Skip to main content

Agent development lifecycle

Use the Agent Development Lifecycle (ADLC) to move an agent from an idea to ongoing use. The lifecycle helps you define success, build with appropriate safeguards, and assign ownership for continued improvement.

Scale review to risk​

Match the review process to who will use the agent, what it can access, and the impact of its work.

Agent typeScopeSuggested review
PersonalIndividual productivityKeep the scope narrow and test the main use case.
Team utilityInternal team workflowsDocument the design and ask a peer to review it.
EnterpriseDepartment or company-wide workflowsPlan formal evaluation, review, and launch.

Add review when an agent will be shared broadly, access sensitive or restricted data, use write tools, or support a high-impact workflow.

Follow the six stages​

Treat these stages as a guide, not as rigid release gates. Use the task guides for detailed instructions.

  1. Plan and design your agent: Define the problem, target users, and scope.
  2. Evaluate agent quality: Set success criteria and decide how to measure results.
  3. Build safe and maintainable agents: Set clear boundaries for instructions, knowledge, permissions, and write tools.
  4. Test and launch your agent: Validate behavior and permissions, then prepare users to work with the agent.
  5. Manage drafts, versions, and rollbacks: Make changes through drafts and version history.
  6. Govern, monitor, and maintain agents: Track adoption and quality, review ownership, and improve the agent as needs change.
Was this page helpful?