Bubble.io’s AI Agent feature is expanding beyond new apps to the platform’s entire installed base. Co-founder and co-CEO Emmanuel Straschnov confirmed the rollout timeline in the company’s April AMA: agents now work in blank apps and templates created after April 1st, Gold agency partners and ambassadors are getting access for existing apps over the next few weeks, and all users will have agent access by the end of June.

The expansion addresses a limitation that has dogged AI-assisted development tools broadly. Most work best on greenfield projects. Retrofitting agent capabilities into established applications with complex logic, accumulated data, and production dependencies is a harder problem.

Straschnov pushed back directly on the idea that agents are only useful for new builds. “I strongly disagree,” he said during the AMA, according to the recap. He pointed to the trajectory of coding agents like Claude, which started with simple tasks and grew to handle complex, established codebases. Bubble’s agent, he argued, is on the same arc. The vision is an agent capable of coordinating overhauls across multiple pages in large, long-running applications. “It’s not there yet,” he acknowledged. “We need to work on this.”

Visual Workflows vs. Code Generation

The AMA’s sharpest moment came when Straschnov articulated Bubble’s core competitive thesis against vibe-coding tools. The issue, he argued, isn’t the AI or the prompting experience. It’s the output format.

“Even though AI has made manipulating code easier, more fun and more approachable for humans, because you’re talking natural language, it is not the right language to describe a business application,” Straschnov said, per the AMA recap.

Bubble’s alternative is a visual programming language that describes application logic in readable steps: if a user clicks a button, create a database entry, send an email, charge a credit card. When things break in a vibe-coded app and the AI can’t fix the issue, developers get stuck in loops because the generated code is hard to read. With visual workflows, Straschnov argued, developers can see what’s happening and intervene directly.

Bubble’s AI Agent generates and edits this visual language rather than traditional code. That distinction positions the company differently from tools like Cursor, Replit, or Lovable, which generate and modify source code.

Platform Context

Straschnov also disclosed that apps across the Bubble platform processed over $1 billion in revenue last year, with individual businesses generating tens of millions annually. Mobile usage is growing quickly among new users, with the editor split between web and mobile projects getting “pretty close, actually.” HIPAA compliance is targeted for the second half of 2026 on dedicated and Enterprise plans.

The staged rollout to existing apps is Bubble’s answer to a question every agent platform faces as the market matures: can agents built for new projects handle the complexity of production systems? By June, Bubble’s entire user base will be the test case.