[Feature update] Data type generation now available in the AI Agent

Hi everyone,

My name is Rutvij and I’m the product marketing manager for AI at Bubble.

Last month, we shared that the Bubble AI Agent could create and edit dynamic expressions from a simple prompt. Today, I’m excited to share that the Agent can now generate and modify data types and fields. It will also create privacy rules as appropriate. See it in action here (we had fun with this video!) →

This is the next step in our journey toward an AI Agent that can build every aspect of your app — design, logic, and data — all part of creating a platform where you can switch between prompting the AI Agent and making precise edits with visual tools.

What’s new

The Bubble AI Agent (beta) can now automatically create and modify data types and fields when you prompt it to build features. The Agent can update your database schema, add new data types to keep your data organized, create relationships between different data types, and add fields to existing types.

Unlike the standalone data type generation tool we released a few months ago, the Agent has the full context of your app and addresses feedback many of you shared — you wanted the ability to use AI to edit and evolve your data structure as your app grew, not just create from scratch.

Note: The Agent is only available for AI-generated apps. If your app wasn’t built with AI, you can continue to use the existing data type generation tool on the data tab. We’re working on expanding Agent access to all apps in Q1.

What this update means for your builds

You can now prompt the Agent to create or modify data types and fields.

Create new data types with built-in security

Say you’re building a gym management app. You might need to add additional data types beyond what was initially generated, like a Gym data type.

  • You can prompt: "Add a Gym data type with fields: Name (text), address (text), capacity (number), opening_hours (text).”

  • The Agent will create the Gym data type with all those fields plus appropriate privacy rules.

Modify existing data types

You can also add fields to data types that already exist:

  • Prompt: “Add description, logo, and website url fields to the Gym data type”

  • The Agent will add the new fields to your existing Gym data type.

You can immediately view and edit any generated data types and fields in the data tab. Switch between prompting and visual editing whenever you want — you’re not limited to just one way of building.

Current limitations

This release focuses on core data types functionality and has a few constraints:

  • Creating option sets isn’t supported yet

  • The Agent will not delete fields if prompted

  • The Agent cannot modify existing field types (e.g., change Salary to be a number instead of text)

What’s next

With each update, we’re getting closer to a full-stack AI Agent experience. Generating and editing UI, expressions and now data types, paves the way for workflows. We’re rolling it out this month so expect to hear more soon.

Try it out

If you have an AI-generated app, you can start using the Agent to generate data types today. We’d love to see what you build and hear your feedback in the forum.

Happy building!

— Rutvij

4 Likes

This still doesn’t feel like real progress toward “full-stack AI” until it works on non-AI apps, and until there’s clarity on when an app is still considered AI-manageable.

Right now, “AI-generated” is based on how the app started, not how it actually behaves. There’s no indicator, no warning, and no way to tell what the Agent still understands versus what it doesn’t.

That’s made worse by the Agent’s very limited effective context window. Once an app gets past a certain size, you hit the ceiling fast. Long before you can describe anything close to a real production app.

Until the Agent can work on existing apps, and until there’s some visible signal showing whether an app is still safe for the Agent to touch, this feels more like a side track than a true full-stack workflow.

Need this on non AI apps

Good points. We definitely hear you and the community and we’re working towards removing the AI-app only limitation. Additionally, I DM’d you so we can find time to talk more about this if you’re interested.

2 Likes

Probably the most important part of this. Think about the experience of adding something like the US states. Extremely common across apps.

Manually inputting them one by one, sorting them one by one with up/down arrow clicks, etc. It feels like it’s out of the 1990s.

Just use API Objects instead. It’s already known API Objects are great for such things and that LLM can create quickly an entire JSON array for them.

Yeah I don’t see any downsides to this.

It’s that you lose the option set global accessibility in dynamic expressions that make them easy to work with…other than that, use of api objects for such things only has upsides

Hmm yeah option sets are pretty cool.