Codex Sites Lands With Eight Builder Partners. Webflow's Bet Is What Happens Next.

OpenAI launched Codex Sites today with Webflow as one of eight partners. Here's what prompt-to-live-site means for developers, and why the Cloudflare connection is the part nobody's talking about.

Last week I said the floor of the web-building market was eroding. Today OpenAI shipped the thing that does the eroding.

Codex Sites turns a prompt into a live, hosted website. Webflow is one of eight partners standing next to it at launch.

The easy take is "AI builders are eating the bottom of the market." True, but boring. The part I think actually matters is quieter: the agentic web is consolidating onto infrastructure owned by two or three companies. Webflow has bet its survival on being the layer above all that. And if you look at who's moved where, that bet looks deliberate, not panicked.

What actually launched

Inside a ChatGPT Business or Enterprise workspace, Codex can now build and deploy a website for you. OpenAI hosts it, it runs on Cloudflare Workers, and you get a live URL back. Not a mockup. The real thing.

Webflow is one of eight launch partners: Vercel, Wix, Base44, Replit, Lovable, Figma, Webflow, Emergent. OpenAI calls it the start of a "sites partner ecosystem," not an exclusive. Hold onto that, because half my argument depends on it.

The bits that matter, from the docs:

• Every deployment is live. The URL goes public the moment it exists. Want to review first? You have to ask. That default tells you who this is for: people who want a result, not a workflow.
• Access control is on by default, scoped to the workspace.
• It runs on Cloudflare Workers.
• Preview-then-publish, but only if you ask for it.

One caveat I won't let the hype skip. OpenAI's own framing for Sites is internal: dashboards, planners, review boards, little tools you share inside a workspace. They're not pitching it as your next public marketing site.

So when you see the "Webflow killer" takes flying around today, remember that even OpenAI isn't claiming that yet. The threat is real. It's just pointed one ring out from where most people are aiming it.

Webflow didn't get disrupted. It repositioned

Webflow's former CTO, Allan Leinwand, is now Chief Engineering Officer at Cloudflare. He announced it himself. Sites runs on Cloudflare Workers. Webflow is an OpenAI partner.

Line those three up and the "Webflow is scrambling to survive AI" story falls apart.

This isn't a company caught flat-footed. The person who built Webflow's engineering foundation now runs engineering at the company hosting the infrastructure the whole agentic stack sits on. And Webflow had already rebranded itself "the agentic web marketing platform" months ago, back at its March acquisition of Vidoso and its Adobe partnership. The pieces were placed early, and on purpose.

My read: the three-company stack isn't a coincidence of overlapping press releases. OpenAI generates, Cloudflare hosts, Webflow runs the serious marketing layer. It's the same people who understand both sides of the integration ending up on both sides of it.

Where the facts stop and I start, plainly:

• Confirmed: Leinwand's move, Sites running on Workers, Webflow being a partner.
• My interpretation (about 60/40): that this is coordinated, rather than three companies independently landing on the obvious architecture.
• Unknown: whether the partnership ships a real "graduate your Sites deployment onto Webflow's CMS" path, or whether that's still a slide in a deck.

Confident on the facts, opinionated on what they mean, upfront about the seam.

Webflow isn't the target. Replit and Lovable are

Here's the thing the "Webflow killer" headlines get backwards. Webflow is barely in the blast radius. The companies that should have had a genuinely bad day are the ones whose entire product is the thing Codex Sites just shipped: Replit, Lovable, Bolt, v0.

Their whole pitch is prompt-to-app. OpenAI just built a competent version of that, put it inside ChatGPT, and pointed it at a captive audience. These aren't small companies, either, which is what makes it striking:

• Lovable: ~$400M ARR, around 8 million users.
• Replit: 50M+ users, chasing $1B revenue by year-end.
• Bolt: ~$40M ARR.
• v0 (Vercel): 4M+ users.

Real businesses. And every one of them sits under 50 million users while ChatGPT runs at roughly 900 million weekly. That gap is the whole story.

Framer is a slightly different case, and worth separating out. It's not a prompt-to-app tool, it's a design-led visual builder, sitting in Webflow's lane rather than Replit's. So Codex doesn't replicate it head-on. But it's arguably more exposed than Webflow on the exact axis my argument turns on: Framer and Webflow fight over the same design-led marketing work, and the market consensus is already that Framer wins on design while Webflow wins when CMS and structure matter. Strip out the design flourish and Codex can now do the quick, good-looking marketing page, which is closer to Framer's core than to Webflow's. Notably, Framer wasn't one of the eight launch partners. I wouldn't read that as a deliberate snub on the evidence I have, but the absence is interesting given how squarely it overlaps the work Sites is aimed at.

Distribution beats product, and OpenAI has the distribution

This is the part I keep coming back to. OpenAI didn't win this round by building a better builder. By most accounts the dedicated tools still produce nicer output. What OpenAI has is reach the others can't touch: close to a billion people who already open ChatGPT every week, no new app to download, no new account, no new bill. "Build me a site" is now one prompt away for all of them.

When the product is merely good enough and the distribution is 20 to 100 times larger than the whole rest of the category combined, good enough wins. Replit and Lovable have spent years and hundreds of millions earning users one signup at a time. OpenAI reaches more people before lunch than any of them have in total.

