Three Septembers

1993 and the Eternal September

Before 1993 the internet was closed to commerce. A few hundred thousand of the smartest people alive, mostly .edu addresses, no ads, no customers, and a culture with actual rules. The town square was Usenet, and Usenet had a rhythm. Every September a new crop of college freshmen got their first accounts, stumbled in, broke the etiquette for a few weeks, and got trained by the regulars. By October they were locals. The system could absorb a September. It had been absorbing one a year for a decade.

Then in September 1993 AOL pointed its subscribers at Usenet, and the September never ended. The freshmen kept coming, month after month, and there weren't enough regulars left to train them. The old crowd called it the Eternal September, and they meant it as a funeral. The smart-people internet didn't die. It got buried under the one everybody uses.

The same year Mosaic shipped, the first browser a normal person could use. The guy who co-wrote it was 22. Within a year it became Netscape, and the door AOL had cracked open got torn off the frame. The press was certain it wouldn't matter. Credit cards would never go online. Nobody would use their real name. In 1993 the internet had roughly fourteen million users, and fourteen million sounded enormous. Today it has most of the planet, and the people who mourned the old Usenet were mourning a rounding error.

Facebook, 2004 to 2006, and the .edu Gate

I missed 1993. I was nine. The next one I was in.

In 2004 I was a business student at a small college in New Hampshire and Facebook showed up. It launched at Harvard in February 2004 and spread school by school, and when it reached ours you needed a .edu address to get in. Mine came on July 23, 2005, back when it was still thefacebook.com. Facebook still keeps the date. I checked this week and it told me exactly how many days it's been.

It's hard to explain to anyone who arrived later how small and how good it was. There was no feed. There was a wall you wrote on and profiles you checked one at a time, and that was the whole product. Everybody on it was a real person with a real name at a real school, because the gate made lying pointless. For two and a half years it stayed that way, and inside the gate you could see what no user count tells you. Nobody could put it down. Between classes, in the library, in the dining hall, everyone was on it. And the people who weren't in yet kept asking how to get in. Not whether it was good. How to get in.

That is what product/market fit looks like from the user's chair, before anyone publishes a number. The market was pulling the product out of the company.

Facebook opened to anyone with an email address on September 26, 2006. Its own September, thirteen years after the first. It had roughly twelve million users that year. It has about three billion now, and it swallowed the media, the news, and half of how people talk to each other. MySpace, which had more users than Facebook the day the gate opened, sold for $35 million in 2011.

And the part I actually remember: it got bigger and worse for us at the same time. News Feed launched September 5, 2006, twenty years ago today, three weeks before open registration. Suddenly everything you did was broadcast, and 700,000 of us joined a group demanding it be removed. It wasn't. It became the product. The people who showed up that fall didn't care how it used to work and never would, and there were more of them within months than there had ever been of us. The early users were the last people whose opinion mattered, which is what always happens to the early users.

Two Guys, a Chess Table, and a Stack of Business Cards

That picture is 2006, right after graduation, and it's what I did about it. In school we were running a non-motorized sports magazine and had just gotten our first business cards. I remember holding one and thinking, what if this was virtual. What if you could send it across the internet and it had more than your phone number on it. Your education, your resume, your work. What if it was networked like Facebook was networked.

The business plan I wrote for class became a company. After graduation we raised money and named it SpartX, leaning on the X for exchange. A virtual business card with your resume attached and your work hyperlinked. The slogan was your net worth is your network. Later we built a jobs search that government agencies used to place people, and in 2010 the company was acquired. I was 22 at that chess table. The grand version, exchanging cards phone to phone, was waiting on a phone that didn't exist yet, but that's its own story.

I didn't know I was building on the gated phase before the gate opened. I just knew everyone I met wanted one. I'm putting it here not because it was big, it wasn't, but because it's the receipt. I saw the pattern once from inside and acted on it, which is the only reason I trust myself to see it now. The full story, with the original video and the rest of the receipts, is in 294 Views.

