The short answer
I’ve been getting this question a lot lately.
Six to twelve months ago, my answer was easy: Don’t do it.
Now, I would say… “it depends.”
Mostly on one thing: who has to run the site after it launches.
AI can build a fast, good-looking site in days. I built this one that way. The hard part isn’t the launch. It’s month three, when your team needs to add a page, and it has to look like it belongs.
The models drift and the quality changes. Even if you’re sitting here on the latest model from the major AI labs, that same model seems to behave differently depending on how much compute the provider is putting towards it.
When I built a recent project with GPT 6 Astra, it was significantly better for the first three weeks it was out, and now the output isn’t even the same. Beware of that drift and build guardrails to keep your brand on track.
So you want to rebuild your site with AI, here are the questions I’d ask first
The platform is the last decision, not the first. These come before it.
- What’s making you ask? A competitor’s results on LinkedIn? A slow site? A bill? A designer who left? Each one points somewhere different. Sometimes the real problem isn’t the platform at all.
- Who edits the site after launch, and how often? One person once a month is a different site from a marketing team publishing every week. This answer decides more than any other.
- How long does it take to build a new page today? If it’s two hours in Webflow and the team is happy, that’s a good system. If it’s a day for a developer and takes a week to turn around, that’s the cost you’re trying to cut.
- What do you publish on a schedule? Case studies, blog posts, job listings, locations. Anything repeating needs a content management system people can sign into, whatever the site is built with.
- Who on your team has “the eye”? AI-built sites drift. Pages come out slightly different from each other. Someone has to set the rules and catch the drift. If nobody can, that’s a vote for Webflow with a component system.
- How big and how complicated is it? A five-page site for a consultant, maybe with a blog, is well within reach. A large site with many templates, integrations and editors is where AI-built sites hit a ceiling today.
- How much do you care about owning it? Could you move your site to another host tomorrow? Would you want to? If that matters, it changes the math.
- When is your next refresh? If it’s two years out, you have time to watch the tools mature. If it’s next quarter, decide with what exists now.
How I’d decide
Here’s where those answers usually land.
| If this sounds like you | My call | Why |
|---|---|---|
| Your team edits the site every week, on their own, and nobody has a designer’s eye | Stay in Webflow | Change a line, hit publish, done. Webflow keeps the brand intact without anyone policing it. |
| You’re staying in Webflow but pages take too long to build | Stay, and build a component library | New pages come together from parts, not a bajillion divs. Faster, and still on brand. |
| You have someone with a good eye who can set rules and check the output | An AI-built site is worth a serious look | You’ll get most of the brand consistency today, and rules keep it closer. |
| You want AI-built speed, but your team publishes on a schedule | Code the site, add a content system like Sanity | Pages live in code. Your team signs in and adds a case study the way they would in Webflow’s CMS. Nobody has to prompt anything. |
| It’s a small site that rarely changes | AI-built is a reasonable default | Fewer editors, fewer templates, less drift. Just be mindful of overused “good design” patterns. |
| You’re planning a refresh in the next year and don’t know what you need | Plan the user journey first | Where people land, where they go next, what happens after the form. The platform comes after. |
Two answers override the table. If nobody will check the output, stay where the guardrails are built in. If owning your site outright matters most, lean toward code.
What AI-built sites are good at, and bad at
I built this site with AI. It’s an Astro site, written with Claude, kept in GitHub and hosted on Cloudflare. The brief was basically “make this good enough to win an awwward.” So this part is first-hand.
Good at, today:
- Speed to launch. The animations came together quickly. Work that used to take weeks takes days.
- Performance. Clean code, no builder overhead. Fast pages, especially when hosted on something like Cloudflare.
- Being found. A help center I published here was on Google’s first page within a day, drove 10,000 impressions on Google in 3 days, and saw 120 clicks to this site. I’ll share the details on that another day.
Bad at, for now:
- Consistency. Go from one page to the next and things look slightly different. You have to fight it to keep the brand intact.
- Handoff. I haven’t found a good way to tell a client, “Here’s how you prompt your site into existence and keep it on brand.” I’m working on it.
- Quick edits. In Webflow, you change a line and hit publish. With a coded site, an edit takes longer unless you plan for it.
The gap is closing. I’d guess we’re 3 to 12 months from AI tools holding 90% of a brand on their own for most frontend web design, but there are still major challenges at scale.
The lock-in question
Webflow sees this coming. It’s building AI into its own tools, and it already charges for AI credits. Its next-generation editor, Source, is in beta. They’re saying it’ll be out soon but I’m leaning towards mid-2027. I have yet to get admitted into the beta so I’ll withhold saying more about it.
The open question is the bill. Hosting, plus every change you push through their AI, could become a bottomless pit. You’d be paying one vendor to edit a site that only lives with that vendor.
The other road: your site in your own GitHub repo, on your own hosting, edited with the AI subscription you already pay for. Switch tools whenever you like. That’s data portability, and you don’t fully have it in Webflow today.
Before you commit either way, ask:
- What does hosting cost, and what does each change cost on top?
- Can you use your own AI subscription, or only theirs?
- If you left next year, what would you take with you? The code? The content? Just an export?
- Who would you call if something broke on a Saturday?
What to do next
Webflow is still a good platform, especially when brand control matters. But things are heading toward AI-built sites faster than I expected.
If you’re not sure which camp you’re in, answer the eight questions above. Then bring them to a free 30-minute call. That’s usually enough to know.
If you need the full picture first, your journey, your stack and what to fix in what order, that’s the Systems Diagnostic.