How AEP Works

System flow

1. User and AI solve a task.
2. Team captures what worked as an AEP.
3. AEP is stored in the repository.
4. Future tasks load and apply the AEP.
5. AEP is improved after each successful run.

Simple diagram

Task Execution
   |
   v
Capture Successful Pattern
   |
   v
Save AEP in Repo
   |
   v
Apply AEP to New Task
   |
   v
Refine and Promote

Lifecycle phases in practice

1) Execution

The agent works through a task with normal prompting and constraints.

2) Extraction

After success, you extract only reusable signal:

• key constraints
• essential workflow order
• meaningful success checks

3) Persistence

You store the AEP in paths visible to the active tooling (for example, .codex/aep/ or .cursor/aep/).

4) Reapplication

On future related tasks, the agent loads the AEP before implementation and follows its structure.

5) Refinement

You update the pack based on real outcomes, especially failures and edge cases.

Why this loop matters

The loop compounds quality over time:

• less repeated setup
• faster alignment
• fewer avoidable regressions
• better cross-session consistency