Study guide for Exam GH-600: Developing in Agentic AI Systems

Purpose of this document

This study guide should help you understand what to expect on the exam and includes a summary of the topics the exam might cover and links to additional resources. The information and materials in this document should help you focus your studies as you prepare for the exam.

Useful links Description
Certification renewal Microsoft associate, expert, and specialty certifications expire annually. You can renew by passing a free online assessment on Microsoft Learn.
Your Microsoft Learn profile Connecting your certification profile to Microsoft Learn allows you to schedule and renew exams and share and print certificates.
Exam scoring and score reports A score of 700 or greater is required to pass.
Exam sandbox You can explore the exam environment by visiting our exam sandbox.
Request accommodations If you use assistive devices, require extra time, or need modification to any part of the exam experience, you can request an accommodation.

About the exam

Some exams are localized into other languages, and those are updated approximately eight weeks after the English version is updated. While Microsoft makes every effort to update localized versions as noted, there may be times when the localized versions of an exam are not updated on this schedule. Other available languages are listed in the Schedule Exam section of the Exam Details webpage. If the exam isn't available in your preferred language, you can request an additional 30 minutes to complete the exam.

Note

The bullets that follow each of the skills measured are intended to illustrate how we are assessing that skill. Related topics may be covered in the exam.

Note

Most questions cover features that are general availability (GA). The exam may contain questions on Preview features if those features are commonly used.

Skills measured

Audience profile

You should have subject matter expertise in operating, integrating, supervising, and governing AI agents inside production-grade SDLC workflows and development environments, ensuring reliability, safety, and velocity using GitHub as the system of record and control plane.

Your responsibilities for this role include:

  • Operating agent workflows inside the SDLC

  • Supervising autonomous behavior with GitHub controls

  • Evaluating and tuning agent outputs using scans and artifacts

  • Configuring custom agents

  • Coordinating multi-agent execution safely

You work closely with architects, platform engineers, DevOps engineers, application developers, product managers, and security engineers to develop, deploy, operate, and manage agents that operate within the GitHub platform.

You should have experience with the software development lifecycle (SDLC), workflows in GitHub and controls, and code quality, security, and review practices. You should also have experience with coding agents including GitHub Copilot, MCP servers and agent customization such as custom instructions, custom agents, tools, and Copilot setup steps.

Skills at a glance

  • Prepare agent architecture and SDLC processes (15–20%)

  • Implement tool use and environment interaction (20–25%)

  • Manage memory, state, and execution (10–15%)

  • Perform evaluation, error analysis, and tuning (15–20%)

  • Orchestrate multi-agent coordination (15–20%)

  • Implement guardrails and accountability (10–15%)

Prepare agent architecture and SDLC processes (15–20%)

Integrate agents into the software development lifecycle (SDLC)

  • Identify steps for agents to perform

  • Identify and mitigate common anti-patterns in agents

  • Define inputs, outputs, and success criteria for agents

Define boundaries between planning, reasoning, and action

  • Configure agent planning to be distinct from agent execution

  • Configure an agent to output a structured plan

  • Validate agent plans

  • Prevent agent action until the agent checked and approved

Configure observability and control for autonomous agents

  • Plan and implement the degree of agent autonomy, including guardrails

  • Configure agent to produce inspectable artifacts within standard development tooling

  • Configure human intervention for autonomous agents without slowing delivery

Implement tool use and environment interaction (20–25%)

Select and configure agent tools

  • Identify required tools

  • Configure agent tools

  • Configure agent tool permissions

Configure MCP servers

  • Add an MCP server as a tool to an agent

  • Configure a GitHub remote MCP server

  • Configure the MCP registries

  • Configure MCP allow lists

Integrate agents within development environments

  • Evaluate the execution context for an agent

  • Configure an agent's scope to a specific repository

  • Configure an agent to be invoked in a CI workflow

  • Configure an agent to use branch-based scope

  • Enable an agent to perform autonomous actions, including creating branches and pull requests

