Why Your Suggestions Never Run Out

Whenever I open LinkedIn's My Network tab I stare at a grid labelled Suggestions for you. Faces, headlines truncated with ellipses, Connect buttons waiting like chores. I dismiss one card. Another face slides into the slot before my thumb has left the screen. I connect with someone I half recognise from a conference. The grid does not thin. It refills.

The grid was working exactly as built. What I was looking at is a suggestion surface: a ranked stream of people, accounts, or next media, dressed as work you can finish. Two other terms carry weight in this piece. The checklist illusion is the folk model that says you can clear the row to zero and be done for the week. Slot refill is the product behaviour that answers every action with another candidate. Put the two together and the claim follows. A suggestion surface is a growth recommendation feed, and an empty state is treated as a bug, because a feed like that succeeds by never emptying.

In Why Infinite Scroll Is a Cognitive WMD I wrote about feeds that delete the stopping cue at the bottom of a page. The ceremony deleted here is a different one: the moment when networking, or catching up, is allowed to be over. LinkedIn is only the door. Instagram, TikTok, YouTube, X, and Facebook run the same never-finished posture under different labels. The labels change. The refill does not.

It Looks Like Work You Can Finish

The folk model is professional hygiene. Clear the suggestions. Feel caught up. Treat Connect and Follow as inbox zero for your graph. That story survives because it pays out just often enough to stay believable: a useful collaborator, a creator worth following, a group that actually fits your work. It also survives because nothing on the screen contradicts it. Cards, buttons, a bounded grid on a Sunday evening: that is the visual grammar of a task list, and your hands read grammar faster than your judgement does.

Who benefits from the story sticking is not a mystery. Every invite, follow, and extra minute feeds connection rate, follow rate, and session depth, which means your Sunday ritual becomes someone else's growth ledger. Growth teams test the copy, the ranking, and the mutual-connection thumbnails that make each card feel socially urgent. A face you half recognise is worth more than a stranger, so the ranking learns to hand you faces you half recognise. Advertisers inherit a denser graph and more inventory to buy against. I have sat in rooms where "always something next" was said in the voice people use for care. Keep the suggestion inventory warm. Never leave them with nothing. Said out loud it sounds like hospitality. On the slide it is treated as the absence of negligence, which is how a folk model gets funded as product strategy. The slide asked whether the surface still had someone to offer. It never asked whether the person had finished.

A Checklist Would Be Allowed to Empty

The test that separates the two things is the zero. A shopping list, an inbox, a form: each one is permitted to reach nothing, and reaching nothing is the point. Growth surfaces treat zero as an inventory failure, and the mechanism behind that is not mysterious. LinkedIn Engineering describes People You May Know as a multi-stage funnel that processes hundreds of billions of potential connections daily. The first ranking stage pulls a few thousand candidates from an inventory of billions of items. linkedin-pymk On the My Network tab, an infinite People You May Know cohort is split into paginated pages so the next set of faces is always ready when you scroll. linkedin-my-network That is the architecture of never-done, described in the vendor's own voice.

The rhyme across platforms is product posture rather than shared code. Meta's own transparency page for Facebook People You May Know describes gathering an inventory of profiles, then ranking them by how likely you are to engage. The predictions include whether you will send another friend request in the same carousel. meta-pymk YouTube's published recommendation architecture does the same job for video. A candidate-generation stage retrieves hundreds of items from a corpus measured in millions, then a ranking stage orders what you see next. youtube-recsys-2016 Instagram, TikTok, and X run cousins of that session inventory. None of those systems are built around a finished zero. Netflix and Spotify keep a quieter version outside social, an infinite next in a catalogue that will not admit the shelf is bare. In Optimise Is Maximise With Better Manners I argued that optimising without a stop rule is maximising with better manners. Here the missing stop is the empty grid.

The Pool Is Larger Than Your Patience

Your patience is a human budget. The pool is a machine budget. A dozen Connect cards look like a manageable Sunday afternoon, and they are, which is why clearing them feels like it should count for something. Behind those twelve faces sits an inventory measured in billions of items, narrowed to a few thousand before ranking even begins. The system still processes hundreds of billions of potential connections a day. Nothing about the grid tells you that. The row you cleared was never the pool. It was the last stage of a funnel doing its job on you.

Pagination reads on screen as a courtesy for slow connections. On My Network it is also how an infinite People You May Know cohort arrives as a finite page of faces. linkedin-my-network Twelve is a number a person can hold. Billions is not, so the interface never shows you billions. You measure progress in slots cleared and the product measures success in slots refilled. Both counts rise together, which is why the mismatch survives so long without being noticed. Only one of you is permitted to reach done.

