Migrating from Webflow to Framer with AI agents

If you’ve followed AI lately, you’ve probably seen an agent rebuild a website from a short prompt in record time. We wanted to see what that looks like on a real job: moving a Webflow site to Framer, the kind of project agencies still quote at one to six weeks.

The agent took 85 minutes for a five-page blog and 128 minutes for an eight-page studio site with sliders, videos and a full-screen menu. It had a rough version of each site within half an hour. The rest went into comparing every page with the original and rebuilding the animations. While the agent does most of the heavy lifting, it still leaves the site unfinished. Whether it’s the forms, the animations or the responsive layouts, a person still needs the rest of the day to review it. Rebuilding by hand used to take days. Now it can be done in a day, without compromising on quality.

The agent used our free Framer skills. They are open-source instructions for AI agents that build in Framer, written from real builds: the steps that can’t be undone, the calls that report success and do nothing, the layouts that only break on a phone. Get them on GitHub and give them to Claude Code, Codex, Cursor or any agent that reads skills.

We build Framer sites for a living, so keep that in mind as you read. Both migrated sites are public, so you can open them next to the Webflow originals and judge for yourself.

AI agents can now build in Framer

Framer has a set of Webflow migration guides: a step-by-step plan, a page on keeping your SEO, and one on migrating with agents. The agents guide splits the work between Framer’s own Agent, inside the editor, and external agents that connect from outside. It’s careful about the result: “Treat the Agent’s result as a first build pass, not a finished migration.”

We wanted to see how far an external agent (Claude Code, Codex, Cursor and the like) gets when it does the whole job alone, with nobody in the Framer editor. How good is the first pass, and what does it take to finish?

What we tested

We took two free Webflow cloneables and migrated each one into an empty Framer project. The agent was Claude, connected through Framer’s external agent API, which is how we run framerlabs.com on most days.

• Migration A: TechBytes Blog Cloneable by Toms Design. A blog with a home page, an articles page, tag pages, post pages and a newsletter form.

• Migration B: EMPATHY by Ty Hughey, a studio template cloned about 8,800 times. Six static pages, project and news pages from the CMS, a preloader, scrolling marquee headings, a testimonial slider, a team carousel, background videos, a full-screen menu and a contact form.

For each site the agent saved the published pages, read the CMS content off them, then followed the order of Framer’s step-by-step guide: styles first, then the CMS, components, pages, a publish, and a check at desktop, tablet and phone widths.

Framer’s guide gives the order. Our skills cover what it leaves out. Every trap the agent hit in these two migrations, such as marquees that render invisible or a form button the API refuses to link, is now written into the Framer Labs skills, with what works instead. They’re free, so the next agent doesn’t pay for the same lesson.

The studio site is twice the size of the blog and full of motion, and nearly all the extra time went into matching that motion: the preloader, the team carousel, the menu animation. The early steps took about as long on both, because the agent reused helpers it wrote for migration A, the way your team would reuse its own after a first migration.

How we ran the tests

• Agent: Claude, connected to Framer with Framer’s agent command-line tool and an API key for an empty project created for each migration.

• Source: the published Webflow sites. No Webflow export was used: the CMS content was read from the live pages.

• Timing: from the first look at the source to the last fix, the agent’s own checks included.

• Speed: Lighthouse, the engine behind PageSpeed Insights, with its phone and desktop settings, 5 runs per page, the same network for both versions.

• Results: the migration A and the migration B, on free Framer domains.

What moved, and what still needs a person

Most of the site moved as it was. The CMS kept every slug, so every URL stayed the same. The agent downloaded every image and video and uploaded it to Framer, so nothing is left on Webflow’s servers. Text styles kept their desktop, tablet and phone sizes, and the sliders and marquees are Framer’s own components and effects, which your client or their team can edit without code.

Every early publish had a few problems. The agent found them by screenshotting its own pages at three widths, and each took a minute to fix:

• The studio site’s marquees were invisible until their rotation was set to zero.

• The testimonial slider collapsed to no height.

• Some news articles ran their paragraphs together, because Webflow stored each one as a single paragraph broken by line breaks.

Then the agent compared every page with the original. It measured each text block on both versions at three widths and fixed what differed. It also found eight projects and posts on a second list page that its first capture had missed. Last came the motion, rebuilt with Webflow’s own timing:

• Native Framer. Scroll and hover animations, the preloader, film grain, the menu animation, the team carousel and the video lightbox.