Framer makes this point cleaner than any of them, because it holds the product quality constant. Framer's own AI features already run on OpenAI's models - its translations use GPT-4o under the hood - so it's an OpenAI customer and an OpenAI competitor at the same time. Same engine. The difference is roughly 500,000 monthly users against ChatGPT's 900 million weekly, give or take three orders of magnitude. And Framer clearly knows what's coming: it's already published a "Framer vs Codex" comparison page, the kind of thing you only build when you've decided something is a threat. When the engine is identical and the only variable that's changed is who's standing in front of a billion people, the engine stops being the contest.

And it won't stop at internal dashboards. With that much usage funnelling through one prompt box, websites are an inevitable destination, not a maybe. The only real question is in what form - how good, how public-ready, how soon. The category didn't get a new competitor. It got a landlord who owns the building everyone was renting eyeballs in.

(One tell that the consolidation is already underway: Base44, one of the eight launch partners, is owned by Wix, another of the eight. The "ecosystem" is eating itself before it's finished forming.)

So where does that leave Webflow? Possibly as a paid deployment target

Here's the open question I find genuinely interesting, and I don't have a confident answer, only a hunch.

If OpenAI owns the front door and Cloudflare owns the hosting, what's Webflow's slot in that pipeline? The optimistic version writes itself: Codex generates the site, and when a team needs a real CMS, governance, and a marketing stack, they deploy onto Webflow - Webflow becomes a paid, premium deployment target you graduate into, and it charges to host the serious version. Distribution from OpenAI, money from hosting and the platform. That would be a tidy outcome for Webflow.

But this is speculation, not reporting. Nothing in the launch materials promises a direct OpenAI-to-Webflow deployment path. It's the logical shape of the partnership, given who's involved and where Leinwand sits, but "logical shape" and "shipped feature" are different things, and I've been burned assuming the first implies the second. If that integration shows up - one-click "deploy this Codex site to Webflow, managed and hosted, for a fee" - that's the moment the three-company-stack theory stops being my interpretation and becomes the obvious read. I'm watching for exactly that.

Four predictions

I'm writing these down so I can be wrong in public. That's rather the point of a prediction.

Within 18 months, "build a website" stops being a job below the enterprise line, and becomes a prompt. The brochure-site freelancer isn't being squeezed, that role is being retired. Last week I said the floor was eroding. I'll revise upward: it's gone. The only question is how long the market takes to admit it.

The agentic web centralises onto a few infrastructure owners. Everyone's watching the models. The leverage is in the pipes - Cloudflare's edge and OpenAI's distribution. Five years out, the question won't be "which builder," it'll be "whose Workers, whose workspace." The old web sold itself on decentralisation. This one is shaping up far more concentrated, and Leinwand's move is a tell.

Webflow's bet pays off, but not as a builder. As a system of record. Its value was never the canvas. It's being the structured, integration-heavy layer that AI-generated sites graduate into once a team has to run them for years. If leadership gets dragged into the prompt-to-site demo war instead, they lose. They're a decade ahead on the boring stuff and should defend it like it's the only thing they've got, because it roughly is.

At least two of today's eight partners are irrelevant within two years. Eight partners isn't an ecosystem, it's an audition. OpenAI keeps the ones that own a defensible layer - Figma owns design, Webflow the marketing layer, Vercel deployment - and quietly lets the rest drift. The launch list is a starting gun, not a finish line.

Bigger than Webflow, and messier than the tidy version

Late May was rough:

• ClickUp cut 22% of staff (~290 people), citing a "100x org" where agents outnumber employees.
• Wix cut roughly a thousand, citing an AI transition.
• Webflow cut an undisclosed number ("many," was all the memo said), with the 7am laptop lockout I wrote about last week.

The neat narrative is "the no-code category is collapsing." I don't buy it, and neither should you. ClickUp is project management, not a site builder, and the three cuts are three different bets wearing the same AI-displacement costume.

The detail that actually tells you something: Wix laid off a thousand people and is a Codex Sites partner. So "build your own AI versus partner with OpenAI" is a false choice, plenty of them are doing both. Everyone's hedging both ways, which is what it looks like when nobody - not even the people doing the cutting - knows where this lands.

What to actually do about it

If you're in a Business or Enterprise workspace, Sites is in preview now. Admin enables it, you add the plugin, you prompt @Sites. Use it for internal tools and prototypes. It's good at those, which is what it's for.

What I wouldn't do is reach for it as the marketing site your team has to run for years. That's still platform work, and my answer there hasn't moved since last week: match the tool to the job, disposable versus something a team operates over time.

What's changed is the urgency. The disposable end just got a free, instant, OpenAI-grade option overnight.

• If your value sits near that floor, today was the warning shot.
• If your value is the systems work - the CMS architecture, integrations, automation, conversion logic - today made you more valuable, not less. I'd be putting my rates up, not down.

The agentic web isn't democratising anything. It's centralising, and calling it democratisation. Worth building with that in mind.

(For the full argument on Webflow's pivot and whether the platform's still worth it, see my 29 May piece.)