Best practices
Agent gives better results when you provide a clear goal, useful context, and a result that can be checked.
Write a clear request
Tell Agent what should change, where it should change, and what a correct result looks like.
Include the details that affect the work:
The page, reusable element, workflow, data type, or setting involved.
The current behaviour and the behaviour you want.
The Bubble branch to use.
Important limits or cases Agent must handle.
Anything Agent must not change.
A useful request looks like this:
On the test branch, update the signup page so users see an error when their invite code is invalid. Inspect the current workflow first. Do not change the login workflow.
Ask for a result you can check
A specific result is easier for Agent to build and easier for you to review.
Avoid:
Make onboarding better.
Use:
On the test branch, redirect new users to the setup page after they verify their email. Keep returning users on the dashboard. Test both paths in Preview.
Keep the task focused
Agent works best on one connected piece of work at a time.
Do not combine unrelated changes, such as redesigning a page, changing payments, and updating privacy rules, in one request.
For a larger feature, use clear stages:
Ask Agent to inspect the current app.
Review its plan.
Let Agent make the change on a development branch.
Ask Agent to test the result.
Review the work before merging or deploying.
Use Plan before Build
Start in Plan mode when the task is large, unclear, or risky.
Ask Agent to explain the current behaviour, propose an approach, and identify what could break. Switch to Build after the plan is specific enough to proceed.
Give Agent the right context
Attach information that Agent cannot get from the Bubble project itself. Optionally, connect a custom MCP in the Integrations page to let Buildprint access that information itself.
This can include a design, error message, customer report, specification, or example data. Use a Skill when you want Agent to follow the same instructions across several tasks.
Do not add unrelated files or long background notes. More context is only helpful when it affects the task.
Review and test the result
Do not treat a completed Agent turn as proof that the change is correct.
Read the summary of what changed.
Check the affected pages and workflows in Bubble.
Test the normal path and important failure paths.
Review changes involving privacy, permissions, payments, authentication, APIs, or data structure with extra care.
Start again when the context is messy
Start a new conversation if Agent misunderstands the task, repeats itself, or mixes unrelated work together.
In the new conversation, include the goal, branch, relevant previous work, and the exact next step. A clean summary is usually better than adding more corrections to a confused conversation.