September 1, 2026: AI Enters the Chart

On September 1, OpenAI connected ChatGPT to Epic, the electronic health record that holds current records for more than 300 million patients. The access is read only. Nothing is written back to the chart. UCSF Health is the pilot partner and Epic itself has said nothing about it, which is consistent with a connection made through the interfaces Epic already offers rather than a partnership. The same day, OpenAI added public data sources to the product, including PubMed, ClinicalTrials.gov, RxNorm, DailyMed, and the CMS coverage database.

For two years AI in healthcare was gated by a skill instead of a credential. There was no .edu address to show. The gate was knowing how to use the thing: how to ask, how to check the answer, how to chain the tools together, how to wire them into a workflow without breaking the rules that govern a chart. The people inside read the manual and ran their companies on it. I have been one of them, and I can tell you the gated phase felt exactly like the last one. The operators who weren't in kept asking me how to get in. Not whether it was good. How to get in.

This month the gate changed shape. A clinician in an Epic hospital doesn't have to seek AI out anymore, or learn to prompt, or open a separate tab. It shows up inside the tool they already live in, next to the chart they already read. Every clinician in an Epic hospital is about to meet AI whether they asked for it or not, and most of them never asked.

A read-only pilot is not a flood. Neither was dial-up in 1993, and Facebook in 2006 was a wall you wrote on. The gate opening is the event, not the bandwidth. If the first two Septembers are any guide, what comes next isn't gradual. It's everybody, all at once, and it doesn't go back.

Build Inside. Design for the Flood.

You build while the gate is still closed. Being early is worth nothing on its own. Millions of people were early to the internet and did nothing but post. Mosaic was worth nothing until Netscape gave it a company and distribution. The .edu phase was worth nothing to the people who watched it, and I knew plenty who watched it. What lasts is what you build with the head start, because the head start is the only part that expires. At 22 I did that by accident. This time I did it on purpose. CultureAI runs my companies on first principles, question every requirement, delete what nobody can defend, automate last, and it was built and installed before this month arrived. Not because I predicted the date. Because the whole point of a gated phase is that you don't get to know the date.

When the gate opens, the thing gets bigger and worse for the early users at the same time. The people who showed up on Facebook in 2006 didn't care how it used to work and never would. The same is about to happen to AI in healthcare. The nurse in a driveway and the daughter managing her father's care between shifts are not going to read a manual, and there are going to be a lot more of them than there are of me. The tools that survive the flood are the ones that never needed the craft.

The inventor is the worst forecaster. Edison was certain the phonograph would be used for sermons. It was used for jazz. Nobody who built the early internet predicted what AOL's users would do with it, and nobody who built Facebook predicted what three billion people would do with that. Discount every confident forecast about what AI will do to healthcare, including mine, and watch what clinicians and families actually do with it.

So I'm building for them, not for people like me.

If you run a post-acute company, here is the whole application in one move: the gated phase is closing, which means the window where using this well is an advantage instead of table stakes is closing with it. Use the tools yourself, personally, before your staff meets them by default inside someone else's product. Decide your standards for what AI touches and what it never touches while that's still your decision. The operators who did that with the internet and with social media are not famous for it. They're just the ones who were ready.

Dated Calls, On the Record

Every Culture Lab field note that makes a prediction gets a line here, dated, so it can be checked. Open means not yet resolved.

Scored quarterly. When one resolves, the line gets a grade instead of an excuse.

The Ways I'm Wrong

Gates open and floods sometimes don't come. If a serious patient-safety event traceable to AI output triggers a regulatory clampdown that pulls these integrations out of the chart, the timeline breaks and the prediction with it. If Epic closes the interfaces the connection runs through, the gate re-closes from the inside, which no September in this piece had to survive. And if two years from now adoption looks like voice assistants in the exam room, a technology that opened wide and stalled anyway, then I pattern-matched on the wrong precedent and this piece becomes a good story about two Septembers instead of three. Any of those shows up, the ledger line gets marked wrong in public, which is the point of having one.