All posts

We Left Bubble. Here's Why We Bet on AI-Native Code.

We built our first three client projects on Bubble. Right call at the time — fast prototyping, visual workflows, no backend to manage. For MVPs and validation, it worked.

Then it stopped working.

Where no-code hits its ceiling

The pitch of no-code is speed: drag, drop, ship. For simple CRUD apps, that's real. But complexity doesn't scale linearly in Bubble. It compounds. Every conditional visibility rule interacts with every other one. Every API call needs a plugin or a workaround. Custom logic becomes a maze of nested conditions that nobody can debug six months later.

Hover over this to see what we mean:

The complexity ceiling

What you need
Bubble
AI + Code

Hover to see what happens as project complexity grows

Requirements keep growing. Bubble hits a ceiling. AI-assisted code keeps up.

The turning points

It wasn't one moment. It was a series:

Month 3

Plugin dependency hell

Needed a specific OAuth flow. No plugin existed. The workaround took longer than building it in code would have.

Month 5

Performance wall

Page load times hit 4+ seconds on data-heavy views. No way to optimize — Bubble controls the query layer.

Month 7

The debugging tax

A conditional visibility bug took 3 days to find. The logic was spread across 40+ elements with no search, no breakpoints, no stack trace.

Month 9

AI coding crossed the threshold

We rebuilt a Bubble feature in Next.js using Claude. It took 2 hours. The Bubble version had taken 2 weeks and was still buggy.

What AI coding tools actually change

This isn't "AI replaces developers." It's closer to: AI removes the tax on turning ideas into code. The bottleneck used to be syntax — translating your mental model into working code. That bottleneck is almost gone.

No-code workflow
Visual editor for everything — even things that shouldn't be visual
Plugin marketplace for basic functionality
Black-box hosting — can't optimize what you can't see
Vendor lock-in — no export, no portability
AI + Code workflow
Describe what you want. The code gets written.
Full access to npm, APIs, databases — no artificial limits
You own the code. Deploy anywhere. Optimize everything.
Git, tests, CI/CD — real engineering practices

The speed myth of no-code

People assume no-code is faster. For the first week, it is. But track total time over a project lifecycle — including debugging, workarounds, performance tuning, and features you couldn't build at all — and code wins. Especially now.

We measured it across our last three projects:

Project delivery time (weeks)

Same scope. Different approaches.

12 weeks 5 weeks
MVP V1 Launch V1 + Iterations V2 Scope
Bubble (time compounds)
AI + Code (time stays linear)

The gap widens with every iteration. By V2, the Bubble project spent more time on workarounds than features.

What we're not saying

Bubble isn't bad software. For visual prototyping and simple apps, it's genuinely good. We're not dunking on no-code. We used it, learned from it, and it served us well for a specific stage.

But we outgrew it. And AI coding tools made the transition from no-code to code painless instead of terrifying.

Where we landed

Our stack now: Next.js, React, SwiftUI, Cloudflare Workers, SQLite/D1. We write code with AI assistance, deploy with CI/CD, and own every line. Faster, more flexible, and — critically — debuggable. When we build native macOS apps, we get access to system APIs that no-code platforms can't touch. When we need hardware-backed security, we write Swift that talks directly to the Secure Enclave.

The irony isn't lost on us: we left a platform that promised to make coding unnecessary, and the thing that made us comfortable leaving was AI that makes coding effortless.

The future isn't no-code. It's code that writes itself when you tell it what you need.