• Small code overrides. A few cursor effects that Framer has no setting for, such as images that follow the mouse over the news list.

What’s left for a person is short, and it takes the rest of the day:

• Choosing where each form’s submissions go, in Framer’s editor. The agent builds the forms but can’t connect them.

• Setting the site language, which the agent can’t reach.

• Accepting a few differences: Framer lists load more items instead of showing “1 / 2” pages, and some spacing is a few pixels off.

• Checking what no screenshot covers, such as keyboard focus and contrast.

Is the Framer version faster?

Start with real visits. HTTP Archive measures Chrome visits to millions of sites every month. In August 2026, Framer and Webflow sites passed Core Web Vitals on mobile at almost the same rate, 70.6% and 71.2%. On desktop Framer was ahead, 87.1% to 82.1%.

By metric, Framer sites do better on paint and server response, and Webflow sites do better on interaction (INP), 88.6% good against 80.2% on mobile. These numbers describe how thousands of people build with each tool, not what one careful build gets.

Then our own sites. We ran Lighthouse, the test behind PageSpeed Insights, five times on three pages of each version of the blog, with its desktop settings and with its phone settings, which model a slow 4G line and a processor four times slower than a laptop’s. On desktop, the Framer version showed its first text after about half a second and the Webflow original after about 1.2 seconds. Both feel instant. On the slow phone the gap widens: 1.2 to 1.3 seconds against 2.5 to 4.3.

Before it draws anything, Webflow’s page fetches four files from four servers: its stylesheet, Google’s font loader, jQuery and the site’s script. Framer puts its styles inside the page itself, so every round trip a slow line adds costs Webflow and not Framer.

The rest of migration A’s results:

• Largest image. On the home and post pages, the Framer version’s main image finished downloading in under 2.6 seconds, then waited about as long again for its fade-in, which only starts once Framer’s script has run. It showed at 5.1 to 5.3 seconds, against 3.6 to 4.5 on Webflow, which runs the same fade-in. Before we added the animations, the two were level, so matching every animation has a cost on a phone.

• Score. Lighthouse scored the Framer version 78 to 79 on a phone and 93 to 96 on desktop, against 65 to 78 and 80 to 91 for the original.

• Layout shift. Webflow’s articles page moved around as it loaded, by 0.368 on phones, far past Google’s 0.1 limit for good. The Framer version stayed near 0.02.

Migration B, the studio site, is a heavier test, and now a fair one: both versions run the same preloader, film grain and fade-ins. On a phone, both are slow:

• Largest element. About 10 seconds on Framer and 11 to 13 on Webflow, because the preloader holds everything back on both.

• First content. Still sooner on Framer, 1.1 to 1.2 seconds against 2.3 to 3.5.

• Score. 55 to 63 against 48 to 52 on a phone, and 82 to 93 against 72 to 80 on desktop.

One caveat: Lighthouse’s browser can’t play the Framer version’s mp4 video, so that file never downloaded in its runs, while the original loaded its video in two formats. That’s why we leave page weight out of this comparison. A design this heavy is slow on any platform. If speed matters, the preloader and the grain are the first things to question.

Both Framer versions run on free custom domains from Framer, which visitors see without Framer’s editor overlay, as on a paid site. Neither has its language set yet, which costs a few accessibility points. It’s one field in Framer’s site settings.

Webflow vs Framer plans

Both sell a plan per site, billed monthly or, for less, yearly, plus seats for the people who edit it. These are the plans most marketing sites land on, at the yearly prices on Framer’s and Webflow’s pricing pages.

Priced for four typical sites, Framer is cheaper to start and cheaper with a team, and Webflow is cheaper for a content site that fits on one plan.

A small site with no CMS costs $10 on Framer Basic and $15 on Webflow Basic. A blog costs $10 on Framer, while Webflow’s Basic plan has no CMS, so the same blog needs Premium at $25. Webflow includes far more CMS room for that price, 20,000 items against Framer Pro’s 2,500, so a large content site comes out cheaper on Webflow. Add people and it swings back: a designer seat is $20 on Framer and $39 on Webflow.