  • Configure an agent to handle environment-specific constraints

Operate agents with safe execution paths and robust error handling

  • Implement error handling

  • Implement retries

  • Implement rollbacks

  • Implement escalation paths

  • Implement traceability and accountability for agent actions

Manage memory, state, and execution (10–15%)

Implement agent memory strategies

  • Choose between short-term, long-term, and external memory

  • Scope agent memory to task-relevant information

  • Define memory expiration, pruning, and reset rules

Persist agent state and manage context drift

  • Capture task progress and decisions as durable artifacts

  • Resume agent work without repeating steps or diverging from prior decisions

  • Detect and correct drift during extended agent execution

Ensure continuity of agent memory and state across tools and environments

  • Share agent state

  • Prevent conflicting context

  • Prevent stale context

Perform evaluation, error analysis, and tuning (15–20%)

Define success criteria and evaluation signals for agent tasks

  • Specify expected outcomes and operational constraints for agent tasks

  • Identify qualitative and quantitative evaluation signals to evaluate agents

  • Align evaluation criteria with development intent

  • Generate evaluation signals by using automated scanning tools

Analyze agent failures and identify root causes

  • Identify failures by using logs, plans, traces, outputs, and workflow artifacts

  • Classify root causes, including reasoning errors, tool misuse, and context or environment issues

Tune agent behavior based on evaluation results

  • Revise instructions, workflows, or constraints

  • Refine memory usage

  • Refine tool usage and tool access

Orchestrate multi-agent coordination (15–20%)

Operate and manage multi-agent workflows

  • Apply an orchestration pattern to coordinate multiple agents

  • Configure agent isolation for parallel execution

  • Detect and resolve agent conflicts, including overlapping code changes, duplicated effort, and contradictory outputs

Configure observability for multi-agent behavior by using logs, artifacts, and operational signals

  • Configure multi-agent workflows to produce artifacts suitable for review and audit

  • Document key decisions, handoffs, and outcomes across agents

  • Perform post-hoc analysis of multi-agent behavior

Detect and respond to multi-agent failures and degraded behavior

  • Identify failed, partial, or stalled agent executions

  • Respond to degraded behavior or coordination across agents

  • Implement multi-agent recovery patterns, including rollback and human-in-the-loop

Manage the lifecycle of agents within multi-agent workflows

  • Add agents to existing multi-agent workflows

  • Update, reconfigure, or replace agents without disrupting active workflows

  • Retire agents while preserving auditability and workflow continuity

Implement guardrails and accountability (10–15%)

Define autonomy levels

  • Classify agent actions by operational, security, and compliance risk to right-size human interventions

  • Assign autonomy levels to maximize delivery speed while remaining compliant with organizational security and Responsible AI standards

Implement guardrails and human-in-the-loop workflows

  • Identify the subset of actions that require human judgment

  • Block actions that violate defined security, compliance, or Responsible AI policies

  • Scope permissions and execution contexts to enforce least-privilege access

  • Require explicit authorization or controlled paths for irreversible or compliance-sensitive changes

  • Preserve execution velocity by minimizing approvals that do not materially reduce risk

Study resources

We recommend that you train and get hands-on experience before you take the exam. We offer self-study options and classroom training as well as links to documentation, community sites, and videos.

Study resources Links to learning and documentation
Get trained Choose from self-paced learning paths and modules or take an instructor-led course on Microsoft Learn – Foundations of Agentic AI in GitHub, Designing Agent Architecture and SDLC Integration, Tooling, MCP, and Agent Execution Environments
Find documentation Prepare agent architecture and SDLC processes
Implement Tool Use and Environment Interaction
Manage Memory, State, and Execution
Perform Evaluation, Error Analysis, and Tuning
Orchestrate Multi-Agent Coordination
Implement Guardrails and Accountability
Ask a question GitHub Community Discussions
Get community support GitHub Blog
Follow GitHub Twitter
LinkedIn
Instagram
Find a video YouTube