of a species with a documented taste for public punishment. Critics including James Russell have questioned whether Ekman's emotion categories are as cleanly universal as his framework claims, which makes the entire 'we designed contempt in' argument harder to sustain if the underlying science is contested. [communicationcache](http://www.communicationcache.com/uploads/1/0/8/8/10887248/is_there_universal_recognition_of_emotion_from_facial_expressions-_a_review_of_the_cross-cultural_studies.pdf) That objection has real weight, right up to 2021, when Facebook's own internal research, revealed by whistleblower Frances Haugen, showed the company knew its algorithms were amplifying divisive and contemptuous content and chose growth targets over redesign. Ignorance is a defence. A board presentation is not. #### What Happens If You Keep Scrolling Contempt is the one emotion research consistently finds most correlated with irreparable breakdown, in marriages, in organisations, and in democracies, because unlike anger it does not leave a door open for repair. We have scaled that emotion across every surface humans use to form opinions, hire people, and raise children. If the AI systems being built on this data are never retrained on cleaner signal, the next time an algorithm silently decides you are not a cultural fit for a job you were qualified for, it will be making that call with a model trained on fifteen years of people sneering at each other for engagement points. [thinkingfeelingbeing](https://thinkingfeelingbeing.com/2025/08/24/the-science-of-facial-expressions-paul-ekman-universality-and-the-debate-over-emotion/) --- ### Why Do Algorithms Ignore The Many And Worship The Few? [Read more](https://www.gaganmalik.io/en/newsroom/why-do-algorithms-ignore-the-many-and-worship-the-few) By Gagan Malik Everyone in tech worships the funnel. Awareness, consideration, decision, retention, referral. You've seen that funnel slide in every strategy deck you've ever touched. It is the founding religion of modern marketing and, in short form video, it is a liability. There is no gentle staircase. There is only a gladiator pit where every video walks in and most die in three seconds. The true unit of strategy is not the campaign. It is the repeatable format that wins that reflex swipe. And right now the most ruthless version of that format is the elimination game, where one option survives and the other gets buried. #### The $47,000 Launch Budget That Lost To A Phone Camera Last year I was consulting a lifestyle product startup, freshly seeded, 90 days to prove traction before the next check. They signed off 32,000 dollars (about £25,000) on a full launch content strategy: brand film, product explainers, polished case studies. Then layered 15,000 dollars (about £11,800) in paid social on top. Meta, LinkedIn, pre-roll. The goal was 500,000 views in 30 days. While the agency was colour-grading the brand film, one of the junior designers on my team built a rapid fire elimination bracket on his phone. Ten product features head to head. Twenty-eight seconds. Zero cost. He posted it on a Tuesday without telling anyone. By Friday it had 480,000 views. The paid campaign had generated 31,000 views across twelve videos at a cost-per-view that made the founder go very quiet on our next call. One data point is not a controlled trial. But when your zero-dollar organic post nearly hits the entire campaign target in four days, that is not luck. Data from thousands of YouTube Shorts shows that 50 to 60 percent of viewers drop off within the first three seconds, and on TikTok, videos that hold 70 to 85 percent retention past that window receive 2.2 times more total views than those that do not. Attention in short form is not bought through a media plan. It is won in the first three seconds or not at all. [opus](https://www.opus.pro/blog/ideal-youtube-shorts-length-format-retention) #### Algorithms Hunt Outliers, Not Audiences The algorithm is not a megaphone for your brand. It is a pattern recognition system that bets on statistical anomalies, and it does not care how much you spent on your brand guidelines. Shortimize's 2026 analysis identifies what they call an Outlier Score: a video's views divided by an account's median views. A video pulling 140,000 views on an account that typically gets 20,000 is a 7x outlier. That is what triggers amplification. Gurkha Technology's hook research puts a number on the threshold: strong hooks achieve a 70 to 80 percent 'watched past three seconds' rate, which correlates with significantly higher algorithmic distribution. Their clearest example of such a hook is a simple binary 'X vs Y' setup. The inference, and it is an inference rather than a proven causal chain, is that the elimination bracket is purpose-built to spike that outlier behaviour repeatedly. Every new matchup resets the curiosity loop. The algorithm does not care about your brand narrative. It cares whether your format generates abnormal retention curves. [shortimize](https://www.shortimize.com/blog/how-to-find-viral-video-patterns-in-your-niche) #### What If Your Comments Section Is Your Best Ad Budget? The elimination bracket grows your account by starting fights in your comments so you do not have to. The moment two options face off on screen, you manufacture winners and losers, and people cannot help themselves. Short form performance research shows that videos designed to invite debate in their structure drive higher engagement rates, which in turn lifts platform distribution. That is a correlation, not proven causation, but it is consistent across multiple platform studies. Duolingo is the clearest proof of concept at scale. They built their entire TikTok presence on conflict, absurdist humour and combative skits pitting their owl mascot against every cultural rival it could find. Followers grew from 50,000 to over 10 million, with an 11 percent engagement rate against an industry average of 2 to 3 percent. Their philosophy was explicit: 'The comment section is our social brief'. Conflict invites comment. Comments signal engagement. Engagement signals distribution. Engagement is no longer a KPI. It is collateral damage. [benchmarkemail](https://www.benchmarkemail.com/blog/short-form-video-marketing-trends/) #### Formats Are Assets, Videos Are Tests Your content roadmap is not a calendar. It is a format library, and most teams are spending serious money building the wrong one. Shortimize's pattern framework is explicit: mine a format, run a ten-video sprint to validate it, keep only what consistently beats your baseline, and iterate from there. Structure is the durable asset. Individual videos are disposable test cases. Apply that logic to the elimination bracket and you can pressure-test ten variations across any niche in 14 days: software tools, pricing models, industry opinions, product features. The variables are hook line, opening visual and proof style. The underlying logic of 'one survives, one dies' stays constant. In short form, the company that owns the most repeatable, high-performing formats has a structural distribution advantage over the one burning budget on one-off campaigns. [shortimize](https://www.shortimize.com/blog/how-to-find-viral-video-patterns-in-your-niche) #### Darwin, Not Don Draper Think of this less like Mad Men and more like how winning tactical formations spread through professional football. When a system proves it works at the highest level, every club in the league adopts a version of it within two seasons. The formation becomes the asset. The individual players running it are disposable test cases. The elimination bracket works exactly the same way. It is not a one-time trick. It is a replicable system that compounds. Every video you make in that format generates data, comments, and retention signals that make the next iteration sharper. By the time your competitor starts experimenting with the format, you have already run fifty versions and know exactly which variables convert. Tech teams keep trying to win with talent and budget. The bracket is a system. Systems beat talent at scale. [gurkhatech](https://gurkhatech.com/short-video-hook-formula-guide/) #### 'But We're A Serious Brand' The strongest objection is a fair one. Serious products are complex. B2B buyers run long decision cycles. You cannot explain a data platform, a medical device, or a fintech compliance stack through a TikTok elimination game. And if you try, you risk cheapening the very credibility your sales team spent years building. That is a legitimate concern, not a failure of imagination. Here is the counter-example. Ryanair sells a high risk, heavily regulated, operationally complex service in one of the most scrutinised industries in Europe. They are not a lifestyle brand flogging hoodies. Yet their TikTok strategy is deliberately crude, combative and often absurd. Behind the memes, they run one of the most cost-disciplined operations on the continent with record passenger volumes to prove it. The content does not explain the complexity. It earns the first click. Depth lives downstream. You use the bracket to buy the attention. You use the sales call to justify the price. [foryouadvertising](https://www.foryouadvertising.com/blog-posts/ryanairs-tiktok-social-strategy-a-masterclass-in-viral-video-marketing) #### The Stakes Short form platforms are now structurally rewarding elimination formats that generate outlier retention curves, visible conflict, and high comment velocity. The implication is that every dollar you spend producing slow, narrative-heavy content in 2026 is a dollar quietly transferred to the competitor who figured out the bracket six months before you did. Ignore this and you will not get a postmortem or a panel slot at a conference to explain what went wrong. You will just be gone. [gurkhatech](https://gurkhatech.com/short-video-hook-formula-guide/) --- ### The Unemployment Rate Is Fine. That Is the Trap. [Read more](https://www.gaganmalik.io/en/newsroom/the-unemployment-rate-is-fine-that-is-the-trap) By Gagan Malik Three years of AI headlines and unemployment is near record lows. The office is still full. The doomers were wrong again. The chatbot took nobody's job. Pour yourself a flat white and open your portfolio dashboard. Here is the problem with