Redirects decide the plan for most migrations. Framer includes them from the Pro plan, $30 a month, so if any URL changes on the way, budget for Pro. These are platform fees only. As in our Framer vs Astro comparison, none of these estimates include developer or Framer expert time.

How to keep your rankings

The biggest risk in a migration is losing search traffic, and most of it comes from changed URLs. Framer’s SEO guide covers the same ground. This is the checklist we would follow:

1. Crawl the Webflow site first: every URL, title, meta description, canonical, h1 and the sitemap.

2. Keep slugs identical where you can. In both our migrations, every URL stayed the same.

3. Redirect every URL that changes to the page that replaces it. If a whole section goes away, its old pages can point to the closest page that’s left, the home page included. Framer includes redirects from the Pro plan.

4. Carry over titles, descriptions and social images, and bind them to CMS fields on collection pages.

5. Set the site language in Framer’s site settings.

6. Check structured data, such as Article markup on posts, and the sitemap Framer generates.

7. Move every file still served from Webflow, videos and PDFs included. The agent downloads and re-uploads them too. Afterwards, search the published pages for any link still pointing at Webflow’s servers.

8. After the domain switch, submit the sitemap in Google Search Console and watch coverage and 404 errors for a few weeks.

A migration is also a chance to fix what nobody noticed. TechBytes has two h1 headings on every page, and EMPATHY marks every word of its marquees as h1. The Framer versions have one each. A person checking pages by eye rarely spots these. An agent reads the HTML of every page.

When to stay on Webflow

• You sell products through Webflow Ecommerce. Framer has no ecommerce plan.

• Your CMS is large. Framer Pro includes 2,500 items and sells more in blocks, up to 40,000. Webflow Premium includes 20,000.

• Your developers or your infrastructure rely heavily on Webflow’s API. That work has to be rebuilt, not migrated.

• Your team knows the Webflow Editor well and the site rarely changes. Staying spares them the learning curve of a new tool.

Studios that can help with a Webflow to Framer migration

You can hire a studio, a freelancer, or do it in-house with an agent, as we did here. If you go in-house, start with our free Framer skills. Framer lists studios and freelancers in its Experts directory. A few publish their prices.

Whoever you pick, ask them five things:

1. Will every URL stay the same, and who writes the redirect map?

2. How will the CMS content move, and who checks references and rich text after import?

3. What gets rebuilt by hand: interactions, forms, search?

4. Which Framer plan will the site need?

5. Who watches Search Console after launch, and for how long?

Where Framer is clearly better

Webflow has a built-in store and Framer doesn’t. Webflow also wins on a large CMS and on the price of translation. Here is where Framer is clearly ahead, each point from the two pricing pages or our own tests.

• A design canvas. Webflow styles elements through classes, the way CSS does, which gives control and takes learning. Framer styles each layer on the canvas and shares text and colour styles, so a designer who knows Figma can change a Framer site on day one. If you think visually, this is the reason to switch.

• Staging and branches for $30. Framer Pro includes a staging environment and branches with shareable previews. Every Webflow site gets a webflow.io copy to preview, but branch staging and page branching come with the Team plan at $2,500 a month. If you show your client changes before they go live, this is the biggest gap between the two pricing pages.

• Analytics included, testing for less. Every paid Framer plan shows traffic analytics, with 30 days of history on Basic and 90 on Pro. Webflow sells Analyze as an add-on from $9 a month. A/B testing starts at $50 on Framer and $299 on Webflow.

• Cheaper seats. $20 a month for a designer and $10 for a content editor, against $39 and $15 on Webflow.

• Faster first paint. In both migrations the Framer version showed its first content on a phone in about 1.2 seconds, against 2.3 to 4.3 for the Webflow originals, and Lighthouse scored it higher on every page we tested. Across real sites, Framer passes Core Web Vitals on desktop more often, 87.1% against 82.1%.

• An agent API that does the migration. Framer’s external agent API, free while in preview, migrated both test sites. Webflow also has an MCP server for agents; we didn’t test it.

If your sites are marketing pages, a blog and a team that designs, Framer costs less and gives you staging for $30. To try a migration yourself, give your agent our free Framer skills. If you’d like us to look at your Webflow site, get in touch.

Prices checked on each vendor’s pricing page on October 14, 2026. Field data from the HTTP Archive technology report for August 2026. Every source is listed below.