Bubble: our review

Bubble is used to build a web application on a visual canvas: screens, database and business rules come together with clicks. Recently, a beta AI agent writes an entire first version from a sentence, then hands control back for visual editing. The tool targets fast prototypes as much as durable internal applications.

What it actually does

Describe a need — a portal where your clients track a case, with login and notifications — and the agent sets up a database, screens and the rules linking them. The result isn't a mockup: it's a running application that you then keep adjusting by mouse, without going back through the prompt for every change. Underneath, you find the classic Bubble: a visual programming system that's existed for fifteen years, where each rule is placed as a block rather than a line of text.

Artificial intelligence, yes, but exactly where

AI steps in at two precise points: initial generation from text, and a conversational assistant that adds a feature or explains a point of the ongoing project. Everything else — business logic, workflows, the database — remains a classic visual programming system. Bubble wasn't born with AI; it received a layer of it, like most tools in this comparison.

What's at stake if you ever leave

That's the question few comparisons ask, and it deserves digging into before committing. Bubble's documentation is clear: you keep ownership of your data and can export it as CSV or through the API at any time. You don't keep the application: it "can only run on the Bubble platform," in the words of its own documentation. Migrating to another tool therefore means rebuilding the screens and rules from scratch, not transferring a code base. A guarantee exists if the vendor disappears — the source code would then be published under an open-source licence, enough to keep running the application on a Bubble server managed by you — but this safety net, rare in no-code, doesn't replace a real export. This is exactly the kind of question we dig into during an audit and advisory engagement before recommending a no-code tool to a client.

Where else it falls short

Two other points to check before signing. Standard hosting runs in the United States, with no option to choose Europe: a firm in the legal or healthcare sector, bound by a strict data-residency requirement, has to go through the Enterprise offer, quoted on request, to host elsewhere. And the privacy policy owns up to it: exchanges with Bubble's AI tools may be used to train and improve those same tools — a frank answer, but different from a tool like n8n, which rules it out.

Our recommendation

Bubble remains defensible for an application prototype or an internal tool, built quickly and without a developer. Reserve it for projects where you accept, from the start, staying with it: there's no real way out. An organisation bound by a strict data-location requirement, or that wants to keep a realistic migration option, should look at tools that separate the database from the visual layer.