that. Anthropic's economists Maxim Massenkoff and Peter McCrory, writing in their February 2026 paper "Labor Market Impacts of AI," measured actual AI usage against theoretical capability across every major occupational category. The headline finding everyone repeated: no unemployment spike. The finding nobody mentioned: hiring of workers aged 22 to 25 in high-exposure roles dropped 14 percent from late 2022 onwards. The jobs did not disappear. They stopped being created. Those are two different problems, and you are currently holding only one instrument. [anthropic](https://www.anthropic.com/research/labor-market-impacts) #### You Checked the Unemployment Stats and Felt Relieved. Put the Dashboard Down. Harvard Business Review, in a February 2026 survey of 1,006 global companies, found that firms are cutting headcount projections because of AI's anticipated capability, not its demonstrated performance today. Pre-emptive structural reduction. Not a response to evidence. A bet placed quietly, in advance, on a future that has not fully arrived yet. [hbr](https://hbr.org/2026/01/companies-are-laying-off-workers-because-of-ais-potential-not-its-performance) The unemployment rate measures the stock of existing jobs. It does not measure the flow of new ones that were never opened. When a firm hires fifteen graduates instead of twenty-five, the rate does not move. Ten people who would have been hired become unrecorded, uncompensated, and invisible to every dashboard the optimists are currently pointing at. The instrument is not broken. It is measuring the wrong thing. That is worse. #### Where Did All the Juniors Go? I know this move because I made it. In Q3 2025, I cancelled two hires: Aisha, a junior UX designer, and Callum, a full-stack engineer fresh out of his bootcamp. I redirected $20,000 (£15,000) of combined headcount budget into AI tooling instead. In December 2025, I finished a design sprint and a working prototype alone in five days and sent the final files. The surprise was not the speed. It was what I noticed when I hit send: no Slack thread, no handoff call, no one else in either folder. I had built the whole thing in a room with no one in it and had not registered that until the door closed. I kept the full margin. Aisha, twenty-two, first-class design degree from Northumbria University, graduated July 2024, sent eleven applications between August and December and received zero offers. Not because her portfolio was weak. Two of those companies had quietly removed the junior design tier from their hiring plan before she graduated. Callum, twenty-four, three years of self-taught development and a GitHub full of side projects, reached final rounds at two firms and watched both roles evaporate in the same quarter his offer was supposed to come. Neither of them appears in any unemployment statistic. Both are, on paper, fine. I did not think about either of them once until I sat down to write this sentence. #### The Hire That Never Happened Is Still a Career That Was Taken. The Atlantic, in a February 2026 feature on AI and the labour market, reported that Dario Amodei, the chief executive of Anthropic, publicly expects 10 to 20 percent unemployment as AI capabilities reach full deployment. That is not a critic's prediction from the outside. That is the founder of the product doing the displacing, describing what he expects his own product to eventually do at scale. [theatlantic](https://www.theatlantic.com/magazine/2026/03/ai-economy-labor-market-transformation/685731/) The distance between 'no unemployment spike yet' and '10 to 20 percent at completion' is not a reason to exhale. It is a timeline with most of its travel still ahead. Entry-level roles that are not being posted today are the leading edge of a structural adjustment that headline statistics will not register for another three or four years. By then the counterfactual is invisible, the policy window has closed, and the debate is purely academic for everyone who needed the pipeline to be open and found it shut before they arrived. #### The Alarm Is Not Going Off. That Is the Problem. Carbon monoxide is colourless, odourless, and undetectable without the right instrument. You feel fine. Slightly tired, maybe. The room looks completely normal. The first signal that something is wrong is frequently the last one you are conscious enough to act on. The problem was never invisible. You were checking for smoke. Carbon monoxide does not trigger a smoke detector. That is not what kills you. Forbes, in a March 2026 methodological critique of the Anthropic paper by analyst Hamilton Mann, argued that the study cannot measure what it does not track: wage compression, stalled promotion velocity, and the systematic disappearance of junior pipeline roles. None of these appear in unemployment data. The unemployment rate is the smoke detector. It is not going off. The junior hiring pipeline has been running 14 percent below its pre-2022 level for three years, per the Anthropic paper's own figures, and every government alarm is perfectly, uselessly silent. The gap between the instrument and the reality it fails to capture is not a technical footnote. It is where Aisha and Callum live. [forbes](https://www.forbes.com/sites/hamiltonmann/2026/03/08/anthropics-study-does-not-measure-ais-labor-market-impacts/) #### The Case for Patience Is Genuinely Strong. Here Is Why It Is Not Enough. Let me hold the optimist position at full strength, because it has earned it. The Peterson Institute for International Economics, in a March 2026 review of the AI labour research base, argued that the evidence is genuinely thin, that the Anthropic paper measures one company's product usage rather than economy-wide displacement, and that every major automation wave in recorded history produced net job creation rather than net destruction. The historical null hypothesis is that new categories of work emerge to absorb displaced workers. History has a strong track record on this. I am not going to pretend that argument is weak, because it is not. [piie](https://www.piie.com/blogs/realtime-economics/2026/research-ai-and-labor-market-still-first-inning) The counter is specific, not rhetorical. Daron Acemoglu and Pascual Restrepo, in a peer-reviewed paper published in Econometrica in 2022, found that automation accounted for 50 to 70 percent of the rise in US wage inequality between 1980 and 2016, and demonstrated that net job creation requires a sufficient supply of new complementary tasks for displaced workers to absorb. When new task creation slows, automation produces sustained wage depression and reduced employment share even without touching the headline unemployment number. The historical optimism is not wrong in principle. It is conditional on new work arriving fast enough, and for the right people. Neither condition is confirmed by the current evidence. #### Three Sentences Before You Close This Tab. The Financial Times noted in March 2026 that legal, finance, and management sectors sit at 80 to 90 percent theoretical AI exposure and under 15 percent observed usage, with most of the adoption curve still ahead. When that gap closes, it will close fast, because the deployment barrier is a software procurement approval and a six-week onboarding cycle, not a factory retooling. If you are Aisha or Callum, portfolio current, waiting for the pipeline to reopen, the smoke detector will not tell you when the room has already changed. [ft](https://www.ft.com/content/2cd79c2c-d1f0-4417-a06f-2572868ee858) --- ### Release v1.13.13 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-13-13) Voice chat UX redesign (Ask), connecting state, push-to-talk, first-time voice picker, AI speaking strip, optional fullscreen voice view; Ask UI improvements. Release 1.13.13 delivers the voice chat UX redesign on the Ask page: explicit voice session with clear states (Idle → Connecting → Listening → Submitting → AI speaking), always-visible voice entry, dedicated connecting state with retry, push-to-talk (Hold to talk), first-time voice picker, “Gagan is speaking” strip, and error recovery with Try again / Open settings. Optional Perplexity-style fullscreen voice view adds a voice-reactive particle sphere, sources strip, and live transcript. Ask UI gets category dropdown, dual-state voice/text toggle, Sources and follow-ups when streaming completes, and layout fixes. #### Added - Voice chat UX redesign (Ask): Explicit voice session with state machine; voice affordance always visible when supported - Connecting state: “Getting your microphone ready…” with spinner; Cancel; timeout and error handling with retry - Listening (hands-free): Stop & send, Exit voice, waveform, live transcript; focus animated ring; Gear only in voice mode - Push-to-talk: Hold to talk when activation is push; pointer/touch and Space keyboard - First-time voice picker: Choose voice on first tap; preset options, preview; then Connecting - AI speaking strip: “Gagan is speaking” bar with Stop playback and Exit voice; hands-free re-listen after TTS - Error recovery: Card-level alert with “Try again” and “Open settings” - Perplexity-style voice view (optional): Fullscreen UI with particle sphere, sources strip, mute, live transcript, Exit chat - Ask UI: Category dropdown, Cog/Mic/Settings2; follow-up filtering and Show more; dual-state voice/text toggle; Sources and follow-ups when response done streaming; suggestion chips and message-link chips #### Changed - Voice entry: Voice button always visible when supported; Gear removed from idle form—only in voice-mode UI - Voice ring: Shown only when voice active; hidden on error; waveform stagger and softer animation; list numbered + spacing - Ask: Left-align follow-up chips; citation number and list on same line; prepareSendMessagesRequest includes messages in body - Connecting definition: Use voiceConnecting for isConnecting so Connecting state displays correctly #### Fixed - Voice: isConnecting definition, animated voice ring restore, error state and retry/settings CTAs - Ask: API error body parsed as JSON; error alert with Try again; memory/voice i18n keys for production --- ### Release v1.12.1 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-12-1) Article Listen bar (TTS), Share article popover, byline row on all newsroom posts, playback