GH-700: the new GitHub Certified: Agentic AI Business Professional certification

GH-700 validates that you can use GitHub and GitHub Copilot to design agentic workflows and apply human oversight to their decisions. According to the announcement, the certification shows four abilities: organizing and tracking work with GitHub, producing reliable outputs with GitHub Copilot, building agentic workflows and automations, and reviewing what those workflows produce so you can approve, revise or stop them.

The audience is clearly business-side. The study guide names product managers, program managers, technical project managers and business analysts who want to improve the quality, speed and scale of their daily work without deep software development experience. You are expected to have used generative AI, to manage projects, products or processes, and to make or influence decisions about quality, speed, risk and control. Familiarity with GitHub helps but is not required.

The beta has been open since 20 October 2026. The first 300 candidates who take the exam by 9 November 2026 pay 80% less by entering the code GH700B8F5 when prompted for payment during registration. Seats are first come, first served, and you must take the exam, not just book it, by that date. The discount is not available in Turkey, Pakistan, India or China; in fact, the official beta exams page says candidates in those countries cannot take beta exams at all.

If you use the code, you also receive a 25% voucher for your next exam about two weeks after GH-700 goes live. If you miss a discounted seat, you can still take the beta at full price.

As of 29 October 2026 Microsoft has published neither the certification page nor the exam page on Microsoft Learn (both addresses return an error). That means there is no official price, duration or language list yet. What the study guide does set is the passing score, 700 out of 1,000, and the general rule: localized exams are updated about eight weeks after the English version, and if the exam is not available in your language you can request 30 extra minutes.

The heaviest block is designing and implementing agentic workflows and automations. It asks you to identify use cases that benefit from an agentic workflow, decompose them into processes and tasks, specify decision criteria, create the workflow files (schedule, permissions, output, folder structure, instructions and self-critique loops), configure triggers, prompts, models and tools, and work with skills (installing external ones, creating your own, calling them) and with MCP servers to give the workflow more context.

The GitHub platform domain mixes governance with work management: budget guardrails and token usage, secrets management, privacy and risk controls, Responsible AI principles, branch strategies, choosing between models and when to use MCP servers, alongside GitHub Projects (templates, views, custom fields, access), issues, pull requests and wikis.

The Copilot domain is the most hands-on: picking a model for the task, managing sessions (start, resume, rename, pin and archive), @ mentions, slash commands such as /explain, /fix, /tests, /doc, /clear, /delete and /chronicle, writing prompts that define the input data, content and response format, reverting with checkpoints and creating custom agents with guardrails. The last domain, human-in-the-loop, covers where human judgment is needed, the interactive, plan and autopilot modes, escalation paths and how to approve, reject, revise, troubleshoot, monitor or stop a running workflow.

GH-700 and GH-300 share a foundation: both revolve around GitHub Copilot, model choice, prompt writing, Responsible AI and governing the tool across an organization. The difference is the angle. GH-300 looks at Copilot from the developer's side; GH-700 looks at it from the person who organizes the work and decides what gets automated, with much more weight on Projects, issues, agentic workflows and human oversight.

If you already hold GH-300 or are preparing it, what you know about models, prompts, custom agents and policies carries over almost entirely; what you need to add is agentic workflows (files, triggers, permissions, skills and MCP) and the human-in-the-loop part. CertifyDeck's GH-300 questions cover that common ground on the current official outline, and you can start free.

If you are starting from scratch, a sensible order is: GitHub Projects, issues and pull requests first (the area the exam takes most for granted), then Copilot Chat and custom agents, and finally agentic workflows built around a real task from your job: a weekly report, issue triage or a document review. Microsoft has not yet published a Practice Assessment or a dedicated learning path for GH-700; the announcement points to the study guide, training partners and Microsoft's article on preparing for beta exams without dedicated material.