Usenet used to have a September problem. Every year, new students got their first internet accounts and showed up in the newsgroups without knowing the rules. The regulars would grumble, explain the etiquette, and by October or November things calmed down.
Then, around 1993, commercial providers (AOL is the one everyone remembers) let everyone else in. The newcomers never stopped coming. The regulars called it the September that never ended, or Eternal September.
Chris Bolton wrote about it in a post about missing when the internet was fun. His point: the early internet worked partly because getting on it took effort, and newcomers were few enough to pick up the manners. Once anyone could get on, the manners didn’t survive.
Software is having its own Eternal September right now.
Writing code used to require clearing a bar: learn a language, fight a build system, read a lot of Stack Overflow. Now you describe an app to an agent and have something running in minutes. I’m not complaining from the outside. Herald, my latest app, was built with heavy AI help.
Mostly that’s good. But the rest of the software world was built for the old pace, and it’s showing. If GitHub has felt flaky this year, here’s why: in April, its CTO wrote that they’d started building toward 10x capacity in October 2025 and by February realized they needed to design for 30x, mostly because of AI agents. (GitHub even used the Eternal September comparison itself in February, for maintainers buried in AI-generated pull requests. So I’m not the first to notice.)
the hoops
Earlier this year I built Herald, a small app that shows the one event you care about next and counts down to it. The last time I published an Android app as an individual, back in 2018, the process was roughly: pay $25, fill in the listing, upload, wait a day or two.
This time, my new personal developer account had to:
- verify my legal name and address, and confirm email and phone with one-time codes
- prove I have access to a real Android phone by scanning a QR code in the Play Console app
- run a closed test with at least 12 testers opted in for 14 days straight before Production would unlock
- apply for production access and wait up to a week for review
And because Herald has a paid tier, my full address shows on the Google Play listing.
Google says the testing rule is there to protect people from low-quality apps, malware, and fraud, and I believe them. The rule is from late 2023, before agents wrote whole apps, but the problem it was made for has only grown. App releases across the App Store and Google Play were up a reported 60% in Q1 2026 compared to the year before, and AI coding tools are the leading explanation. When an app takes an afternoon to make, more people make one, and plenty of those are junk or worse.
So the gatekeeper does the only thing it can: make publishing expensive again. Not in skill this time, but in time and other people.
The regulars are now a policy team. Like before, the rules meant for the flood land on everyone, including people with a real app to ship. Finding 12 people with Android phones willing to keep a test build installed for two weeks is harder than it sounds. I was lucky enough to have a network I could lean on for that, but you don’t have to dig far to find plenty of developers miffed about this requirement. People use Google Groups to swap installs, and some sell testers by the dozen. Which tells you how well it filters the people it was meant to stop.
the shiny new thing
The flood doesn’t only come from new people. It pulls at the people who were already here.
Arc was the first browser in years people actually switched to from Chrome. Sidebar, Spaces, tabs that closed themselves after a while. People who used it wouldn’t shut up about it.
In May 2025, The Browser Company said it had stopped actively building Arc to focus on Dia, an AI-first browser. Josh Miller’s letter said Arc asked too much of most people and AI was the real opportunity. It also said Arc wasn’t being shut down.
On Oct 22, 2026, they announced Arc shuts down Dec 22. The reason given is security, with AI making browser upkeep harder and Dia built for that. I’d assumed AI was behind all of it, and it turns out it’s right there in both announcements.
Dia has since picked up a lot of what made Arc good: sidebar tabs, Spaces (now Profiles), split view. You don’t have to pay for the Arc parts, and you can ignore the AI. Still, I don’t know anyone who switched. People who loved Arc didn’t ask for an AI browser, and people who want one have plenty of options. They traded a browser people loved for one that fewer people asked for.
That’s the temptation of AI in one story. It lets you be more ambitious, which is good. But it also makes what you already have look small next to what you could build. So you start the new thing and drop the old one. Anyone with a folder of half-finished side projects knows how this goes. I have one.
google did it first
None of this is new. Google’s been doing it for years. Killed by Google lists 300+ products, Reader, Inbox, and Stadia among them.
The explanation you hear from people who worked there is promotions. Zach Lloyd, who led Google Sheets before starting Warp, wrote about this in 2022: launching something new was one of the best ways to get promoted, so new products kept appearing while existing ones were neglected or shut down. Engineers who wanted to polish what they had lost out to projects that’d look better at promotion time. One former employee said they had a name internally: “Launch, Promo, Abandon.”
My read: someone who cares builds the thing, ships it, gets rewarded, moves on. Whoever inherits it gets no credit for keeping it good, so it slowly gets worse until someone shuts it down.
AI gives everyone Google’s incentives. Starting something new used to be expensive, so you thought twice before dropping what you had. Now an agent can get you a working first version in a weekend. Every team, and every solo developer, gets to make Google’s choice again and again: keep improving the thing people use, or start the exciting new one.
what’s scarce now
Eternal September didn’t end the internet. It changed what was scarce. Before, getting online was the hard part. After, it was attention, and people willing to teach newcomers.
Software is going through the same change. Code isn’t the scarce part anymore. What’s scarce is everything around it: maintainers with time to review pull requests, app reviewers who can keep up, platforms that stay up, and teams that stick with what people already use.
So my plan is boring. Keep working on the same few things instead of starting something new every time an agent makes that look easy. Code is cheap now; staying is the hard part.