Hands-on experience with coding agents
RecommendedWorking with GitHub Copilot agents, MCP servers, custom instructions and custom agents, tools and Copilot setup steps.
GitHub Certified: Agentic AI Developer (GH-600) Exam Guide
Validates knowledge of running AI coding agents in GitHub workflows, including tool use, state, evaluation, orchestration and guardrails.
GH-600 tests practical agent operations on GitHub: where agents fit in the development lifecycle, how they use tools and MCP servers, how they keep state, how you evaluate and tune them, how several agents work together, and which guardrails keep them accountable. Expect to read real configuration and logs. Start with the three official modules, then build agent workflows yourself using the docs the study guide links for each domain.
Understand the exam
Syllabus
Prepare agent architecture and SDLC processes15–20%
- Integrate agents into the software development lifecycle (SDLC)
- Define boundaries between planning, reasoning, and action
- Configure observability and control for autonomous agents
Example scenario: Your team wants an agent to pick up labelled issues and open pull requests. How would you make it produce a plan that someone approves before it changes code?
Implement tool use and environment interaction20–25%
- Select and configure agent tools
- Configure MCP servers
- Integrate agents within development environments
- Operate agents with safe execution paths and robust error handling
Example scenario: An MCP server that reaches a production database should only be available to agents in one repository. Where would you configure that, and how would you check it?
Manage memory, state, and execution10–15%
- Implement agent memory strategies
- Persist agent state and manage context drift
- Ensure continuity of agent memory and state across tools and environments
Perform evaluation, error analysis, and tuning15–20%
- Define success criteria and evaluation signals for agent tasks
- Analyze agent failures and identify root causes
- Tune agent behavior based on evaluation results
Example scenario: An agent's pull requests keep failing the same code scanning check. Which artifacts would you read, and is the root cause more likely its instructions, its tools or its context?
Orchestrate multi-agent coordination15–20%
- Operate and manage multi-agent workflows
- Configure observability for multi-agent behavior by using logs, artifacts, and operational signals
- Detect and respond to multi-agent failures and degraded behavior
- Manage the lifecycle of agents within multi-agent workflows
Example scenario: Two agents are assigned related issues that both touch the same configuration file. How would you stop them colliding?
Implement guardrails and accountability10–15%
- Define autonomy levels
- Implement guardrails and human-in-the-loop workflows
Official GH-600 study guide · Developing in Agentic AI Systems
Format
- Duration
- 2h
- Passing score
- 700 / 1000
- Delivery
- Test centre or online
- Languages
- English
Question types
Choose one answer, or select every correct option when the question says how many to pick.
CertUp example
A team wants an agent to explain its intended changes and wait for a reviewer to approve them before it edits any files. What should they configure?
Show answer
A. Separating planning from action and requiring approval in between keeps the agent from changing code before a person has reviewed the plan.
Move items into the right places on screen, such as matching a situation to the control that fits it.
CertUp example
Match each situation to the guardrail that fits it.
Show answer
Delete production data → require explicit human approval; Reads documentation → allow without approval; Two agents edit the same file → isolate them on separate branches; Growing tool list → enforce an MCP allow list.
Eligibility and booking
Sign in to Microsoft Learn with a personal Microsoft account, then schedule through Pearson VUE. Exams booked with a work or school account are lost if you leave that organization.
Make a study plan
Read the study guide and its linked docs first, then build small agent workflows so you can read their configuration and logs fluently.
Most candidates spend 35–65 hours. Plan for the low end if you already run Copilot agents, MCP servers and Actions workflows; add time if multi-agent work, guardrails or Actions YAML are new to you.
- 1Understand the examRead the syllabus and format, check you’re eligible and pick a target date.2 h
- 3Learn the core conceptsWork through one main course or book, topic by topic.24 h
- 4Practice with mock examsTake timed practice tests and review every mistake.12 h
- 5Prepare for exam dayBook the exam and check the ID and delivery rules.2 h
Learn the core concepts
Microsoft Learn has three official modules for the first domains. For the rest, the study guide links GitHub Docs pages for each skill; work through them with a test repository open.
Best official starting pointFoundations of Agentic AI in GitHubBest for agent designDesigning Agent Architecture and SDLC IntegrationBest for MCP toolingTooling, MCP, and Agent Execution EnvironmentsRead real artifacts A beta candidate found the exam far more about reading agent files, MCP configuration, logs and Actions snippets than the modules suggest. Practise with real files.
Practice with mock exams
Use the free community study guide for flashcards, labs and a mock exam, and add a paid question set if you want timed practice.
GH-600 Study Guide and Mock Exam
GH-600 Practice Pack
GH-600: Developing in Agentic AI Systems - Practice TestsAre you ready?
- You can read an agent or MCP configuration file and say what it allows
- You can explain where human approval sits in an agent's pull request flow
- You can diagnose a failed agent run from its session log
Prepare for exam day
Use the checklist for your delivery route.
Before testing from home
Before travelling to the test centre
- Run the Pearson VUE system test on the same computer and network you will use
- Check in between 30 minutes before and 15 minutes after your start time
- Use a closed, private room with a clear desk and no second monitor
- Have a government ID whose name exactly matches your certification profile
- Bring a government ID whose name exactly matches your certification profile
- Arrive early to complete check-in
- The exam lasts 120 minutes. Breaks are allowed but the clock keeps running, and you cannot return to earlier questions after a break.
- You see a pass or fail result at the end of the exam.
- If you fail, you can retake after 24 hours, then after 14 days, up to 5 attempts in 12 months.
Advice from exam takers
Expect configuration and logs, not definitions
A detailed beta write-up says most questions asked you to read YAML, JSON, CLI and session logs, agent files, MCP configuration or Actions snippets and pick the right field or reading. A commenter called it much harder than GH-300.
GitHub Actions knowledge carries a lot of weight
A candidate who scored 805 after the beta said the exam leaned heavily on Actions and YAML configuration, and a beta candidate reported many questions on Actions workflows and Copilot CLI. Brush up on workflow syntax first if it is not part of your job.