Hello, you are using an old browser that's unsafe and no longer supported. Please consider updating your browser to a newer version, or downloading a modern browser.
GitHub Certified: Agentic AI Developer is the GitHub credential for people who run AI agents inside a real development workflow. This October 2026 guide covers the GH-600 exam, its six domains and their weights, the 120-minute format and passing score, the $165 fee, how the exam came out of beta, and the roles it fits.
GitHub Certified: Agentic AI Developer, earned by passing exam GH-600, tests whether you can put AI agents to work inside a software development lifecycle and keep them under control. The exam is titled Developing in Agentic AI Systems, which is why you'll see two names for one credential. Its subject is agents that plan, call tools, open branches and raise pull requests, with GitHub acting as the system of record and the control plane.
The credential is GitHub-branded, and Microsoft administers it. You register with a Microsoft Learn account, schedule through Pearson VUE, and the record lives in your Microsoft Learn profile. Microsoft opened GH-600 as a beta exam in May 2026, and its June 2026 credentials roundup set general availability for July. By October 2026 the certification page no longer carries a beta label.
GH-600 has no prerequisite certification, but Microsoft expects real experience first: the SDLC, GitHub workflows and controls, code review and security practices, and coding agents such as GitHub Copilot, MCP servers and custom agents. The study guide and exam page are published on Microsoft Learn.
Agent steps, success criteria, structured plans, autonomy levels and human intervention. 15 to 20%.
Tool permissions, MCP servers and allow lists, CI invocation, branch scope, rollbacks. 20 to 25%.
Short-term, long-term and external memory, durable state, context drift. 10 to 15%.
Evaluation signals, scan results, root cause analysis, instruction and tool tuning. 15 to 20%.
Orchestration patterns, isolation, conflict resolution, recovery and agent lifecycle. 15 to 20%.
Risk-based autonomy levels, human review, least privilege, policy blocks. 10 to 15%.
Four things that set the Agentic AI Developer exam apart from the rest of the GitHub track in 2026.
GH-300 tests how you use and govern Copilot while you're the one driving. GH-600 picks up once an agent opens its own branch, runs inside a CI workflow and raises a pull request with nobody typing each step. The study guide asks you to scope that agent to a repository or branch, set its tool permissions, and build the retries, rollbacks and escalation paths for the day it gets something wrong.
GH-600 runs 120 minutes, 20 more than any other GitHub exam. Microsoft also bills it at $165 in the United States, the same as its Azure role-based exams, while GH-900, GH-200, GH-300, GH-500 and GH-100 cost $99. Plan for the extra time and the extra cost before you book.
Implement guardrails and accountability carries 10 to 15 percent of the exam on its own, covering risk-based autonomy levels, least-privilege execution, and blocking actions that break security or Responsible AI policy. Our walkthrough of the OWASP Top 10 for Agentic Applications covers the failures those controls exist to stop.
Microsoft's GH-600 page says GitHub certifications are valid for two years. GitHub is moving them onto Microsoft's recertification process so holders can renew without retaking the full exam, and certifications that expire before that process is live get a six-month extension. GH-600 is not a DoD 8140 credential and does not appear on the Approved Qualifications Matrix. The current matrix is published at the DoD Cyber Exchange.
Its weight is practical instead. Teams that let coding agents open pull requests need someone who decides what those agents may touch and who reads back what they did. GH-600 is the GitHub credential built around that job.
The certification, the GH-600 exam, and what happens after you pass, as of October 2026.
GH-600 sits at the agent end of the GitHub track. Where you go next depends on whether your agents live in a DevOps pipeline, inside a security program, or on Azure. These three are the logical next steps.
Microsoft's new expert credential for AI-assisted DevOps across Azure DevOps and GitHub went into beta in October 2026, with general availability planned for December. You earn it with AZ-401 plus either GH-300 or AZ-104, so GH-600 doesn't count toward it, but the agent supervision skills carry straight over. AZ-400 retires February 1, 2027.
GH-600 asks you to generate evaluation signals with automated scanning tools. GH-500 is the exam about those tools. Its July 2026 outline covers Secret Protection, supply chain security and Code Security with CodeQL, plus security operations and administration of the security suites.
