From Bubble to AI coding: why one of the world's leading Bubble.io agencies changed how it builds

For five years, Bubble.io was our answer to almost everything.

We started Momentum in 2020 as a specialist Bubble agency. We grew a team of 25 around it, delivered more than fifty projects on it, trained hundreds of people through Momentum Academy, and sat on the Bubble certification committee. If you’ve read anything in our archive, you’ve read us at our most Bubble-convinced — and most of what we wrote was true, and much of it still is.

So this isn’t a takedown of no-code. It’s an honest account of why an agency with every commercial reason to keep selling Bubble now builds with AI coding tools — mostly Claude Code — and what that means if you own a Bubble app today.

What Bubble actually solved

Be clear about why Bubble won in the first place, because it explains what’s changed.

Traditional software development was slow and expensive. A meaningful web app meant engineers, DevOps, months of runway — and most SMEs simply couldn’t buy in. Bubble collapsed that: one capable builder could ship a real product in weeks. The trade-offs — platform lock-in, workload-unit pricing, performance ceilings, no code export — were a fair price for access to software you otherwise couldn’t afford to build.

We built a business on that trade being worth it. For years, it was.

What changed

AI coding didn’t make Bubble worse. It made the alternative radically cheaper.

The thing Bubble abstracted away — writing code — stopped being the bottleneck. With Claude Code, a small senior team ships real, production-grade software at no-code speed. Projects that took a thousand agency hours take fifty. And the output isn’t a proprietary app locked inside a platform; it’s a clean codebase in your own GitHub repository, on mainstream frameworks, that any developer — or, increasingly, your own staff with an AI beside them — can read, change and extend.

That last part matters more than the speed. The clients coming to us now aren’t asking “can you build it cheaper?” They’re asking for something Bubble structurally can’t give them: software their own team can keep evolving with AI. If an AI wrote the code, an AI can amend the code. Change requests stop being invoices.

Once we saw our own delivery numbers, the conclusion wasn’t optional. We shrank the team, retrained around AI-first development, and rebuilt our practice the same way we’d built the first one: by going deep early on the tool we believed would define the next decade.

When Bubble is still the right call

Our archive is full of us telling people not to rebuild things that work — that advice survives the transition.

Stay on Bubble if your app is stable, your costs are predictable, your team knows how to maintain it, and the platform isn’t blocking anything you actually need. A working system is an asset; rebuilding it for ideological reasons is how you set money on fire. We still maintain Bubble apps, still rescue struggling ones, and still tell some clients “don’t migrate yet.”

Start planning a move when you recognise yourself in more than one of these: workload-unit costs climbing with usage; performance you can’t tune past; integrations that fight you; features you defer because the platform makes them painful; a team that wants to use AI on the codebase and can’t, because there’s no codebase to use it on.

How a migration actually works

Bubble doesn’t export runnable code, so an honest migration is a staged rebuild — and the method matters more than the stack:

1. Map what exists. Data model, workflows, integrations, the behaviour users actually rely on. Your Bubble app becomes the richest spec you’ll ever have.
2. Rebuild AI-first. Mainstream frameworks, AI-assisted development, your repo from day one. This is where the 20-to-1 hours difference shows up.
3. Migrate the data. Scripted, verified, rehearsed before the real run.
4. Run in parallel. Old and new side by side until the new system has earned trust.
5. Cut over without losing anything. Users, links and search equity all carried across with proper redirects — we’ve done this for our own sites, not just clients’.

Then the part Bubble could never offer: we hand over a codebase your team can genuinely own, train them to work on it with AI, and stay a phone call away. No retainer required to change your own product.

The uncomfortable, useful truth

The hardest thing about this transition wasn’t technical. It was admitting that the thing we were best in the world at had stopped being the best answer. Most agencies in that position defend the old model — we watched vendors quote the same price 18 months into the AI boom and insist all future changes run through them.

We’d rather be early twice than defensive once. We were early on no-code in 2020. We’re early on AI coding now. The through-line is the same promise we started with: get capable software into the hands of businesses that couldn’t otherwise afford it — and this time, hand them the keys too.

If you’re weighing up where your Bubble app goes next, talk to us. We’ll give you the same honest read we gave ourselves.