WORKFLOW / 03
Building agent.
Follow an AI development workflow from acceptance criteria through implementation, checks and human review.
INSIDE THE WORKFLOW
Agentic software development from brief to review
Agentic development works best when a change has a concrete scope and observable acceptance criteria. Planning, implementation and review remain distinct steps, with the person able to inspect the work before accepting it.
Where it can help
- Product features and prototypes
- Maintenance and focused refactoring
- Test and documentation workflows
A generated change is not automatically production-ready. It needs repository-specific checks, review and an appropriate release process.
- 01Planner
Define the change
Convert the request into acceptance criteria and identify the part of the application that needs to change.
Result Search by name, keyboard access and an explicit empty state.
- 02Builder
Build the feature
Implement a focused change that follows the existing project conventions. This is a simulated walkthrough, not a coding environment.
Result A sample search component and filtering behaviour.
- 03Reviewer
Validate behaviour
Check the behaviour against the criteria, including edge cases. Results shown here are illustrative.
Result Review scenarios for matching, no results and keyboard use.
- 04You
Your checkpoint
A person reviews the implementation and validation before accepting the change.
Result Approval to prepare the sample delivery summary.
- 05Builder
Prepare the handoff
Summarise the change and its limitations. Deployment is a separate, explicitly authorised action.
Result A concise implementation handoff.
Connect this pattern to your work.
Start with a specific task, the right context and a clear outcome.