and byline fixes. Release 1.12.1 adds the Article Listen bar (listen/pause, progress, playback speed 0.5x–3.0x) and Share article popover to newsroom article pages. The byline row (By Gagan Malik, Listen, Share) now appears on all post detail pages (not only WRITINGS), with reliable playback on Chrome/Safari and byline fallback when subtitle is empty. #### Added - Article Listen bar: TTS listen/pause, progress, and playback speed (0.5x–3.0x) on newsroom article pages - Share article popover: Share menu (LinkedIn, X, Facebook, WhatsApp, Reddit, email, copy link) on article header - Article listen bar tests: Unit tests for empty state, Listen button, and play → progress UI #### Changed - Article Listen bar: Prime voices and resume synthesis on user gesture; set utterance lang/volume for reliable playback (Chrome/Safari) - Share button: Icon-only trigger (size-icon) to match listen bar - Newsroom article page: Integrate ArticleListenBar and ShareArticlePopover in header; plain text from getArticlePlainText(post) - Newsroom posts (all categories): Listen bar and Share on all post detail pages (not only WRITINGS); byline row By Gagan Malik + Listen + Share with text-sm font-medium; same row layout for byline and actions; byline fallback when subtitle is empty or whitespace so the row is always visible on every post (excluding stories and videos) #### Fixed - Article Listen bar: Playback error state and Playback unavailable message when synthesis fails - Playback speed: Speed selection applied to active utterance when changed while playing - Newsroom byline: Byline By Gagan Malik now always shows on post pages when subtitle is empty or whitespace-only --- ### Release 1.12.0 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-12-0) Stripe Customer Portal, llms.txt, About refresh (education, Techstars), SEO Person schema, PWA crawler fix, clean story slugs and 301s. Release 1.12.0 adds the Stripe Customer Portal and pricing alignment, llms.txt for LLM crawlers, and a refreshed About page with education (LBS), bio, and Techstars coverage. SEO and AEO improvements include Person JSON-LD disambiguation, hero heading hierarchy, and metadata aligned with LinkedIn. PWA and docs get static PNG icons and cleaner structure; story slugs are cleaned with 301 redirects. #### Added - llms.txt generator: Build-time llms.txt for LLM crawlers; includes education (LBS) and site context - 725M Bug Hunt: Newsroom article and related updates - Stripe Customer Portal: Customer portal for subscription management; pricing alignment - Video cards pagination: Pagination for video cards in newsroom/archive - PWA icons: Static PNGs for Newsroom, Ask, and docs (RSS, input bar, design system) #### Changed - AEO: Sr-only copy updates for answer-engine optimization - Stories: Clean story slugs with 301 redirects from old URLs - SEO Person JSON-LD: mainEntityOfPage and entity disambiguation - About: Redesign; Lessons learnt section - SEO & nav: Hero heading hierarchy; nav Ask Gagan; Who/What/How consolidated to About - Metadata: Aligned with LinkedIn (titles, descriptions) - Analytics: Moved to locale layout; Vercel/pnpm build fix - PWA: Icons generated as static PNGs #### Fixed - PWA: Service worker bypasses cache for crawler document requests so search engines and AI crawlers receive fresh HTML from the edge (SEO) - About: Chunk load error on About page - Favicon: Next.js app icon convention - Privacy: Hide behaviour corrected where needed - Video overflow: Video card overflow on small viewports - Ask: Integration test fix #### Removed - Docs: Introduction and Related docs sections --- ### Release 1.11.11 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-11-11) YouTube videos in newsroom, YouTube Data API integration, video transcripts in RAG, archive filter, external link handling. Release 1.11.11 brings YouTube videos to the newsroom, integrates the YouTube Data API for build-time channel content, adds video transcripts to RAG for the Ask page, introduces a Videos filter in the archive, and improves external link handling for video cards. #### Added - YouTube videos in newsroom: Videos section displays content from @gagan_malik via YouTube Data API v3 - YouTube Data API integration: Build-time fetch of channel videos; YOUTUBE_API_KEY env var (optional; falls back to manual entries when unset) - Video transcripts in RAG: Ask page retrieves YouTube video transcripts; EMBED_VIDEO_TRANSCRIPTS=false to disable - Archive filter: Videos option in archive type dropdown (/newsroom/archive?type=video) - External link handling: Video cards open YouTube links in new tab with aria-label and external-link icon #### Changed - Newsroom cards: External YouTube hrefs use instead of Link - Next.js images: i.ytimg.com and img.youtube.com added to remotePatterns for YouTube thumbnails - seed:embeddings: Fetches and chunks video transcripts when EMBED_VIDEO_TRANSCRIPTS is enabled (default) --- ### Release 1.10.0 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-10-0) RSS feed, What's New banner, PWA badge, ghost-style newsroom buttons, footer RSS link. Release 1.10.0 adds an RSS feed, a What's New banner on the home page, PWA badge for new content, and refines the newsroom with ghost-style buttons and an RSS link in the footer. #### Added - RSS feed: /feed.xml route (RSS 2.0) with newsroom content; link rel='alternate' in layout head - What's New banner: Home-page banner showing new updates since last Newsroom visit; dismiss stores count in localStorage - PWA badge: App badge shows new content count when PWA is installed; clears on Newsroom visit - Preview route: /preview/whats-new-banner for design/dev preview of What's New banner - localStorage keys: newsroom_last_visit_at, whats_new_dismissed_count for What's New and PWA badge (documented in Privacy) #### Changed - Newsroom: Ghost-style buttons for "View Archive" and "More Stories" - Footer: RSS Feed link in Resources - Documentation: LEGAL_PAGES.md enumerates localStorage keys; PRODUCT_BACKLOG PB-44 done --- ### Release 1.9.0 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-9-0) i18n locales (hi, ar, es), region selector, FAQ accordion, region-specific pricing, RTL for Arabic. Release 1.9.0 adds Hindi, Arabic, and Spanish locales with full translations, a footer region/language selector, FAQ accordion on the About page, region-specific pricing support, and RTL for Arabic. #### Added - i18n: Hindi, Arabic, and Spanish locales with full translations - Region/language selector: Footer dropdown (Apple-style) with regions grouped by geography - Legal: LegalLocaleNotice on Privacy and Terms for non-English locales; region-specific legal docs structure - RTL: Arabic locale support via LocaleDir (dir='rtl', lang) - Region-specific pricing: Checkout API accepts regionCode; optional STRIPE_PRICE_*_REGION env vars - FAQ accordion: About page FAQ as collapsible accordion with card fill - Pricing sticky bar: Pins below nav (top-16); moves to top-0 when nav hides on scroll #### Changed - Footer: Legal links row (Privacy | Terms | Site Map); removed Get in touch and LinkedIn text links from Connect - Sitemap: All pages included for en, hi, ar, es locales - Backlog: PB-40 (region selector), PB-41 (local currency pricing), PB-42 (legal translations) --- ### Release 1.8.0 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-8-0) Skeleton screens for all pages, archive filters fix, sticky header on newsroom pages. Release 1.8.0 adds skeleton loading screens across the site, fixes the archive filters sticky nav bug, and keeps the header visible on all newsroom pages. #### Added - Skeleton screens for all pages: Home, Newsroom, Site Map, Docs, Legal (terms/privacy), Ask preview, Stories, Archive - Newsroom skeleton preview route (/newsroom/skeleton-preview) for QA - Sticky header on all newsroom pages: index, detail articles, and archive (with handoff to filters bar) #### Changed - Archive filters: Replaced Radix Select with DropdownMenu (modal={false}) to fix sticky nav disappearing when opening dropdowns - Nav visibility: Header stays visible on /newsroom and /newsroom/[slug]; on archive, header hides when filters bar reaches it - Mobile: Writings and Stories hero cards match other cards on mobile - Redirect: /portfolio redirects to /stories --- ### Release 1.7.0 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-7-0) Mobile responsive audit, touch targets (44pt), Ask history as Sheet on mobile, floating toggle. Release 1.7.0 delivers a mobile responsive audit with improved touch targets, Ask page layout on small screens, and responsive polish across Stories and Newsroom. #### Added - Mobile: Floating history toggle on Ask page (fixed top-right overlay) - Ask: Right-side Sheet for conversation history on mobile - Ask: 16px horizontal padding on mobile for chat content #### Changed - Mobile responsive audit: root overflow-x-hidden, hero responsive typography, pricing table scroll, sticky bar layout - Touch targets: 44pt minimum — footer icons, hamburger, theme toggle, testimonials play/pause and dots, Ask voice/send buttons - Stories: Work with logos CTA visible on mobile (no longer hover-only) - Design impact stats: grid layout on mobile (2 columns), flex on desktop - Ask: History sidebar uses Sheet on mobile when below lg breakpoint - Newsroom: Responsive hero and bento min-heights - Archive: Reset button touch target size --- ### Release 1.5.0 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-5-0) Favicon, Open Graph images, RAG postbuild automation, vibe coding fix, cursor rule. Release 1.5.0 adds favicon, Open Graph images for social previews, and RAG automation. #### Added - Favicon: app/icon.png from the logo (replaces Vercel default). - Open Graph images: Newsroom articles now use their cover image; a default logo is used for pages without images. - RAG: postbuild seed automation — Embeddings are reseeded on every Vercel deploy. - Cursor rule: rag-seed-embeddings.mdc (a reminder to run the seed after newsroom changes). - Newsroom: Release 1.4.1 update post with a mechanical blueprint cover. #### Changed - RAG: The system prompt has been extended for "vibe coding," articles, and writing queries. - RAG: The seed script now skips gracefully when DATABASE_URL is missing and supports AI_GATEWAY_API_KEY. - Social link previews: Correct cover images are