How to Control AI Agents in GitHub | GH-600

Your AI agent writes the plan, does the work, then updates the plan to match — that's grading its own homework. Here's how to control AI agents in GitHub. Part of my GH-600 (Developing in Agentic AI Systems) series. When agents work like a team for you, you're still accountable for what they do. This video shows the GitHub guardrails that let the agent move fast without crossing the line. What you'll learn: - Why an agent-written plan isn't verification (and the 3 plan confidence levels) - Human-in-the-loop vs agentic mode — and when you actually need a plan - Pre-committing the plan so you can prove it wasn't changed - GitHub rulesets: require PRs, block force pushes, re-review after new pushes - CODEOWNERS for protected paths + read-only token permissions (contents: read) - GitHub environments with required reviewers and secrets - Record vs Control — the CCTV analogy that ties it all together Chapters: 0:00 Agents as Your Team: Who Is Accountable? 0:39 GH-600 Series & Where This Fits 1:16 Recap: Plan Template & Copilot Instructions 1:46 3 Plan Types & Their Confidence Levels 2:43 Problem: The Agent Rewrites Its Own Plan 3:16 Human-in-the-Loop vs Agentic Mode 4:28 When You Actually Need a Plan 4:49 Issue Template: Scope & Out-of-Scope 5:23 Pre-Commit the Plan (Tamper Check) 6:46 6 Controls to Keep Agents in Line 7:49 Demo: GitHub Rulesets 9:00 CODEOWNERS File for Protected Paths 9:44 Read-Only Token Permissions in YAML 10:04 Environments, Reviewers & Secrets 10:52 Record vs Control: The CCTV Analogy 12:14 Summary: How the Controls Work Together Resources mentioned: 📦 Demo repo (clone it and follow along): https://github.com/gauravkhuraana/gh600 Watch next: ▶️ Full GH-600 playlist (includes the previous videos on the plan & issue templates): https://www.youtube.com/playlist?list=PLGr9VtBrfNqg Connect: 🎓 Courses: https://gauravkhurana.com/courses 🌐 Website: https://gauravkhurana.com 📅 1:1 mentoring: https://topmate.io/gauravkhurana #GitHubCopilot #AgenticAI #GH600 #SoftwareTesting