On June 16, 2026, Framer shipped a version of its website builder that will let Claude Code, Cursor, or Codex reach into your live marketing site and edit it. Not a mockup. Not an export. The real pages your visitors load. Framer 3.0 took the number one spot on Product Hunt that day, and most of the coverage fixated on the flashy half of the release: AI agents that design pages for you on the canvas. That is the part that demos well. It is not the part that changes how a solo creator actually runs a website.
The part that matters is quieter, and it is borrowed straight from how software teams have protected production systems for twenty years.
What Framer actually put in the box
Framer 3.0 bundled three things, announced by founders Koen Bok and Jorn van Dijk. First, Agents that live on the canvas and handle 20-plus jobs: generating a page from scratch, rebuilding a layout from a screenshot, fixing responsive breakpoints, organizing styles, writing code components, managing CMS collections, drafting and improving copy, generating SEO metadata, and auditing a site for broken links and accessibility problems.
Second, Branching. You point an agent at a copy of your site, let it make changes there, review exactly what moved, compare versions, and publish only the branch you approve. Your live site never flinches while the experiment runs.
Third, a rebuilt Community with a Marketplace, galleries, awards, contests, and a social feed. Framer says it paid creators $6.5 million in 2025, up 200 percent year over year, and takes no cut of those earnings. Roughly 7,000 creators already sell templates and components there.
For scale: Framer reports 188,000-plus companies using it across 200 countries, and 364 million monthly visitors landing on more than 4 million published sites. This is not a fringe tool experimenting with AI. It is a large hosting platform changing how its sites get built. You can read the full launch note on Framer’s blog.
The MCP hook is the real headline
Here is the line that got buried. Through the Model Context Protocol, external agents like Claude Code, Cursor, and Codex can now operate inside a Framer project. If you already run one of those tools for coding work, your website stops being a separate silo you have to log into and click around. It becomes another surface your agent can read and change. Framer laid out the full agent capability list on its dedicated Agents page, and it is broader than the launch demos suggested.
Think about what that collapses. The old creator workflow for a site update went: open the builder, find the page, remember how the CMS was structured, make the edit by hand, check every breakpoint, publish. The new one is a sentence typed into an agent you already have open: “Add a testimonials section to the pricing page using the three quotes in this doc, match the existing card style, and keep it responsive.” The agent does it as native Framer work you can still inspect and refine, not as opaque code you can never touch again.
I have watched teams try to bolt AI onto content systems for years, and the thing that usually breaks is the handoff. The AI generates something in its own format, and now a human has to translate it back into the real platform. Framer skipped that step by making the agent edit the actual project. That is a genuinely different design than tools that spit out a static draft and wish you luck.
Branching is the grown-up feature, and I mean that as praise
Twenty years in IT operations taught me one reflex above all others: nobody touches production without a way to roll it back. You use a staging environment. You review the diff. You ship on approval, not on vibes. That discipline is why banks and telecoms do not fall over every time someone fixes a typo.
Consumer website builders have never had it. You edited live, you prayed, and if the AI hallucinated a broken layout at 11pm, your homepage was broken until you noticed. Handing an autonomous agent write access to a live site under those rules would have been reckless, and I would have told any client to keep the AI features switched off.
Branching is what makes the agents usable for anything that pays your bills. It is a staging environment and a diff review wearing friendlier clothes. Let the agent go wild on a branch, read what it changed, throw the branch away if it went sideways, merge it if it nailed the brief. That is the same change-management loop a fractional COO would insist on before letting any new system near revenue-generating infrastructure. Framer is the first no-code builder I have seen ship it as a first-class feature rather than an enterprise afterthought. If you take one thing from the whole release, take this: turn the agents on, but never let them publish to your live branch directly.
The credit meter deserves an honest look
Framer moved every plan onto AI credits, and this is where the peer-to-peer honesty kicks in. Credits map to what a task costs to run, so your bill tracks what you ask for. Generating a landing page runs about 300 credits. Making a page responsive is roughly 150. A large edit is around 100, a small one about 50. Full details live in Framer’s pricing note and its credits help page.
The allowances: Free gives you 500 credits a day, capped at 1,000 a month, good for about two landing pages. Basic at $10 a month gets 1,000 credits monthly, around five pages. Pro at $30 a month gets 3,000, roughly ten. Editor seats got cheaper too, cut from $40 to $20, with a $10 content-editor seat for people who only touch the CMS.
Metered AI is fair in principle, and it is also the pricing model most likely to surprise you. A creator who “just asks the agent to try a few versions” of a homepage can burn a month of Pro credits in an afternoon without noticing, because each regeneration is a fresh 300-credit hit. Budget it the way you would budget any usage-based service: know your monthly ceiling, watch the burn rate for the first two weeks, and buy add-on credits deliberately rather than in a panic mid-project. If you have read our take on the $100-a-month creator tech stack, fold Framer’s real usage cost into that math before you commit.
Where the money actually is for creators
The Community rebuild is not decoration. Seven thousand creators selling templates and components, $6.5 million paid out last year, and zero platform cut is a real income line, not a hobby. If your skill is design, Framer just made it easier to package that skill into products other people buy, with a marketplace, contests, and a reputation system attached. That is a different path to revenue than chasing brand deals, and it compounds: a good template sells while you sleep. It pairs well with the owned-audience logic we covered in how to make money as a creator in 2026.
The agents change the economics on the other side too. If building a client site used to take you three days, and an agent plus a branch review now gets you a solid draft in an afternoon, your effective hourly rate on freelance web work jumps. The creators who win here are not the ones who let the AI do everything; they are the ones who use it to skip the tedious middle and spend their time on the taste and strategy an agent cannot fake.
Where Framer still loses
Honesty means naming the gap. Webflow spent the first half of 2026 building the opposite strength. It shipped a native Claude MCP connector in February and an AEO product in April aimed squarely at getting your site cited by AI search engines, with its CEO calling Webflow an “agentic web marketing platform.” If your priority is being found and quoted by AI answer engines over the next two years, Webflow is currently the deeper bet. Framer is faster and more pleasant for design and conversion pages; Webflow is built for the long game of managing and optimizing a site at scale. We break the two down in detail in Webflow vs Framer, and the 3.0 release narrows the design gap without closing the optimization one.
And the agents are not magic. They are excellent at the tedious, well-defined jobs (breakpoints, alt text, broken-link audits, first drafts) and mediocre at the judgment calls that make a page actually convert. Treat them the way you would treat a fast junior designer: brief them tightly, review everything, and keep the strategy in your own head. If you already run Cursor or Copilot for code, the same instincts port over cleanly.
What I would do this week
If you have a Framer site, turn the agents on and run exactly one experiment: create a branch, ask an agent to rebuild your weakest page, and review the diff before you merge anything. You will learn more about where these tools help and where they flail in twenty minutes of that than in an hour of reading launch coverage, including this one. If you do not have a Framer site but you have been meaning to build a portfolio or a landing page, the free plan plus one afternoon is now a legitimate way to ship something real.
Just keep the branch discipline. The agents are the demo. Branching is the reason you can trust the demo with a site that matters.
Recent Posts
ChatGPT Work and Claude Cowork Just Turned AI Into Your Back Office
OpenAI's ChatGPT Work and Anthropic's Claude Cowork both landed in July 2026. A fractional COO's honest read on what a solo creator should actually hand an AI agent, and what to keep on your own desk.
Meta's Muse Image Can Put Your Public Photos in Someone Else's AI Art. No Notification.
Meta's Muse Image is an agentic AI generator that searches the web, writes code, and self-refines. It can also pull public Instagram photos into someone else's AI art with no notification. Here's...