now used for newsroom articles, with a logo fallback. --- ### Release 1.4.1 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-4-1) Against One-Click Coding article, citation pills with favicons (newsroom + Ask page), publication dates on Writings, new cover image. Release 1.4.1 adds a new Writings article and improves citation UX across the site. #### Added - Newsroom: "Against One-Click Coding: Reclaiming Craft for Personal Projects" article (replaced migration article) - Newsroom: citation pills with favicons (newsroom article + Ask page) - Newsroom: new cover image for Against One-Click Coding article #### Changed - Newsroom: Writings bento cards show publication date instead of relative time - Newsroom: citation links styled as pills with favicon, Globe fallback --- ### Release 1.4.0-rc.1 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-4-0-rc-1) Deslop, refactors (getRelatedItemsForPost, WritingsBentoCard composition), Ask test TooltipProvider fix, related items (3 max, sorted desc, hidden when empty), removed Kyle Hanagami story. Release 1.4.0-rc.1 focuses on code quality, refactors, and test fixes. #### Added - Newsroom update post for v1.4.0-rc.1 #### Changed - Deslop pass across the codebase - Refactored getRelatedItemsForPost and WritingsBentoCard composition - Ask integration test — TooltipProvider fix - Related items section: 3 max, sorted desc, hidden when empty - Removed Kyle Hanagami story from newsroom --- ### Release 1.4.0 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-4-0-rc-0) Ask page: case study citations with thumbnails, Show more/less chips toggle, ASSISTANT_QUESTIONS.md catalogue, helper text and Button size variants. Release 1.4.0 adds Ask page improvements and Button size variants. #### Added - Ask page: case study list queries now return one citation per study with thumbnails; each bullet links to /stories/[slug] - Ask page: "Show more" / "Show less" ghost toggle to expand suggestion chips - ASSISTANT_QUESTIONS.md catalogue for assistant prompts #### Changed - Ask page: helper text renamed to "Here's what you can ask me" - Button: added xl and icon-xl size variants (fixes build for About and Rolex pages) --- ### Release 1.3.0 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-3-0) Cookie consent (GDPR), release plan docs, Ask integration tests. Release 1.3.0 adds a cookie consent banner for GDPR compliance, release plan documentation, and integration tests for the Ask page. #### Added - Cookie consent banner with 'Necessary only' and 'Accept all cookies'; GA4 loads only after consent; delayed display (3s or first scroll); links to /privacy - Release plan documentation (RELEASE_PLAN.md) and cross-references in product docs - Ask page integration tests --- ### Release 1.1.0 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-1-0) Basic homepage, seven full case studies, Stories hero with Ask Gagan CTA. Release 1.1.0 introduces a streamlined homepage, seven full case studies (Apple, Google, Presto, Aviva, Lloyds Bank, EE, John Lewis), and a Stories hero with the Ask Gagan primary CTA. #### Added - Basic homepage (hero, trusted by, selected work, about teaser) - Case study template with takeaways, company sidebar, hero quote - Article JSON-LD on story detail pages - Stories hero with 'Selected work with context' value prop and Ask Gagan CTA #### Changed - Home primary CTA to /stories - Story detail TrackedLink on CTA strip and nextSlug navigation --- ### Release 1.0.0 is now live [Read more](https://www.gaganmalik.io/en/newsroom/release-1-0-0) Initial release: Home, About, Stories, Newsroom, Pricing, Ask, Analytics, SEO. Release 1.0.0 marks the initial launch of the personal website. #### Added - Home with value prop, hero, work carousel, principles - About page with How I Lead, case studies, testimonials - Stories (case studies) with project pages - Newsroom - Pricing with Stripe checkout - Ask page (AI chat) - Analytics (Vercel, GA4), SEO (metadata, sitemap, canonicals, JSON-LD) - Dark mode and responsive layout --- ### Against One-Click Coding: Reclaiming Craft for Personal Projects [Read more](https://www.gaganmalik.io/en/newsroom/against-one-click-coding) By Gagan Malik Let's talk about your personal website. I know what it looks like. You built it on a Sunday afternoon. You fired up a vibe-coding app—maybe Claude Code, maybe Lovable, maybe one of those platforms that promises "from prompt to production in 60 seconds." You typed "Make me look like a visionary but approachable tech founder," pressed enter, and went to make a flat white. And it looks fine. It's got the trendy brutalist typography. The dark mode toggle. The slight gradient on the hover state. It looks exactly like the websites of the other 400,000 guys who asked an AI to make them look visionary but approachable this month. You outsourced your digital identity to an LLM optimizing for the median acceptable aesthetic. Congratulations. You are now perfectly average. I'm building my personal site right now. I'm doing it the hard way. I'm in an IDE. I'm fighting with Git. I'm staring at a CLI. Am I doing this because I'm a masochist? No. I'm doing it because I understand the difference between a product and a craft object. And if your personal website isn't a craft object, why does it even exist? Let's get the economics out of the way first. I am not anti-automation. In my day job, I use AI tools to collapse effort constantly. If I need a throwaway landing page to test a value prop for a new SaaS product, I am absolutely letting an agent build it. Time is money, and I am not spending £500 of my time hand-coding a button for an MVP that might die on Thursday. But your personal website is not a throwaway MVP. It is your canonical footprint on the internet. It is the one digital asset you own outright in an era of rented land. When you use a low-code or no-code platform for this, you aren't just saving time. You are accepting a fundamental trade-off: you are trading agency for speed. You think you are the conductor, but you are actually the passenger. You are constrained by their abstractions, their component libraries, their pricing tiers, and their definition of what a website should be. As developer Jay Little put it recently, the promise of low-code is a lie when it comes to long-term maintainability. You are taking on technical debt disguised as convenience. [jaylittle](https://jaylittle.com/post/view/2023/4/low-code-software-development-is-a-lie/) More importantly, you are missing out on the actual value of building something yourself: the friction. There's an old John Dewey concept—often summarised as "we learn by doing". Dewey's actual point was sharper: we do not learn from experience alone; we learn from reflecting on experience. [structural-learning](https://www.structural-learning.com/post/john-deweys-theory) You don't reflect when a machine generates 4,000 lines of React for you in three seconds. You just deploy it and hope it doesn't break. When I'm in my IDE, setting up my own routing, configuring my own build tools, and making deliberate commits via the CLI, I am forced to be intentional. Every folder structure is a design decision. Every dependency is a liability I have to accept. I am staying close to the grain of the medium. This is software as a craft, not a commodity. [8thlight](https://8thlight.com/insights/why-software-development-is-a-craft) And let's talk about the ROI of that friction. Yes, it costs me more hours up front. But what does it buy me? It buys me full-stack literacy. When things break—and they will break—I have the levers to fix them. I'm not waiting in a Discord support channel for a startup with a $50M valuation and an 18-month runway to patch their platform. It buys me portability. My Git repo is mine. I can move it to Vercel, to Netlify, to a Raspberry Pi in my living room. Try exporting your highly customized, database-linked project out of a proprietary no-code builder. It's like trying to get your data out of a casino. But mostly, it buys me taste. Vibe-coding tools optimize for output speed. They encourage you to ship the first plausible draft. But plausible is the enemy of precise. When you build the stack yourself, the constraints force you to make choices. You inject your actual personality into the architecture, not just the CSS. [rebeccabilbro.github](https://rebeccabilbro.github.io/low-code-no-code/) We are entering an era where AI can write competent code instantly. Competent code is about to have a market value of zero. What will have value is taste, architectural judgment, and a deep understanding of how systems actually fit together beneath the abstractions. You don't build those muscles by typing prompts into a chat box. You build them in the trenches. You build them by getting your hands dirty in the terminal. Your personal site shouldn't just be a billboard. It should be your lab. It's the one place where you don't have to compromise with a product manager, a client, or a venture capitalist. Stop treating your digital identity like a fast-food order. Reclaim your agency. Open your terminal. Build it yourself. It's going to be frustrating. It's going to take longer. But when you're done, you'll actually own it. And in 2026, owning your own infrastructure might be the biggest flex there is. --- ### The $725 Million Bug Hunt [Read more](https://www.gaganmalik.io/en/newsroom/the-725-million-bug-hunt) By Gagan Malik Let me tell you about the most expensive hobby in tech that nobody talks about at dinner parties: chip verification. Not the sexy kind of chips. Not the ones Jensen Huang waves around on stage like a newborn at a christening. I mean the 18 months before that, when thousands of engineers sit in cubicles running simulations, hunting bugs in circuits smaller than a human hair, praying to whatever god they believe in that the thing works the first time. Spoiler: it almost never does. The latest data from the Siemens EDA/Wilson Research Group study landed like a brick through a window. First-time silicon success has plummeted to **14%**. That is not a typo. Fourteen per cent. Down from 24% two years prior and roughly 30% historically. Three out of four chip projects are running behind schedule. And the average cost to design a single chip at the 2nm node? **$725 