What AI Agent Runbooks is
An open, community-driven repository that packages proven enterprise delivery knowledge for AI agents built on Microsoft technologies — so delivery teams spend their time on customisation and business value instead of rediscovering foundational patterns on every engagement.
Born from delivery, not from documentation.
This repository came out of real enterprise engagements, where Microsoft Cloud Solution Architects and Technical Specialists kept hitting the same gap: no structured, reusable, validated guidance for building AI agents at scale. Product documentation explains what a feature does. A runbook tells you what to build, in what order, what will go wrong, and what to do about it.
Same shape every time
Every scenario follows an identical folder and file structure, so any engineer can pick one up mid-engagement and deliver it.
Proven in production
Content is written from deliveries that actually reached production on Copilot Studio, Microsoft Foundry and Microsoft 365 Copilot.
Build once, reuse everywhere
Architecture decisions, grounding, governance and cost control are documented as patterns and shared across scenarios.
What is an AI Agent Runbook?
A structured, end-to-end implementation guide for one specific agent use case, purpose-built around a real business scenario and carrying everything needed to understand, design, build and deploy the agent in an enterprise environment — concept to production, repeatable.
Where these agents are built
Guidance spans three Microsoft AI platforms — often combined within a single scenario.
Microsoft Copilot Studio
Build, test and deploy conversational agents and copilots across Microsoft 365 and custom channels, with knowledge sources, topics, actions and connectors.
Microsoft Foundry
Enterprise-grade platform for building, evaluating and deploying custom models and agent solutions using Azure OpenAI and Azure AI services.
Microsoft 365 Copilot
The native AI experience inside Microsoft 365 apps, extended through declarative agents, connectors and Copilot extensibility scenarios.
How the catalogue is organised
Scenarios are grouped by business domain and agent type. Patterns are grouped by engineering theme and apply across domains.
Scenario platforms
Scenarios span HR, IT service desk, legal, finance and ERP, built on the platform combinations below.
Pattern themes
Patterns capture the decisions that repeat on every engagement — documented once, reused everywhere.
Four folders, one purpose each
The repository is deliberately shallow — find the folder that matches your question and start there.
00-overview/Repository objectives, agent taxonomy, structure and navigation. Start here if you are new.01-scenarios/End-to-end runbooks per use case. Start here if you are building a specific agent.02-patterns/Technical patterns, architecture decisions and best practices. Start here if you need building blocks.03-references/Templates, limitations, links and architecture references. Start here if you need reference material.A delivery path, not a reading list
Pick a use case and start building.
The catalogue is the fastest way in. If you already know the architecture question you need answered, go straight to the patterns instead.
Read this before you use anything here
A community-driven project, not an officially supported Microsoft product, provided as-is. Four things to know before you build on this content.
Read the full disclaimer
Please read the following disclaimers carefully before using any content in this repository.
1. Content accuracy and currency
The content, configurations, architecture diagrams, and implementation guidance included in this repository reflect information and product capabilities available at the time of authoring. Microsoft products and services — including Microsoft Copilot Studio, Azure AI Foundry, and Microsoft 365 Copilot — are continuously updated, and features, interfaces, APIs, and behaviors may change without notice following a product update or release.
Users are responsible for verifying that the guidance in this repository remains applicable and accurate against the current version of the relevant Microsoft products and services before use in any project or production environment.
2. No Microsoft support
This repository is a community-driven project and is not an officially supported Microsoft product. Content is contributed and maintained by Microsoft employees and community members acting in a voluntary capacity.
Microsoft Support (CSS) does not provide assistance, troubleshooting, or incident response for issues arising from the use of content in this repository. For official product support, please use the Microsoft Support portal or your designated Microsoft support channel.
3. Deployment responsibility
Any deployment of AI Agent solutions derived from or informed by the content in this repository is the sole responsibility of the customer or user performing the deployment. This includes, but is not limited to:
- Ensuring compliance with applicable laws, regulations, and organizational policies
- Validating configurations, integrations, and data handling practices in your target environment
- Managing access controls, security configurations, and data privacy requirements
- Testing and validating agent behavior prior to production release
Microsoft and the contributors to this repository assume no liability for any damages, data loss, service disruption, or compliance violations resulting from the use or deployment of content from this repository.
4. General use at own risk
This repository is provided “as-is”, without warranty of any kind, express or implied. The authors and contributors make no representations or guarantees regarding the completeness, reliability, suitability, or availability of any content within this repository for any particular purpose. Use of this repository and its contents is entirely at your own risk.
Source of record: 00-overview · Disclaimer. Trademark guidance is covered by Microsoft’s Trademark & Brand Guidelines.