Implementation

Once research is complete and a report is marked Actionable, an implementation agent builds the change and opens a pull request for you to review.

Sandboxed execution

The agent clones your repository into a secure, sandboxed environment. It then:

1. Creates a new git branch (for example, ai-fix/billing-null-check).
2. Applies the code changes.
3. Runs any local tests, if configured.

Pull request generation

The agent pushes the branch to your remote and opens a pull request:

• Description – an AI-generated summary of the problem, the proposed change, and the research reasoning.
• Labels – PRs are automatically labeled (for example, LLM context or PostHog Agent) so reviewers know where they came from.

Asking the AI about a report

You can use the Ask AI button on any report to ask questions or give guidance to the agent. Your message kicks off a discussion task, and the agent responds in context. Use this to clarify requirements or steer the approach before any code is written.

Ask AI is available on every report, including resolved and archived ones.

Reviewing and merging

Your job is to review the code the same way you would any other pull request, then click Merge when you're happy with it. If something needs tweaking, you can chat with the implementation agent directly and it'll make the changes you ask for.

Nothing ships without you. The inbox is where you stay in control of what self-driving does.

Next step

Run into something unexpected? See how to debug it.