Sergey Brin

The pattern they actually represent

Look at what Brin was publishing in the mid-nineties. A 1995 SIGMOD paper on detecting copied documents across large collections. Two 1997 SIGMOD papers out of Tucson, on market basket data and on association rules. His own Stanford home page states the through-line plainly: a major research interest is data mining. None of that is a search engine. Search is what fell out of it.

Then read the 1998 paper. "The Anatomy of a Large-Scale Hypertextual Web Search Engine" opens by presenting Google as a prototype of a large-scale search engine. It does not describe a plan to build one. It gives the address: the prototype, with a full text and hyperlink database of at least 24 million pages, was available at google.stanford.edu. Then it publishes the mechanism in full, in what the authors call the first such detailed public description to date.

The research and the working product sit in one document. A reader could stop at the abstract, type the address and use the thing.

The crossing is on Google's own history page: in August 1998 Sun co-founder Andy Bechtolsheim wrote Larry and Sergey a check for $100,000, and Google Inc. was officially born. Brin's Stanford home page, still on a university server, says two things at once. Ph.D. student in Computer Science at Stanford. Currently I am at Google. There is no handoff in the record. The research did not finish and then turn into a company; the two overlapped, and the overlap is the lesson.

It kept working that way. In 2004 he co-signed a letter telling public shareholders the company would fund projects with a ten per cent chance of paying off. In December 2023 his name is in the contributor list of the Gemini technical report — an author list again, twenty-five years on.

The blockage it speaks to

Read this one if your last six months are a folder of papers, courses and notes, and nothing in it opens for anybody else.

The shape is specific. You are learning the thing properly before you build with it. Every session leaves something genuinely new in your head, so none of it feels wasted, and there is always another paper. Meanwhile the work it was all for has no address and no user.

Two forces hold that in place, and neither is laziness. The first is a payout mismatch: reading returns something the same afternoon, while a rough shipped version might not get one user this month. The second is quieter. Unfinished work cannot be judged. A folder of notes has never once been ignored. Give it an address and someone can ignore it, and that is information you did not ask for.

Brin is the right study here, and not for the reason you would expect. He did not stop learning in order to build. The publication list keeps going. The Stanford page still says student. He gave the unfinished work an address while it was still unfinished, so the using ran alongside the research rather than after it.

That is the difference between a folder and a prototype. A prototype is not a finished thing. It is an unfinished thing a stranger can reach.

Three moves you can steal

Give the unfinished thing an address. One page, one link, one form, this week. The point is not polish; the 1998 paper called Google a prototype while it was indexing 24 million pages. An address turns work you are doing into work that can be used, and usage is the only feedback that is not your own opinion of your own progress.

Publish the mechanism, not the conclusion. Brin and Page did not announce that their ranking worked. They described how it worked, and said it was the first public description of its kind. Writing down how your thing works is the fastest way to find the part that does not, and it is what lets other people build on you rather than merely admire you.

Stop scheduling the crossing. Nowhere in the record is there a date where the research ends and the company starts; the home page says student and says Google in the same breath. So drop the "once I finish this course" clause. Pick one session a week where the reading has to leave behind something another person can use, however small, and let the two run in parallel.

Where the pattern breaks

The bill for this pattern is not that research bets fail. It is that a prototype meeting the public also meets people who never agreed to be in the study. Glass was built at X; by April 2012 the team was wearing it out in public, and in April 2013 the living lab opened to anyone through the Explorer Program. In 2013 the Office of the Privacy Commissioner of Canada signed a joint letter about Google Glass alongside 36 counterparts. The device found its actual job later and elsewhere: by X's own account the team moved to Google in 2018 to scale enterprise work, and again in 2019 to put the device on hands-on workers. Shipping early buys you evidence. It does not tell you who the thing is for, or what it costs a bystander who never chose it.

What to do this week

Take the thing you have been reading about and give it an address by Friday. One page, one link, one form — the smallest version a person who is not you can open and use for something. Call it a prototype in the first line. That word is load-bearing: it is what the 1998 paper called Google while Google was already answering queries.

Then write two paragraphs of how it works and put them on the same page. Not what you plan to build. The mechanism as it stands today, including the part you already know is weak.

Send it to three people who have the problem it addresses, and watch what they do with it rather than what they say about it.

None of the reading has to stop. The only change is that from this week there is a thing with an address, and the reading has somewhere to land.