The GitHub Certified: Agentic AI Business Professional (exam GH-700) is one of the newest certifications released by Microsoft. It's so new, in fact, that the exam is in beta until December 2026. I recently sat the exam, so here's an insider's perspective on what to expect as you prepare.
Who the exam is for
Microsoft says the exam is aimed at product managers, program managers, technical project managers and business analysts, and specifically not at a heavily technical or developer audience. In my estimation, it's aimed at the GitHub and GitHub Copilot power user. If your work in GitHub is mostly moving tickets around on project boards, expect this exam to go well beyond that. The certification focuses on using agentic AI tools in GitHub to create workflows and increase productivity.
I was curious how others who had taken the exam felt about it, so I checked the AzureCertification subreddit. The general sentiment was that the exam is more technical than people expected. Some questioned why it's aimed at business professionals, and a few wondered whether it would make it out of beta at all.
Exam details
| Cost | $165 |
|---|---|
| Passing score | 700 |
| Time | 100 minutes |
| Questions | Around 61 |
| Score turnaround during beta | Up to six weeks |
Preparing for a beta exam
The challenge with Microsoft's beta exams is that there are no Learn modules to help fill gaps in your knowledge. Microsoft offers deep voucher discounts on beta exams for this reason, and largely expects candidates to already have a firm grip on the concepts. This article takes the same approach: it assumes you've worked with these tools, and focuses on the specific concepts I saw on my version of the exam.
The questions are mostly scenario-based, and there are two case studies. Memorizing terms won't help. You need to know how to apply them.
What to study
- GitHub slash commands such as
/chronicle,/clear,/explain,/fixand/doc. I saw these in two questions. - How pressing Esc, double Esc and Ctrl+C affect a running process.
- How the folder structure of an agentic workflow is set up in GitHub.
- What goes in key markdown files like
agent.md, and whereskill.mdfiles are stored. - Where to place an agent to make it available to the whole repository.
- When to use a fast, cheap model versus a reasoning model.
- What built-in models like Ask and Plan do.
- The difference between orchestrator and worker agents.
- CLI commands like
gh aw logs,gh aw audit <run-id>,gh aw status,gh aw runandgh run cancel. - How to monitor agentic workflows while they run, and where to find the reasoning and tools that were used.
- How to redirect an agent mid-session when it isn't doing what you want, and what stopping it means for work already done.
- How to read workflow template syntax, including where triggers, permissions, tools, engine, safe-outputs and approval gates fit.
- What should trigger a human review versus letting the agent act.
- When to use Autopilot, Interactive and Plan.
- When to use built-in templates like Kanban and Feature Release.
- Setting up issues in a project: creating, assigning, linking, tagging and creating dependencies.
What wasn't on the exam
A few topics I expected but didn't see:
- Nothing about wikis
- No content on self-critiquing loops
- Only one question about MCPs
Best of luck with your Agentic AI Business Professional exam.
Watch the video
Video coming soon.