million**. [semiengineering](https://semiengineering.com/first-time-silicon-success-plummets/) You now have better odds of a startup surviving five years than your chip working first time. And when you lose, you do not lose your shirt. You lose a mid-size office building. #### The 70% problem nobody pitches Here is the quiet part out loud: **verification consumes over 70% of a chip project's time and headcount**. Not design. Not manufacturing. Verification. The bit where you check whether the thing you drew actually does what you think it does. [blogs.sw.siemens](https://blogs.sw.siemens.com/verificationhorizons/2025/09/03/why-first-silicon-success-is-getting-harder-for-system-companies/) And within that 70%, the nastiest, most manual, most soul-destroying task is **debugging**. Finding the root cause of a failure in a haystack of billions of transistors. Post-silicon debug alone can run **$15 to $20 million and six months** per project. Pre-silicon debug is no better; it is just cheaper to cry about. [web.eecs.umich](https://web.eecs.umich.edu/~imarkov/pubs/conf/iccad07-fogclear.pdf) Harry Foster, chief verification scientist at Siemens EDA, put it bluntly: "What we're doing is not working. We need to significantly increase productivity". This is not some blogger. This is the person who has run the industry's definitive verification survey for over two decades. [semiengineering](https://semiengineering.com/first-time-silicon-success-plummets/) #### Why investors should care (but do not, yet) Every AI model you use, every copilot, every autonomous vehicle, every data centre GPU cluster, sits on top of silicon. Jensen Huang himself frames AI as a five-layer cake: energy, chips, infrastructure, models, applications. If the chip layer breaks, the four layers above it do not wobble. They collapse. [blogs.nvidia](https://blogs.nvidia.com/blog/davos-wef-blackrock-ceo-larry-fink-jensen-huang/) And right now, the chip layer has a **structural crack called the verification productivity gap**. Complexity is scaling exponentially. Verification capacity is not. The gap was first identified in the early 2000s. We are now in what Siemens officially calls **"Verification Productivity Gap 2.0"**. Same disease, nastier symptoms. [verificationacademy](https://verificationacademy.com/topics/planning-measurement-and-analysis/breaking-the-bottleneck-overcoming-the-verification-productivity-gap/) Meanwhile, the EDA market, the tools that make chip design possible, captures roughly **2 to 3% of semiconductor revenue**. Two to three per cent! The entire industry that enables a $600 billion chip market earns less than what Apple spends on marketing. Liyue Yan at Boston University nailed it: EDA has a value capture problem. They create enormous value, then negotiate like they are selling second-hand furniture at a car boot sale. [semiwiki](https://semiwiki.com/eda/363537-eda-has-a-value-capture-problem-an-outsiders-view/) #### The money is moving I will be honest: I dismissed this space for too long. I looked at EDA and saw legacy toolchains, three dominant incumbents, and margins that made my eyes water for the wrong reasons. I was wrong. The market has shifted under my feet. The EDA market hit roughly **$20.8 billion in 2026**, growing at 8 to 9% CAGR. Verification and sign-off is the **largest segment at 26% share**, about $5.4 billion, and growing faster than the average. Cloud EDA is a **$3.7 billion** sub-market on its own. [mordorintelligence](https://www.mordorintelligence.com/industry-reports/electronic-design-automation-eda-tools-market) And then ChipAgents happened. Founded in 2024, the company just closed an oversubscribed **$74 million** raise led by TSMC-backed Matter Venture Partners, alongside Bessemer, Micron, MediaTek, and Ericsson. They report **140x year-over-year ARR growth** and deployment at **80 semiconductor companies**. Their pitch: agentic AI that reads specs, generates verification assets, and does automated root-cause analysis. Basically, they are trying to eat the debug bottleneck alive. [businesswire](https://www.businesswire.com/news/home/20260217568914/en/ChipAgents-Raises-$74M-to-Scale-an-Agentic-AI-Platform-to-Accelerate-Chip-Design) #### The framework you actually need Stop thinking of verification as a line item in chip design. Start thinking of it as **risk infrastructure for the entire AI economy**. Every week a tape-out slips, a hyperscaler's GPU cluster ships late. Every respin burns **$25 million and three to six months**. Every escaped bug that reaches production costs **100x** what it would have cost to catch in simulation. [community.cadence](https://community.cadence.com/cadence_blogs_8/b/fv/posts/how-to-maximize-productivity-and-lower-cost-for-enterprise-prototyping) The companies that compress debug cycles do not just save their customers money. They accelerate the clock speed of AI itself. That is the pitch. That is the thesis. The verification productivity gap is not a semiconductor problem. It is an everything problem wearing a lab coat. And the founders who close it will not just build good businesses. They will quietly become the most important infrastructure layer in the AI stack that nobody on X (Formerly Twitter) has heard of. Yet. --- ### Your Calendar Is Being Mugged, and You Are Holding the Door Open [Read more](https://www.gaganmalik.io/en/newsroom/the-lost-art-of-refusal) By Gagan Malik In Fargo Season 5, there is a boardroom shot that should be stapled to every HR handbook. Lorraine Lyon (Jennifer Jason Leigh), who played a central antagonist-turned-ally as the cold, wealthy "Queen of Debt," sits at the head of the table like a final boss, and behind her is a gigantic piece of wall art that says "NO" in letters big enough to qualify as company policy. Because modern work has a weird superstition. We treat "No" like it is profanity, then we act surprised when our time, our attention, and our sanity get strip-mined by other people's priorities. I know this because I lived it. I once said "sure" to a "quick" set of stakeholder catch-ups that turned into a weekly standing meeting, plus a pre-meeting, plus a "quick sync after". That harmless politeness cost me 3 hours a week for a quarter. Call it 36 hours. At a conservative £200 an hour of fully loaded senior time, that is £7,200 of corporate value converted into vibes and posturing. This is the quiet part out loud. You are not busy. You are under-priced. #### Convenience is the drug, Yes is the side effect Tim Wu called convenience "the most underestimated and least understood force in the world today". He is right, and it is not just about Deliveroo and next-day shipping. It is about decision-making. [The New York Times](https://www.nytimes.com/2018/02/16/opinion/sunday/tyranny-convenience.html) Convenience removes friction. Friction is the moment you pause, think, and sometimes say no. Remove friction and your default becomes compliance. That is why the "Accept" button is one tap, and declining feels like you have to write a short apology letter to the United Nations. Wu quotes Evan Williams, Twitter co-founder, saying "Convenience decides everything." If that line does not chill you, you have not looked at your diary recently. [The New York Times](https://www.nytimes.com/2018/02/16/opinion/sunday/tyranny-convenience.html) Your calendar is not a productivity tool anymore. It is a demand-capture system. #### The economics of being "nice" Businesses love your reluctance to say no because it is free labour. Your politeness subsidises their ambiguity. The corporate version is what I call the "infinite backlog scam". There is always more work than capacity, so the system survives by getting you to personally absorb the overflow. You call it being helpful. Finance calls it margin. And then you wonder why you cannot do deep work, why your projects ship late, and why you feel permanently behind. Some workplaces institutionalise this. In one analysis of "Yes culture", research cited there notes that 55% of companies deny employees true autonomy. If that is even directionally accurate, then the average employee is not a knowledge worker. They are a highly educated order-taker with a laptop. [LinkedIn](https://www.linkedin.com/pulse/culture-yes-just-say-sarah-hendricks) That is not a talent strategy. That is a burnout factory. #### Why ghosting replaced "No" Now we get to the most dystopian part. Ghosting has become more convenient than saying no. Not just in dating. In hiring, in sales, in partnerships, in internal decision-making. People disappear because it is the cheapest refusal on the market. Here is the logic. Saying "No" has a social cost. You risk being disliked. You risk conflict. You risk being seen as "difficult". Ghosting has a lower immediate cost. You say nothing, you feel nothing, you move on. Convenience culture makes this worse because it trains everyone to expect instant agreement. So when you do not want to agree, you pick the path that avoids the emotional admin. Silence. [The New York Times](https://www.nytimes.com/2018/02/16/opinion/sunday/tyranny-convenience.html) Ghosting is what happens when a society loses the language of clean refusal. It is cowardice, packaged as efficiency. It also destroys trust, which is hilarious because trust is the only thing that scales. #### The "No" advantage Warren Buffett is often quoted saying that the difference between successful people and very successful people is that the very successful say no to almost everything. People hear that and think it is motivational wallpaper. [PMC/NCBI](https://pmc.ncbi.nlm.nih.gov/articles/PMC7332800/) It is not. It is basic economics. Time is finite. Attention is finite. Energy is finite. If you keep selling your finite resources for cheap approval, you will stay poor in the only currencies that matter. Saying no is not negative. It is prioritisation with a backbone. #### A framework that does not require a personality transplant You do not need to become rude. You need to become explicit. 1) **Install friction on purpose.** If the request is not obviously a "hell yes", do not answer in real time. Give yourself a buffer. That buffer is where your agency lives. 2) **Force the trade-off.** Say: "I can do this. What should I drop?" This converts vibes into prioritisation. It also reveals whether the requester actually cares. 