GH-600: Agentic AI Developer

Exam at a glance

What the exam validates

GH-600 validates that you can operate, integrate, supervise and govern AI agents inside a production software development lifecycle, using GitHub as the system of record and control plane. It is a judgement exam: most questions describe an agent doing something — or about to — and ask which control, file, setting or escalation path is right. You are expected to read custom agent profiles, hook configurations, MCP JSON, Copilot CLI flags and GitHub Actions YAML and spot what is missing or unsafe.

Skills measured

The weighting is fairly even, but not flat. Skill 2 — tools, MCP and execution environments — is the single heaviest area, and Skills 1, 4 and 5 each carry as much as the two smallest areas together. Memory and guardrails are the lightest, yet guardrail thinking runs through every other skill.

Bars are drawn to scale against the largest possible weighting, 25%.

Start here and final review

The control plane in one picture

Almost every GH-600 question sits somewhere on this diagram. An agent runs inside a bounded sandbox, and everything it produces flows through the same branch → pull request → checks → review path as a human's work. If you can say which box a control lives in, you can usually eliminate half the answers.

flowchart TB
  INTENT["Developer intent<br/>issue · prompt · automation trigger"] --> AGENT

  subgraph CP["GitHub — system of record and control plane"]
    direction TB
    subgraph SANDBOX["Agent sandbox — runner · firewall · scoped token"]
      direction LR
      AGENT["Agent<br/>cloud agent · CLI · custom agent"] --> TOOLS["Tools<br/>built-in · MCP servers · skills"]
    end
    AGENT --> BRANCH["Branch and commits<br/>copilot/… · signed · co-authored"]
    BRANCH --> PR["Draft pull request<br/>plan · evidence · session log"]
    PR --> CHECKS["Checks and scans<br/>CI · CodeQL · secret scanning"]
    CHECKS --> REVIEW["Human review<br/>CODEOWNERS · required approvals"]
  end

  REVIEW --> SHIP["Merge, then deploy<br/>behind an environment gate"]

Suggested study order

flowchart TB
  F["Agentic Foundations<br/>mental model and vocabulary"] --> S1["Skill 1 · Architecture and SDLC<br/>15–20%"]
  S1 --> S2["Skill 2 · Tools and MCP<br/>20–25% — the heaviest"]
  S2 --> S3["Skill 3 · Memory and state<br/>10–15%"]
  S2 --> S4["Skill 4 · Evaluation and tuning<br/>15–20%"]
  S3 --> S5["Skill 5 · Multi-agent<br/>15–20%"]
  S4 --> S5
  S5 --> S6["Skill 6 · Guardrails<br/>10–15%"]
  S6 --> CFG["Configuration reference<br/>files and flags"]
  CFG --> TIPS["Exam tips and caveats"]

The official learning path

Course GH-600T00-A is a one-day instructor-led course, and its self-paced content is split across six Microsoft Learn modules. They are short and scenario-driven; read each one alongside the skill page it maps to.