A hands-on course for experienced developers who want to delegate well-scoped engineering work to OpenAI Codex and verify the results. Practice a disciplined review, repair, and validation loop using a real application repository.

Level: Expert. Estimated study time: 20 hours including labs. Prerequisites: a programming language, Git, command-line tools, and automated testing.

Course outline

1. Task design and setup (2 hours): define scope, acceptance checks, and a reproducible development environment.
2. Repository context and instructions (3 hours): inspect architecture, select relevant context, and write concise project guidance.
3. Feature implementation (4 hours): plan a change, develop it incrementally, and inspect diffs.
4. Repair and validation loops (4 hours): reproduce failures, add regression tests, repair code, and rerun checks.
5. Review and safe delegation (3 hours): assess findings, protect secrets, limit permissions, and document uncertainty.
6. Capstone (4 hours): deliver a tested feature or modernization change with a review-ready handoff.

Learning outcomes

• Define agent tasks with clear boundaries and measurable outcomes.
• Use test feedback to improve an implementation.
• Evaluate generated changes and present evidence for review.

Independent Eustad training; not affiliated with or endorsed by OpenAI. Codex access, subscriptions, and usage charges are separate from tuition. Estimated duration includes independent practice. Features may change.

Technical reference: OpenAI: iterative repair loops with Codex.