If your agents run on Azure rather than inside GitHub, AI-103 covers building them with Python and Microsoft Foundry. Generative AI and agentic solutions is its heaviest domain at 30 to 35 percent. It's a 120-minute, $165 Microsoft associate exam with a free annual renewal.
GH-600 comes after the Copilot stage of the GitHub track. Microsoft's audience profile assumes Copilot experience, so most candidates arrive with GH-300 or daily Copilot use behind them, and from here the path splits into DevOps, security and cloud agent work.
GitHub and Copilot basics
Agents inside the SDLC
DevOps, security or cloud agents
Two questions to answer before you commit: can you sit it, and should you pursue GH-600 specifically. Here's a straight answer to both.
Anyone can register and sit GH-600.
GH-600 has no prerequisite exam. You create a Microsoft Learn account, register with a personal Microsoft account rather than a work account so the record stays with you if you change jobs, and schedule through Pearson VUE. Microsoft rates the exam intermediate and expects SDLC, GitHub workflow and code review experience before you sit.
The outline assumes you've configured agents yourself.
Microsoft's audience profile asks for hands-on experience with GitHub Copilot, MCP servers, custom instructions, custom agents and Copilot setup steps. As of October 2026, Microsoft Learn lists no practice assessment for GH-600, only the exam sandbox, so your practice has to come from real repositories. Connect a remote MCP server, give an agent a CI workflow to run in, and break it on purpose before you book.
GH-600 strengthens a role rather than qualifying you for one on its own. These are the six roles where supervising AI agents inside a GitHub workflow is becoming part of the job.
Builds and runs the agents themselves. GH-600 tests the parts of the job that decide whether an agent can be trusted in production: structured plans, tool permissions, memory rules and evaluation signals. Microsoft lists this role first on the certification page.
Works next to agents that open branches and pull requests in the same repositories. GH-600 covers how to scope those agents, read their plans and artifacts, and tune their instructions when the output comes back wrong.
Owns the CI workflows agents run inside. GH-600 covers invoking an agent from a workflow, branch-based scope, retries and rollbacks, which sits on top of the Actions material the GH-200 GitHub Actions exam tests.
Runs the shared tooling every team builds on, including MCP servers, registries and allow lists. The tool use domain, the heaviest on GH-600 at 20 to 25 percent, reads a lot like this role's backlog.
Decides where agents belong in a delivery process and how much autonomy each one gets. The architecture and guardrails domains cover risk-based autonomy levels and where a human approval actually earns its place.
Microsoft also tags data engineers for GH-600. Pipelines that agents build or modify still need durable state, traceable actions and a rollback path, and those are the habits the exam checks.
Here's how GH-600 lines up against the Copilot exam most candidates take first, the Actions exam for the pipelines agents run in, and Microsoft's own agent developer associate on Azure.
| Agentic AI Developer (GH-600) | GitHub Copilot (GH-300) | GitHub Actions (GH-200) | Azure AI Apps and Agents Developer (AI-103) | |
|---|---|---|---|---|
| Issuer | GitHub, administered by Microsoft | GitHub, administered by Microsoft | GitHub, administered by Microsoft | Microsoft |
| Level | Role-based, intermediate | Role-based | Role-based | Associate |
| Focus | Running and governing AI agents in the SDLC | Copilot features and governance | CI/CD with GitHub Actions | Agents and AI apps on Microsoft Foundry |
| Current Outline | May 2026 (first release) | August 7, 2026 | January 2026 | April 16, 2026 |
| Exam Time | 120 min | 100 min | 100 min | 120 min |
| Passing Score | 700 of 1000 | 700 of 1000 | 700 of 1000 | 700 of 1000 |
| Cost (US) | $165 | $99 | $99 | $165 |
| Validity | 2 years | 2 years | 2 years | 1 year, free renewal |
| Best For | Engineers running AI agents inside a development workflow | Developers and team leads using Copilot | Developers and DevOps engineers who own pipelines | Python developers building agents on Azure |
Microsoft prices exams by the country where you test, so fees vary outside the United States. GitHub certifications don't appear on the DoD 8140 Matrix, so their recognition is market-driven rather than DoD-listed. The two-year GitHub credentials are moving onto Microsoft's recertification process, so check your expiration date in your Microsoft Learn profile.
