One billion tokens later

I've claimed my public Cursor profile. Cursor now publishes a handle with agent counts, model preferences, streaks, and a lifetime token tally. Mine is live at cursor.com/@gaganmalik. The number that gets forwarded in group chats is one billion lifetime tokens. That sounds like a flex until you notice what it measures: hours spent in an editor with an agent attached, not proof that any of the output was worth shipping.

As of July 2026 the dashboard reads like a lab notebook I did not know I was keeping: about a billion lifetime tokens, 231 agent runs, a longest session of four and a half hours, and a thirty-seven-day streak that ended when real life got in the way. Two hundred and thirty of those runs stayed local; one went to the cloud. Composer 2.5, Claude Sonnet 4.5, and Sonnet 5 sit at the top of the model list.

Receipts, Not Resumes

The split is not accident. Local runs keep the repo on my machine, the git history mine, the deploy path through GitHub to Vercel unchanged. I wrote about why that stack matters in Against One-Click Coding: competent code is approaching a market price of zero; taste and architectural judgment are what compound. A public profile that only showed deploy counts would miss the point. Tokens are the receipt for staying inside the medium, not a substitute for reading what shipped.

That matches how I ship this site: Cursor for intent-to-code, GitHub for traceability, Vercel for deploys. Not a one-click wrapper that owns the repo. The portfolio case study documents the same loop: deliberate commits, fast deploys, friction retained on purpose because the friction is where the decisions live.

What the Dashboard Cannot Show

The dashboard is hours logged, not judgment proved. It will tell a visitor I ran agents for four and a half hours straight. It will not tell them whether I deleted three quarters of a diff before merge, whether I rejected a plausible component because it violated the design system, or whether I rewrote a paragraph until the claim matched the warrant. Public metrics are toolchain transparency: proof that I work inside the loop, not a ranking you should hire from.

The counterargument is fair. Publishing token counts invites comparison theatre, the same leaderboard energy that makes developers optimise for vanity metrics instead of shipped quality. A billion tokens means nothing if the output is slop. The profile only earns its keep when paired with artefacts you can inspect: writings, case studies, commit history. Metrics without artefacts are a résumé line; artefacts without context are a portfolio thumbnail. You need both, and you still have to read the work.

Why I Claimed the Handle

I did not claim the handle to win an argument about AI productivity. I claimed it so anyone evaluating how I work could see the loop, not just the landing page. Hours logged buy you a seat at the keyboard; judgment is still what decides what ships.