3) **Use the Positive No.** William Ury's central point is that a strong No is anchored in a deeper Yes. You are not rejecting the person. You are protecting your priorities. "No, because I am committed to X" lands very differently from "No, I'm busy." [YourStory](https://yourstory.com/2025/05/7-books-that-teach-boundaries) 4) **Stop rewarding ghosting.** If you lead teams, treat silence as a bug, not a personality trait. Make explicit responses the norm, even if the answer is no. This is how you rebuild trust. Back to that Fargo painting. The word on the wall is not decoration. It is a boundary made visible. Most of us do the opposite. We keep our boundaries invisible, then act shocked when people walk straight through them. [Mandy Canada, "NO," painting featured in Fargo Season 5](https://ppl-ai-file-upload.s3.amazonaws.com/web/direct-files/attachments/images/26871576/e136344f-ed9a-405c-b155-7f094c0f7fcf/625861165_18252574738292400_2061853049943288809_n.jpg) Your life does not need more productivity hacks. It needs one sentence you can say without sweating. "No." --- ### In Defence of Thinking [Read more](https://www.gaganmalik.io/en/newsroom/in-defence-of-thinking) By Gagan Malik #### I Got This Wrong Last year I was building a product. Real work. The kind that requires you to hold ten variables in your head simultaneously and make judgement calls you cannot undo. I was also checking LinkedIn every forty minutes. I told myself I was staying informed. I was tracking the impressions on my last post. I was reading replies. I was doing, at a generous estimate, 45 minutes of actual thinking per day, wrapped inside eight hours of feeling busy. I did not notice for a long time. The feed is very good at feeling like work. That is the whole design. What broke the illusion was simpler than I expected. I sat down one morning to explain a technical problem to my team, a topic I had spent three weeks "following" online, and I had almost nothing to say. Not because I had not been paying attention. Because I had been paying the wrong kind of attention, to the wrong things, at the wrong depth, for a very long time. That was not distraction. That was managed ignorance, delivered at scale, with a clean UI. #### The Argument I Cannot Shake British adults now spend 7 hours and 27 minutes a day in front of screens. Ten of those minutes go to news. The rest goes to the feed. [IPA TouchPoints 2025](https://ipa.co.uk/news/touchpoints-2025) The global entertainment and media industry will hit £2.8 trillion in annual revenue by 2029. The attention economy, the business built on redirecting your cognition away from anything difficult, was worth $400 billion in 2025, up 372% since 2020. Meta alone generated $160.6 billion in advertising revenue last year. Not by selling you things. By selling your attention to people who want to sell you things. [PwC Global Entertainment & Media Outlook](https://www.prnewswire.co.uk/news-releases/global-entertainment-and-media-industry-revenues-to-hit-us3-5-trillion-by-2029--driven-by-advertising-live-events-and-video-games-pwc-global-entertainment--media-outlook-302511891.html) I keep coming back to Neil Postman, who wrote in 1985 that every communication medium reshapes culture "by promoting certain intellectual pursuits, favoring specific definitions of intelligence and wisdom". He was writing about television. He died before TikTok. But he had already described it precisely: a world where the medium does not carry the message. The medium is the message. And the message right now, running underneath everything, is that thinking is optional. [Slate — Amusing Ourselves to Death](https://slate.com/culture/2025/03/amusing-ourselves-death-neil-postman-ezra-klein.html) He feared Huxley's 'Brave New World' more than Orwell's '1984'. You do not need to ban books. You just need to make them feel like effort in a world engineered to feel effortless. [Slate — Amusing Ourselves to Death](https://slate.com/culture/2025/03/amusing-ourselves-death-neil-postman-ezra-klein.html) #### Why the Game Is Rigged Short-form content generates 70% of all user interactions across digital platforms. So platforms build for short-form. Algorithms learn that outrage and novelty hold attention longer than nuance. Advertisers pay more for held attention. Platforms earn more revenue. The cycle does not pause. [An-Nahar — economy of brain rot](https://en.annahar.com/en/international/world/261805/the-economy-of-brain-rot-when-attention-becomes-a-marketable-commodity) Thought runs on a completely different clock. Reading something serious requires sitting with uncertainty for twenty minutes before any reward arrives. No notification congratulates that patience. No algorithm surfaces it to your followers. Entertainment did not beat thought in a fair contest. It hired behavioural psychologists, ran A/B tests on human neurochemistry, and filed the results as product improvements. As Jay Caspian Kang wrote in 'The New Yorker', the structure of online political conversation, millions of petty arguments, makes genuine dialogue structurally impossible. The Sirens did not overpower Odysseus. They were simply better funded. [The New Yorker — The Sirens Call](https://www.newyorker.com/magazine/2025/01/27/the-sirens-call-chris-hayes-book-review) Among 15 to 24-year-olds in the UK, daily mobile usage is now almost five hours. The feed replaced the broadcast, and unlike the broadcast, the feed has no editor, no watershed, no obligation to the public. It has an engagement metric and an earnings call. That is the whole governance structure. [IPA TouchPoints 2025](https://ipa.co.uk/news/touchpoints-2025) #### The Counterargument I Take Seriously Access matters. A teenager with a smartphone in 2026 reaches more knowledge than an Oxford student could in 1985. The floor is genuinely higher. But the floor is not where any of us are competing. The people I watch extracting real leverage from this moment are the ones who already have strong mental models and use the tools to pressure-test them. They go into the feed with a question and come out with an answer. The rest of us go in with an hour and come out with nothing we can use. Postman noted that Reagan proved credibility no longer meant reliability, only the impression of sincerity. We have not moved past that. We have automated it. [Slate — Amusing Ourselves to Death](https://slate.com/culture/2025/03/amusing-ourselves-death-neil-postman-ezra-klein.html) Raising the floor while the ceiling keeps rising is not progress. It is a more comfortable form of the same gap. #### What I Am Actually Asking Myself We spend 4.5 hours online every day. Roughly half of that is inside products owned by two companies. We are renting our thinking to Alphabet and Meta daily, and paying for the access with our phone contract and our cognitive capacity. [Ofcom — UK adults online](https://www.ispreview.co.uk/index.php/2025/12/ofcom-finds-uk-adults-now-spend-over-4-5-hours-online-each-day.html) I am not telling you to delete the apps. I deleted mine twice. They came back, and I let them. What I am trying to do now, imperfectly, is treat serious reading like a meeting I cannot move. Ring-fence the time. Rebuild the friction deliberately, because friction is not the enemy of thought. It is the condition thought requires. Nobody is building a notification that says: 'you have been entertained enough now, please think.' That one is still on us. --- ### Gen Z Didn't Invent Knowledgemaxxing. They Just Stopped Lying About It. [Read more](https://www.gaganmalik.io/en/newsroom/gen-z-knowledgemaxxing) By Gagan Malik Here is the uncomfortable truth: everyone is knowledgemaxxing. Gen Z just turned it into a bit. Millennials turned it into burnout. Boomers turned it into a LinkedIn post in 12-point Calibri. Knowledgemaxxing is simple. It's the rebrand of something older than Wi‑Fi. Taking in more information than you can possibly action, convincing yourself it's "research", then wondering why your portfolio looks like a charity for unprofitable business models. And yes, I've done it too. In 2021, I spent 3 months "deep diving" Web3. Translation. Eight hours a day in Discords, Twitter Spaces and whitepapers. I told myself I was "building conviction". What I actually built was a 5‑figure position in a governance token that did a friendly 80% drawdown. That is a £14,300 tuition fee I paid to learn one basic lesson. More information is not the same as better decisions. #### The cult of recreational research Here is what's new. Gen Z took "learning" and turned it into lifestyle content. Polyester Zine called it "knowledgemaxxing" and framed research as a leisure activity, "undisciplined and playful" rather than graded and credentialled. TikTok is full of "Spend a research day with me" and "Topics I'm researching this week" videos. It looks like a lo‑fi study vlog. Underneath, it's a quiet revolt against brain rot and bullshit. [polyesterzine](https://www.polyesterzine.com/features/knowledgemaxxing-research-as-a-leisure-activity-and-the-gen-z-yearning-for-intellectual-growth) There is data behind the vibe. Oxford picked "brain rot" as its 2024 Word of the Year after usage spiked 230%, defining it as the supposed decline in mental state from consuming trivial content. That's not a meme. That is a generation diagnosing its own cognitive hangover. [edition.cnn](https://edition.cnn.com/2024/12/02/world/brain-rot-oxford-word-of-the-year-2024-intl-scli-gbr) But here's the contrarian bit. If you think this is a Gen Z phenomenon, you have not looked in the mirror. Millennials invented this behaviour on RSS feeds and Medium. Boomers did it with Sky News, the FT weekend and 14 "must read" business books a year. The packaging changed. The addiction to input did not. American Press Institute data says 79% of Gen Z and Millennials consume news daily and 96% at least weekly. That is not a vibes-based, feelings-first generation. That is a cohort chain‑smoking information. [thevarsity](https://thevarsity.ca/2023/09/10/opinion-gen-z-is-redirecting-the-path-to-intellectualism-not-derailing-it/) The question is not who is reading. The question is who is thinking. #### Your portfolio is a museum of half-read ideas If you want to see the cost of bad knowledgemaxxing, open your brokerage account. Every