Certification Matchups: the published facts, strengths and weak spots of 33 certifications, side by side.
As a Microsoft training partner, we run the GitHub AI Certification Track, a three-day boot camp that covers two exams: GH-300 GitHub Copilot and GH-600. Day two works through all six GH-600 domains with labs in a hosted GitHub organization with Copilot and agent features turned on, and day three runs a timed full-length practice exam for each before you sit them. Official Microsoft Courseware, vouchers for both exams, a free retake and 180 days of instructor support after class are included.
The full GitHub track, the agent risks the GH-600 guardrails domain is written for, and where this credential sits among the other AI certifications in 2026.
The structure of every GitHub exam in 2026, which ones were rewritten this year, what the GH-600 Agentic AI Developer exam adds to the track, and the order that makes sense for your role.
The ten risks OWASP ranks highest for AI systems that plan, remember, call tools and act with delegated authority. Read it next to the GH-600 guardrails domain.
A cert-by-cert rundown of the AI credentials that matter this year, who each one is built for, and where it wins or loses against the alternatives.
The four Anthropic Claude exams, what each one costs, and why only people at Claude Partner Network organizations can sit them today.
How Microsoft certification moved from product exams to job roles, and why the 2026 replacements put Copilot, agents and Microsoft Foundry inside the core job.
How to read an official exam outline, including a Microsoft Learn study guide, and turn its domains, weights and verbs into a study plan. Useful for a new exam like GH-600 with little third-party material.
The six specialized agents Microsoft announced for Azure Copilot, what they can do inside your RBAC and compliance rules, and what that means for operations teams.
The GH-600 exam is organized into six domains, each carrying its own weight. Tool use and environment interaction is the heaviest at 20 to 25 percent, while memory and guardrails are the lightest at 10 to 15 percent each. Click any domain for what it covers.
Where an agent belongs in the software development lifecycle: the steps it performs, its inputs, outputs and success criteria, and the anti-patterns to avoid. Also covers keeping planning separate from execution, having the agent output a structured plan, validating that plan before anything runs, and setting how much autonomy the agent gets and where a human steps in.
Choosing agent tools and their permissions, adding MCP servers (including the GitHub remote MCP server), and configuring MCP registries and allow lists. Also covers scoping an agent to a repository or branch, invoking it from a CI workflow, letting it create branches and pull requests, and building error handling, retries, rollbacks, escalation paths and traceability around its actions. The heaviest domain on the exam.
Choosing between short-term, long-term and external memory, scoping memory to the task at hand, and setting expiration, pruning and reset rules. Also covers recording progress and decisions as durable artifacts, resuming work without repeating steps, catching context drift in long runs, and keeping shared state from going stale or conflicting across tools.
Defining expected outcomes and the qualitative and quantitative signals that show whether an agent met them, including signals from automated scanning tools. Also covers tracing failures through logs, plans, traces and workflow artifacts, classifying root causes such as reasoning errors, tool misuse or context problems, and revising instructions, memory and tool access in response.
Applying an orchestration pattern, isolating agents for parallel work, and resolving conflicts such as overlapping code changes, duplicated effort and contradictory outputs. Also covers audit-ready artifacts and handoff records, spotting failed or stalled runs, recovery through rollback and human review, and adding, replacing or retiring agents without breaking a workflow that is already running.
Classifying agent actions by operational, security and compliance risk and assigning autonomy levels to match. Also covers picking out the actions that need human judgment, blocking actions that break security, compliance or Responsible AI policy, enforcing least privilege, requiring explicit authorization for irreversible changes, and cutting approvals that slow delivery without reducing risk.
Domains and weights reflect the Microsoft Learn GH-600 study guide, dated May 2026 and last updated July 9, 2026. Microsoft refreshes GitHub exams periodically, so confirm the current outline on Microsoft Learn before scheduling.
The questions candidates ask most often when researching the GitHub Certified: Agentic AI Developer certification.