Sunday Networking That Never Clocks Out

What I gained was the feeling of productive catching-up. It was a Sunday hygiene ritual that looked like career maintenance from any angle, including my own, while invite and dwell signals shipped to a ledger I do not own. What I did not receive was a completion ceremony. There was no moment when the product told me that networking was finished for the week, and no version of the screen that could have delivered that news. The hours still looked like professional care from the outside. I closed tabs on work that had an actual end so I could "just clear a few more", as if the grid owed me a clean desk in return.

The cost lands hardest on whoever takes the instruction most seriously. You can clear cards until your thumb aches and still owe the grid another round tomorrow. The people most likely to do that are the ones with the least slack. The junior operator told to build a network before the next review. The person between roles who has been advised to stay top of mind. The hiring manager keeping a pipeline from thinning. They are not doing it wrong. They are doing it exactly as instructed, on a surface that reads their compliance as a reason to serve more. Diligence is the input the system rewards, which means the more seriously you treat the chore, the longer the chore runs. I was not failing at networking. I was feeding a surface that cannot afford to let me finish.

The Finish Line Moves When You Approach

Picture the Connect slot as a finish line painted across a track. You stride at it. Dismiss. Connect. The tape slides forward by exactly the width of one card, every time, so the gap never closes and never widens. Your quadriceps burn on schedule. Your breathing does the thing it does in the last two hundred metres, when the body spends what it has been holding back because the ending is in sight. That surge arrives on time and finds nothing to break.

The exhaustion is real and correctly earned. Only the ending is missing, which is a stranger kind of tired than running too far. A runner who collapses at the line still gets the line. Infinite scroll deleted the bottom of the page. This deletes the scrape of an empty plate. The grid is the race, the slot is the line that will not hold still, and you are not losing because you are slow.

The Strongest Case for Never-Empty Suggestions

Give the counterargument its full weight, because it is strong. Continuous suggestions are how real opportunity arrives: the job you would not have searched for, the collaborator two hops out, the creator whose name you did not know to type. An empty grid would be negligent if the pool still held someone useful to you. Left to ourselves we connect with the people we already know, which quietly reproduces whatever advantage we started with. LinkedIn publishes work on equity in network creation, on who gets recommended and who gets left out of the graph. That is not the behaviour of a company treating discovery as a carnival trick. linkedin-pymk-equity A surface that keeps offering does something no address book has ever done.

Hold that at full strength, then read what the stack is built to improve. LinkedIn Engineering credits People You May Know with building more than half of its professional graph. linkedin-pymk-graph The large-scale recommendation write-up measures the system with online A/B tests aimed at engagement and retention across an inventory scaled past a billion items. linkedin-pymk The My Network work measures itself on how fast the next page of suggestions arrives. Read those together and something is conspicuously absent. No metric in that stack improves when you finish, and no team is rewarded for the row that ends. Opportunity can be real and the empty state can still be forbidden, and both facts live in the same product at the same time. Care that cannot admit a completion state is a growth surface wearing a helpful face.

Because Emptiness Is the Bug

Suggestions never run out because an empty grid would mean the growth surface had failed at the one thing it exists to do. Refill and reshuffle are the specification, working as written, on a screen that has no way to tell you so. The discipline you think you are short of is not the variable in this equation.

A finished zero is not a fantasy of Luddites. It is a product choice: allow a completion state, or keep inventory warm forever. The opportunity case can stay true either way. What changes is whether the interface ever admits you are done.

Give the Grid a Zero

Here is what that choice looks like when you design for a bound instead of for endless inventory. An off switch that means off. Cards grouped by a reason you can read: mutual connections, companies you follow. Dismiss that sticks and downweights similar faces, without a toast that treats your no as a sales objection. A hard remainder (eight of twenty-four, updated two hours ago) and a refreshing state that schedules the next batch instead of pretending there is no end.

تنزيل

None of that deletes discovery. It deletes the checklist illusion. LinkedIn has not shipped this surface. Operators who say they care about opportunity and equity can still demand a zero the metrics do not currently reward.

Close the Grid Mid-Row

Suggestion surfaces are growth recommendation feeds, and refill and reshuffle are the design rather than evidence of a lapse in your attention. Score them as inventory serving someone else's connection and session goals, not as networking you left half finished. Until products ship a zero, whenever I close the app mid-grid the faces are still waiting when I return, and the person who gets the evening back is me.