line item is an artefact from some content rabbit hole. You listened to three AI podcasts and bought the top of a SaaS name trading at 30 times sales. You read a Substack thread on "energy security" and now own a fossil stock that has done nothing but pay you a 3% dividend and emotional regret. You watched one 12‑minute YouTube explain‑like‑I'm‑five on quantum computing and bought an ETF whose top holding is a company that sells more press releases than products. I did this with "metaverse infrastructure" in 2021. The deck looked beautiful. Total addressable market allegedly in the trillions. The P&L was less beautiful. Revenue flat. R&D ballooning. Cash burn heroic. I still bought. Why? Because I had consumed so much content about "the inevitability of virtual worlds" that saying no felt like I was missing the train. Twelve months later, the position was down 72%. The market had not mispriced the future. I had mispriced my own overconfidence. This is the core bug. Knowledgemaxxing without constraint turns into narrative addiction. You think you are doing due diligence. You are actually doing mood boarding. #### Boomers do it with CNBC, Zoomers do it with Substack There is a smug take floating around that Gen Z is uniquely broken by short‑form content. Yes, TikTok has rewired attention spans. Roughly 40% of under 30s now use TikTok as a regular news source and around 60% of TikTok's billion‑plus users are Gen Z. That matters. [stackoverflow](https://stackoverflow.blog/2025/06/18/learn-like-a-lurker-gen-z-s-digital-first-lifestyle-and-the-future-of-knowledge/) But The Atlantic points out that younger Gen Z readers are also checking out more print books from libraries than older cohorts. BookTok, the corner of TikTok dedicated to books, helped push Colleen Hoover to over 2 million copies sold in a single year. NYT‑level bestsellers, off the back of teenagers crying into front‑facing cameras about paperbacks. [theatlantic](https://www.theatlantic.com/newsletters/archive/2023/12/library-gen-z-readers/676963/) So no, they are not illiterate. They just do not care about your canon. Meanwhile Boomers sit on 24‑hour news and call it "staying informed". I grew up in Delhi watching older relatives watch business channels the way teenagers now watch Twitch. Same energy. Different medium. Same output. A lot of noise, a few trades, and a deep sense that the market is unfair when it does not reward your favourite narrative. Gen Z at least has the honesty to label low‑value content as "brain rot" and then build "anti brain‑rot menus" and "doomscrolling cures" to climb out of it. Half of them openly say social media is harming teen mental health and wish they had spent less time on their phones. When was the last time a Boomer admitted CNBC made them panic sell? [stylus](https://stylus.com/pop-culture-media/knowledge-maxxing-serves-anti-brain-rot-low-stakes-intellectualism) #### The real edge isn't more knowledge. It's filters. The quiet part nobody in tech wants to say out loud in a funding meeting. You are not under‑informed. You are over‑indexed on other people's conclusions. The McKinsey and Reuters crowd will tell you Gen Z is "visual‑first" and consumes news through "edutainment" formats like GIFs and quick videos. True. But Millennials are no better. We mainlined blog summaries, Twitter threads and podcast hot takes through the 2010s and called it "staying ahead of the curve". [blog.quintype](https://blog.quintype.com/business/gen-z-news-consumption) The result. Everyone knows the same surface‑level facts. Everyone quotes the same three charts. Everyone makes the same trades a week too late. The edge now is not being the first to know. That was 1998 behaviour. The edge is being the first to stop caring about 90% of the noise. Gen Z's "commonplace books" and offline film logs are interesting because they hint at this. Taking notes by hand. Logging what actually stuck. It is not aesthetic. It is a filter. [thegoodtrade](https://www.thegoodtrade.com/features/what-is-a-commonplace-book/) Boomers had filing cabinets and cut‑out FT clippings. Same idea. The difference is that this generation is starting to recognise that curation is not something The Algorithm does for you. It is something you do yourself if you want to keep your brain and your balance sheet intact. #### How to knowledgemaxx like an adult So what do you do if you are a Millennial or Boomer who has been cosplaying as an analyst via Instagram carousels and quarterly earnings memes? Three rules. They are boring. They work. 1. Put a hard cap on inputs. If American Press Institute says 79% of your cohort is taking in news daily, assume you are already saturated. Pick two primary sources and one longform format. That's it. For me it is one newspaper, one industry newsletter, one podcast. Everything else is dessert, not dinner. [thevarsity](https://thevarsity.ca/2023/09/10/opinion-gen-z-is-redirecting-the-path-to-intellectualism-not-derailing-it/) 2. Assign every idea a price tag. Before you "go deep" on a theme, write down what it will cost you if you are wrong. In capital, in time, in emotional energy. My Web3 tuition fee was £14k. If I had written that down at the start, I would have cut my position size in half and treated the rest as speculative R&D spend, not conviction investing. 3. Separate research from entertainment. If you find yourself consuming content because it makes you feel smart, log it as entertainment, not due diligence. Entertainment is fine. Just do not let it near position sizing. Knowledgemaxxing is not the enemy. It is the default state of an economy where information is cheap and attention is the scarce resource. Gen Z's real innovation is that they have stopped pretending this is purely noble. The rest of us need to catch up. Not by learning more, but by admitting that half our "research" is just sophisticated procrastination dressed in a nice font. Once you see that, you stop trying to out‑read the market. You start trying to out‑filter it. --- ### The Curious Case of the Commute: Why Your Interview is Back in 3D [Read more](https://www.gaganmalik.io/en/newsroom/the-curious-case-of-the-commute) By Gagan Malik In 2024, nearly half of all job seekers used generative AI to game their way through applications and interviews. Half. Employers are not bringing back in-person interviews because they missed you, your firm handshake, or the small talk about your commute. They are bringing them back because your AI is getting the job, and then you have to show up and actually do it. In-person interview requests jumped from 5% to 30% in a single year. That is not a trend. That is a correction. [worklife](https://www.worklife.news/talent/return-in-person-job-interviews/) #### The Arms Race Nobody Won The received wisdom was that virtual interviews were efficient, equitable, and the future of hiring. Wrong on all three counts. What companies discovered is that they had built a system so gameable it was effectively handing offers to algorithms. Google's CEO Sundar Pichai went on the Lex Fridman podcast and said his company is reinstating in-person rounds "just to make sure the fundamentals are there". That is the world's most sophisticated tech operation saying, in plain English: we have a fraud problem, and we cannot solve it over Zoom. Brilliant self-diagnosis. Expensive lesson. [hindustantimes](https://www.hindustantimes.com/business/ai-is-forcing-the-return-of-the-in-person-job-interview-101754992294477.html) #### The Numbers Do Not Lie, But Candidates Do #### Six Percent Is a Lot A Gartner survey of 3,000 job seekers found that 6% had participated in active interview fraud, meaning someone or something else took the interview for them. Gartner also predicts that by 2028, one in four candidate profiles globally will be fake. Think of the virtual interview as a nightclub bouncer who cannot see the queue. The door is open. The bots are getting in. [hindustantimes](https://www.hindustantimes.com/business/ai-is-forcing-the-return-of-the-in-person-job-interview-101754992294477.html) #### The Silent Drop-Off Cisco added in-person rounds after repeatedly encountering final-stage candidates with "something that feels off," in the words of chief people officer Kelly Jones. What proved most diagnostic: the moment the in-person requirement gets mentioned, some candidates simply vanish. Cisco reports this happens regularly. Ghost is the new rejection letter. [hindustantimes](https://www.hindustantimes.com/business/ai-is-forcing-the-return-of-the-in-person-job-interview-101754992294477.html) #### The McKinsey Tell McKinsey began mandating at least one in-person touchpoint roughly 18 months ago, originally to better assess rapport-building skills for client-facing roles. The rise of AI cheating deepened that commitment. When McKinsey changes a process it is not for sentimental reasons. It is because the old process stopped producing reliable outputs. Follow the logic, not the nostalgia. [hindustantimes](https://www.hindustantimes.com/business/ai-is-forcing-the-return-of-the-in-person-job-interview-101754992294477.html) #### I Was Part of the Problem I will own this. Around 2022 I hired a senior creative lead on the strength of three immaculate video interviews. Articulate, sharp, camera-confident. He showed up to his first live client presentation and the client rang me privately afterwards to ask if something was wrong. Same person. Different altitude entirely. I spent the next six months and roughly £40,000 in recruitment fees, notice-period costs, and client-relationship repair learning that I had been screening for the ability to perform on camera, not the ability to do the job. I thought I was running a modern, efficient hiring process. I was running an audition for a presenter role that did not exist. The broader absurdity is this: the companies now insisting you commute in for a final-stage interview are largely the same ones who spent 2020 to 2023 outsourcing their entire hiring stack to keyword-matching algorithms that rejected candidates based on CV formatting. They created the fraud vacuum by removing human judgment from the process. Now they want you on a train at 8am to restore it. The audacity is, at minimum, consistent. [worklife](https://www.worklife.news/talent/return-in-person-job-interviews/) #### Kill, Keep, Create