All Microsoft exams
Microsoft GH-600: Developing in Agentic AI Systems
What the GH-600 exam covers and what it costs, with a free 30-question practice test and an explanation for every answer.
- Level
- Associate
- Topic areas
- 12
- Free practice questions
- 30
- Full question bank
- 100 questions
What the GH-600 exam covers
The topic areas in our GH-600 question bank, with each one's share of the 100 practice questions.
GH-600 cost and format
| Exam code | GH-600 |
|---|---|
| Certification | GitHub Certified: Agentic AI Developer |
| Vendor | Microsoft |
| Level | Associate |
| Status | New |
Last checked October 22, 2026 against the official Microsoft exam page.
GH-600 sample questions
The first 5 questions from the free practice test. Open a question to see the answer and why it is right.
Question 1Implement tool use and environment interaction
An organization wants its agents to use only approved MCP servers. Which configuration enforces this?
- AA note in the README asking developers to use the approved servers
- BAllowing servers found on the public internet at runtime without review
- CRemoving tool access from the agent pending security review
- DAn MCP allow list that limits which servers and registries the agent can add
Show answer and explanationHide answer
Correct answer: DAn allow list enforces the approved set in configuration rather than relying on developer discipline. A README note is advisory, open discovery is unsafe, and removing tools eliminates the capability the agent needs.
Question 2Perform evaluation, error analysis, and tuning
Before an agent starts a task, the team wants to know how success will be judged afterward. What should be defined?
- ANothing, because success is obvious to reviewers reading the output
- BThe agent's display name, icon, and profile color shown in the console dashboard panel
- CExpected outcomes, operational constraints, and measurable signals for the task
- DA promise that the output will look good to reviewers and match the brand
Show answer and explanationHide answer
Correct answer: CDefining outcomes, constraints, and measurable signals in advance allows the result to be evaluated objectively. Names, icons, and vague impressions do not make success testable.
Question 3Orchestrate multi-agent coordination
A coordinator assigns a planning agent, a coding agent, and a testing agent to one feature, with each handing its output to the next. Which orchestration pattern does this describe?
- AA sequential pipeline with explicit handoffs between specialized agents
- BRandom assignment of tasks with no handoff
- CEach agent working in isolation with its output never used
- DEvery agent editing the same file at the same time without coordination
Show answer and explanationHide answer
Correct answer: AA sequential pipeline with explicit handoffs is a common orchestration pattern for specialized agents. Random or isolated work and uncoordinated simultaneous edits do not produce a coherent feature.
Question 4Implement guardrails and accountability
An agent that updates release notes should only be able to write to one documentation folder in one repository. Which principle applies?
- ALeast privilege, scoping permissions and execution context to the minimum needed
- BShared administrator credentials across every repository in the organization
- CMaximum privilege granted to the agent's service account so it rarely fails
- DUnlimited write access to the entire repository tree for speed
Show answer and explanationHide answer
Correct answer: ALeast privilege limits the agent to the folder and repository it needs, which contains mistakes and misuse. Maximum privilege, shared admin credentials, and unlimited access all expand the blast radius.
Question 5Manage memory, state, and execution
An agent stores notes from earlier tasks in long-term memory, and some notes are now wrong. Which practice prevents outdated notes from influencing new work?
- ADefine expiration, pruning, and reset rules for memory entries
- BKeep every note forever with no review
- CDelete the whole memory each time an error occurs regardless of scope
- DCopy all notes into every new task
Show answer and explanationHide answer
Correct answer: AExpiration and pruning rules remove stale entries systematically, and reset rules handle major changes. Keeping everything, copying all notes, and blanket deletion either preserve stale data or lose needed context.
How to prepare for GH-600
- Read the official Microsoft outline first (exam page), so you know exactly what is in scope.
- Give the most time to Implement tool use and environment interaction and Perform evaluation, error analysis, and tuning: together they make up 36% of the questions.
- Take the free 30-question practice test and note which topics you miss.
- Go back to the weak topics, then test again. A practice score shows where to study; it does not predict your exam result.
Full GH-600 question bank
All 100 questions with answers and explanations, as a PDF, an online practice exam, or both.
Both formats together. The best value.
Instant access after payment, with 90 days of updates. Checkout is on our store, AIExamCerts.
GH-600 FAQ
Related exams
| Exam | Vendor | Level |
|---|---|---|
| AI-103: Developing AI Apps and Agents on Azure | Microsoft | Associate |
| AI-200: Developing AI Cloud Solutions on Azure | Microsoft | Associate |
| AI-901: Microsoft Azure AI Fundamentals | Microsoft | Foundational |
| AB-100: Agentic AI Business Solutions Architect | Microsoft | Expert |
| AB-410: Building Intelligent Applications | Microsoft | Associate |
| AB-620: Designing and Building Integrated AI Solutions in Copilot Studio | Microsoft | Associate |