GitHub Certified: Agentic AI Developer, earned by passing exam GH-600, shows you can operate, integrate, supervise and govern AI agents inside production software development workflows, with GitHub as the control plane. The exam itself is titled Developing in Agentic AI Systems. Its six domains run from agent architecture and tool use through memory, evaluation, multi-agent coordination and guardrails.
No. GH-600 opened as a beta exam in May 2026, and Microsoft's June 2026 credentials roundup set general availability for July 2026. As of October 2026 the Microsoft Learn certification page no longer marks it as beta, while the newer GH-700 exam still carries the label. Confirm the status on Microsoft Learn when you book.
It's GitHub-branded, and Microsoft administers it. You register with a Microsoft Learn account, schedule through Pearson VUE, and the certification appears in your Microsoft Learn profile under the title GitHub Certified: Agentic AI Developer. Training Camp teaches it as a Microsoft training partner using Official Microsoft Courseware.
There's no prerequisite certification. Microsoft expects experience with the software development lifecycle, GitHub workflows and controls, and code quality, security and review practices. It also expects hands-on time with coding agents including GitHub Copilot, MCP servers, and agent customization such as custom instructions, custom agents, tools and Copilot setup steps.
The six domains are Prepare agent architecture and SDLC processes (15 to 20 percent), Implement tool use and environment interaction (20 to 25 percent), Manage memory, state, and execution (10 to 15 percent), Perform evaluation, error analysis, and tuning (15 to 20 percent), Orchestrate multi-agent coordination (15 to 20 percent), and Implement guardrails and accountability (10 to 15 percent). Tool use carries the most weight.
Microsoft does not publish a question count for GH-600. You have 120 minutes, 20 minutes more than any other GitHub certification exam, and Microsoft says the exam may include interactive components. Pace yourself against the clock rather than a fixed number of items.
You need 700 on a scale of 1 to 1,000. Microsoft uses scaled scoring, so 700 doesn't mean 70 percent correct, and the passing line accounts for the difficulty of the questions you get. If you fail a first attempt, you can retake after 24 hours.
The GH-600 exam costs $165 USD in the United States as of October 2026, scheduled through Pearson VUE. That's the same price Microsoft charges for its Azure role-based exams, and more than the $99 charged for GH-900, GH-200, GH-300, GH-500 and GH-100. Microsoft prices by the country where you test, so check the exam page for your region. Training Camp's GitHub AI Certification Track includes vouchers for GH-300 and GH-600 in the program fee, along with a free retake.
It's the sensible order, though not a requirement. Microsoft's GH-600 audience profile assumes experience with coding agents including GitHub Copilot, and GH-300 is the exam that covers Copilot's features, data handling and governance. If you already use Copilot's agent features every day, you can go straight to GH-600. Training Camp's GitHub AI Certification Track covers both exams in three days.
No. GitHub certifications don't appear on the DoD 8140 Approved Qualifications Matrix, so GH-600 on its own won't satisfy a cyber-coded position requirement. Its value is practical instead, for teams that let AI agents change code and need someone accountable for what those agents do. See the full DoD 8140 work role paths.
Yes. Microsoft's GH-600 page states that GitHub certifications are valid for two years. GitHub is moving its certifications onto Microsoft's recertification process, which will let holders renew without retaking the full exam, and certifications that expire before that process is live are being extended by six months. Check your expiration date in your Microsoft Learn profile.
GH-600 fits AI engineers, software developers, DevOps engineers, platform engineers, solution architects and data engineers who put AI agents into a real development workflow. Microsoft tags the AI engineer, developer, DevOps engineer, solution architect, data engineer and app maker roles for it. It strengthens a role you hold or are moving into rather than qualifying you for one on its own.
For engineers already running agents against real repositories, yes. It's the GitHub credential built around supervising agents rather than prompting them, its outline dates from May 2026, and it gives multi-agent coordination and guardrails their own scored domains. If you haven't used Copilot's agent features yet, start with GH-300 and come back once you have.
If you're weighing the certification, planning the GitHub AI track for yourself, or arranging Copilot and agent training for a team, tell us where you are and we'll help you map out the right path.