The future is not a wholesale return to three-panel interview rooms with weak coffee and motivational posters. Gartner data shows 68% of interviews remain virtual and that is not reversing. But the direction of travel is settled. [worklife](https://www.worklife.news/talent/return-in-person-job-interviews/) #### - Kill: The assumption that a fully virtual hiring process is inherently rigorous. It was always a cost-saving measure dressed in the language of innovation. - Keep: Virtual rounds for early screening. Fast, inclusive, no room booking required. - Create: A structured hybrid model where in-person is mandatory at the final stage, with candidate travel reimbursed. Cisco and McKinsey are already here. The FTSE 100 follows by 2027, quietly, without a press release. [hindustantimes](https://www.hindustantimes.com/business/ai-is-forcing-the-return-of-the-in-person-job-interview-101754992294477.html) #### Polish the Shoes Stop treating an in-person interview as a logistical inconvenience and start treating it as the only round that counts. Practice making eye contact without a screen as your emotional buffer. Charge the Oyster card. Rehearse the handshake. And if the prospect of sixty minutes in a room with a hiring manager, with no AI whispering answers in your earpiece, genuinely unsettles you, that reaction is the most honest performance review you will ever receive. The commute was always the warm-up. Now it is the test. Welcome back to the room. --- ### Why I Built a Personal LLM Trained on My Own Content [Read more](https://www.gaganmalik.io/en/newsroom/why-i-built-a-personal-llm) By Gagan Malik Everyone says the best AI is the most capable AI. Use the biggest model, write better prompts, and get out of the way. I think that is backwards. #### The Flattening Problem When you use a frontier LLM cold, it pulls your output toward the middle of the internet. I write with a specific voice, one that took years to develop across essays, videos and social media posts. When I fed my drafts into GPT-4 and Claude without context, the outputs were clean but flat. This is not anecdotal. Research published in 'PNAS' in 2025 found that LLMs produce systematically homogenised outputs, with each additional AI-generated piece contributing less unique diversity to a body of work than a human-authored one would. A separate peer-reviewed study found that instruction-tuned models are trained into a particular noun-heavy, informationally dense style that actively limits their ability to mimic other writing registers. The model is not failing. It is doing exactly what it was built to do, which is serve everyone, not you. [pnas](https://www.pnas.org/doi/10.1073/pnas.2504966122) The obvious response is that this is a prompting problem. Give it more context, write better instructions, and the outputs improve. That is true, and I tried it. Better prompts do help. But a 2025 arXiv study tested exactly this assumption and found that even with few-shot prompting, LLMs still struggle with nuanced, informal writing in blogs and forums, which is precisely the register most personal content lives in. At some point you are spending more effort teaching the model who you are than you are saving by using it. That is a broken workflow. [arxiv](https://arxiv.org/html/2509.14543v1) #### Five Years of Thinking, Finally Useful A model fine-tuned on your own work knows your intellectual history, not just your last message. I have five years of writing: essays, transcripts, notes, drafts. That is not just content. It is a record of how my thinking has moved, what I keep returning to, what I have changed my mind about. Fine-tuning a model on a personal corpus is an established technique. A documented case from the academic domain achieved a cosine similarity score of 0.8 between the fine-tuned model's output and the original researcher's writing style, demonstrating that a model can learn not just vocabulary but structural and stylistic patterns from a focused dataset. It made my own archive useful in a way search never could. [blog.gopenai](https://blog.gopenai.com/fine-tuning-large-language-models-to-replicate-scholarly-writing-styles-a-step-by-step-guide-945ac3a6d90f) Some people will point out that retrieval-augmented generation does the same job without fine-tuning. You build a RAG pipeline over your notes and let the model query it. That gets you halfway there. But as both AWS and independent technical analysis confirm, RAG and fine-tuning solve different problems: RAG retrieves existing information, while fine-tuning changes the model's generative behaviour at the level of tone, rhythm and structure. Those are different problems and they need different solutions. A hybrid of both is often where the real value sits. [kairntech](https://kairntech.com/blog/articles/retrieval-augmented-generation-vs-fine-tuning-choosing-the-right-approach/) #### How I Started Sounding Like Everyone Else The longer you write with a generic model, the more you start writing like one. I noticed this over about eighteen months. My drafts got smoother. More structured. Less surprising. This is documented. A Cornell University study published in April 2025 found that when people used an AI writing assistant, their writing converged, with distinct cultural and individual voices becoming more similar. The lead researcher described it as one of the first studies to show that AI use in writing could lead to cultural stereotyping and language homogenisation. A separate study found that teachers rated AI-assisted essays as more fluent and well-structured but thinner in voice and original insight. Designers know this pattern already: you become the tools you use. [news.cornell](https://news.cornell.edu/stories/2025/04/ai-suggestions-make-writing-more-generic-western) The easy answer is that this is a discipline problem. Write your first draft without AI, stay intentional, and the drift stops. Maybe. But the structural problem is that a generic tool has no stake in preserving your voice; it has every incentive to smooth it out. A 'New Yorker' piece from June 2025 put it plainly: large language models are designed to identify patterns within extensive datasets, producing outputs that lean toward consensus, and that tendency shapes the humans using them over time, not just the outputs. Building a model trained on your voice changes the incentive structure of the tool itself. You stop fighting the current. [newyorker](https://www.newyorker.com/culture/infinite-scroll/ai-is-homogenizing-our-thoughts) #### What A.R. Rahman Understood About Synthesis A.R. Rahman did not become the sound of Indian cinema by learning Western orchestration and stopping there. He built a recording studio in his home in Chennai, trained himself in MIDI composition when others were still working with live orchestras, and fused Carnatic classical music, Sufi devotional music and Western pop into a grammar that nobody else could replicate because nobody else had lived the same combination of influences. The tools he used were available to anyone. The synthesis was entirely his. A generic LLM gives you the tools. A personal model gives you the synthesis. #### The Objection Worth Taking Seriously The real objection is this: fine-tuning on a small personal corpus risks amplifying your weaknesses, not just your strengths. If your writing has blind spots or structural tics, a personal model learns those too. You end up with a tool that is very good at sounding like you, including the parts that could use an editor. This is a documented limitation; fine-tuned models carry an increased risk of reinforcing patterns present in the training data and do not surface references to challenge those patterns. [docs.aws.amazon](https://docs.aws.amazon.com/prescriptive-guidance/latest/retrieval-augmented-generation-options/rag-vs-fine-tuning.html) That is true. But it mistakes the use case. A personal model handles voice and continuity. It does not replace editorial challenge; that still needs to come from outside. A musician practising in their own style is not avoiding critique. They are making sure the critique lands on something genuinely theirs. #### What Happens If You Do Not The tools that matter in the next decade will be the ones that know you well enough to be genuinely useful, not just generically capable. If you keep outsourcing your voice to a model trained on everyone else's thinking, you are not augmenting yourself; you are slowly licensing your distinctiveness out. The people who build personal AI infrastructure now will own their intellectual leverage; the rest will spend the next ten years sounding like each other. [news.cornell](https://news.cornell.edu/stories/2025/04/ai-suggestions-make-writing-more-generic-western) --- ### We got accepted into Techstars [Read more](https://www.gaganmalik.io/en/newsroom/techstars-riyadh-2023-class) My cofounders [Ghassan Alabdulgader](https://www.linkedin.com/in/ghassan-alabdulgader/), [Taranvir Singh](https://www.linkedin.com/in/taranhundal/), [Dauren Shaikhin](https://www.linkedin.com/in/dauren-shaikhin-32494538/) and I have been accepted into the Techstars Accelerator Program. [Techstars Riyadh 2023 program](https://www.techstars.com/newsroom/announcing-the-riyadh-techstars-accelerator-2023-class)::pill. At Permitech, we're building AI‑powered industrial workplace safety solutions that improve risk detection, incident prevention, and compliance in high‑hazard environments. It's a validation of the problem we're solving and the team we've built—and a real opportunity to accelerate our path in one of the Middle East's most dynamic startup ecosystems. [Techstars Portfolio · Permitech](https://www.techstars.com/portfolio?q=permitech)::pill. --- --- ## Page metadata Title: About — Gagan Malik | Product Design Executive Description: Gagan Malik — Product Design Executive turning design organizations into revenue engines. 100M+ users, $3B+ revenue influenced. Fractional CXO, board advisory, strategic consulting. Contact via email, LinkedIn, or Ask page. Title: Pricing Description: How much does fractional product design cost here? Plans from Starter to Max (monthly subscriptions) plus a one-day Quick-wins option—see compare table and FAQs. Gagan Malik, Product Design Executive at gaganmalik.io.