What can Draftbit AI agents do?

An agent edits your project’s source, so it can make any change a developer could make there. That covers new screens, shared components, navigation, forms and validation, data logic, and fixes for errors it can reproduce. It can start a project from a description or work inside an app you already have, across as many files as the change needs.

It has tools that go beyond editing files. It can type check and lint the project, add or remove npm packages, update the Expo SDK version and run expo doctor, read the preview server’s logs and restart it, and manage environment variables. If you have connected REST or GraphQL APIs, it can list and call them to inspect live responses before wiring them into app code, and it can create REST endpoints or import them from an OpenAPI spec. It can set up a Draftbit-managed Supabase project, read your publish history and build logs, search the Draftbit docs, and create, update, and comment on tasks on the Tasks page. Before a sensitive or irreversible step, it stops and asks for your approval.

A few limits are worth knowing. MCP and API connectors run only while a chat thread is active, so they help the agent during a build and are unavailable to your published app at runtime. Custom MCP servers require the Pro plan. Fully delegating work to agents that keep running while you are away is listed as Coming Soon.

The agent still needs your judgment. It does not know an unstated business rule or legal requirement, and it can make a technically valid change that is wrong for your users. Give it acceptance criteria, ask for one working flow at a time, and review anything that touches security, data, or billing before you save.