GitHub · GH-600
Agents in the SDLC
GH-600 · Lesson 1 of 16 · 4:10
YouTube loads only after you press play. Watch on YouTube ↗
In this lesson
Prepare for the GH-600 exam (GitHub Certified: Agentic AI Developer). In this lesson: the difference between an assistant and an agent, the plan, act, and evaluate loop with signals from GitHub, which SDLC steps to give agents first, common anti-patterns (no inspectable plan, broad access, diff-only output, trusting a green build) and the contributor model for reviewing agent pull requests, and the task contract: inputs and boundaries, outputs, and verifiable success criteria. Includes three original practice questions.
Preview the topics
- Contoso Pay wants to give agents their first tasks in the lifecycle, with the least risk. Where should the team start?
- An agent’s pull request at Contoso passes every required check, but its description is empty, and nobody knows which files it was meant to touch. What is missing first?
- Contoso’s task says to fix the vulnerable dependency. The agent bumps the direct package, and the alert stays open, because a transitive dependency is still affected. What should the task contract have included?