On September 29, at its DevDay event, OpenAI announced Sites: websites you build inside ChatGPT, with plugins you can embed and MCP events that let a project-board update trigger a drafted plan without anyone asking. According to OpenAI’s own recap, the plugin-and-automation layer for Sites is available to Business, Enterprise, Healthcare, and Edu plans. Not Plus. Not Pro. Not the 1.2 billion weekly users the company cites in its own press materials.

That is the most revealing sentence in the whole release, and almost nobody read it that way.

A website builder you cannot buy

The reaction to the DevDay firehose — the always-on agents, Ultrafast at 300 tokens per second, the Space office suite, a plugin directory living inside ChatGPT — has been a familiar two-step. The enthusiasts announced the web is over. The skeptics announced a walled garden, a land grab, the enclosure of the commons.

Both camps are describing a consumer product. Sites is not one. It is an enterprise SKU, and the tier list is the tell. Individual subscribers cannot buy the thing. The person who actually wants to build a site with AI — the freelancer, the one-man plumbing outfit with a Squarespace login he’s forgotten — is not the customer. The customer is the employer of someone who already has a ChatGPT tab open at work.

That is a far more consequential move than “OpenAI ships a website builder,” and it is boring enough that the commentariat shrugged. The Hacker News thread on Sites pulled 158 points and 186 comments. A font change would have done better.

Shadow IT just became the procurement department

Here is the mechanism, and it isn’t new — it’s just faster. Employees started using ChatGPT on personal accounts two years ago. They pasted in meeting notes, drafted contracts, built internal tools. IT found out later, usually by reading an expense report. That is shadow IT, and every enterprise software company in history has tried to stamp it out, because it is how you lose a renewal.

OpenAI has decided to monetize it instead.

Sites gives the CIO a reason to sign. The user gets a tool. The buyer gets a place for company knowledge to live inside a vendor they cannot easily leave. And the switching cost isn’t measured in dollars — it’s measured in person-years, every internal wiki and onboarding doc and quarterly dashboard rebuilt from scratch. The incumbents charging per seat to host documents nobody reads are the ones who should be disclosing this in their guidance, not the open-web advocates writing blog posts.

A junior IT administrator at a 400-person insurance brokerage put it to me plainly in a Slack message: “Nobody asked me. I found out we were on the Enterprise plan because I saw the invoice.”

The uncomfortable detail for the free-market right is that this is not a better product beating a worse one. It is a distribution advantage beating a procurement process. Those are different kinds of victory, and only one of them is good for anybody outside the winning company.

The walled-garden critique is defending a category that was never public

The standard objection — this silos information that used to live on the open web — assumes the material moving into Sites had a public URL. It didn’t. The onboarding doc has never been indexable. The org chart has never been linked. The quarterly dashboard sat behind a VPN, on a SharePoint instance nobody could search, in a format that broke in Safari.

What is actually being enclosed is the corporate intranet, and the losers are per-seat document vendors, not the reading public. In the short run the user wins, which is precisely what makes this hard to be indignant about — and why the indignation is so thin.

The honest version of the objection is narrower and less flattering: OpenAI is not competing on the quality of its site builder. It is competing on being the only company that can route around IT procurement entirely, because it is already installed in the daily habit of a billion people. That isn’t a moat built by engineering. It’s a moat built by default, and defaults are the hardest thing in enterprise software to displace.

The right keeps aiming at the wrong tier

For a decade, the right-of-center consensus has held that the great threat to competition is the consumer platform — the feed, the search box, the app store. Conveniently, those are the fights with subpoena power, cable hits, and a villain everyone at a dinner party already dislikes.

Meanwhile the boring tier is where consolidation actually happens. Enterprise software has never been competitive. It has been a sequence of monopolies, roughly fifteen years apiece: Lotus, then Microsoft, then Google Workspace, then Slack, then whoever wins the document layer next. Each fall got cheered as a triumph of disruption. Each replacement arrived more concentrated than the one before it, and four years later the cheerleaders discovered they had traded a landlord who raised the rent occasionally for one who raises it every quarter.

Sites is a bet that the next landlord is OpenAI. The genuinely uncomfortable part, for anyone who reads this page and nods along, is that the mechanism is a free market doing exactly what free markets do — and the political energy aimed at Big Tech is pointed confidently at a tier where the migration has already started, and where the receipts will not surface until the seats are gone.

The web will survive. The intranet won’t. Nobody will hold a hearing about it.

Sources

Dots, GPT-6.1 Sol, Ultrafast, a huge Codex upgrade, computer use in the Agents API, and ChatGPT going fully collaborative.

The bigger story? We’re moving from prompting AI to delegating work to AI.

Comment “OpenAI” and I’ll DM you my full 7-minute breakdown 🔗

#OpenAI #DevDay2026 #ChatGPT #AIAgents #TechNews | Coding Sanjai](https://www.facebook.com/100084511125064/posts/openai-just-dropped-20-announcements-at-devday-2026-heres-the-60-second-versiond/1071289679031393)