| Plan AI-powered business solutions | 25–30% | Requirements, AI strategy, ROI |
| Design AI-powered business solutions | 25–30% | Agents, extensibility, orchestration |
| Deploy AI-powered business solutions | 40–45% | Monitoring, testing, ALM, governance |
Plan — 25–30%
The Plan domain is the consulting half of the architect's job. Its three clusters: analyzing requirements (where agents fit task automation, analytics and decision-making, and whether the grounding data is accurate, relevant, timely, clean and available), designing the overall AI strategy (the Cloud Adoption Framework's AI adoption process, multi-agent designs spanning Microsoft 365 Copilot, Copilot Studio and Microsoft Foundry, choosing between extending Microsoft 365 Copilot and building custom agents, prompt libraries and prompt-engineering guidance, small-language-model use cases, an AI Center of Excellence), and costs and benefits — ROI criteria including total cost of ownership, build-versus-buy-versus-extend calls, and designing a model router that sends each request to the most suitable model.
The tell for exam questions here: two defensible options where the scenario's business constraint — cost ceiling, data quality, an existing Dynamics 365 footprint — decides between them.
Design — 25–30%
The hands-on architecture domain, in three clusters. Agents for business solutions: Copilot customization in Dynamics 365 customer-experience and service apps, connectors for Dynamics 365 Sales, Contact Center channel integration, and the taxonomy the exam expects you to keep straight — task agents versus autonomous agents versus prompt-and-response agents — plus Copilot Studio topic design with fallback, grounding-data processing, AI in Power Apps canvas apps, and choosing among standard NLP, conversational language understanding, and generative orchestration. Extensibility: custom models in Microsoft Foundry, agents in Microsoft 365 Copilot, and Copilot Studio extensibility including Model Context Protocol and Computer Use agents that automate apps and websites. Orchestrating prebuilt agents: AI features across Dynamics 365 finance, supply chain, customer experience and service, Copilot for Sales and Copilot for Service configuration, and Power Platform's AI hub.
Deploy — 40–45%, the domain that decides your result
Nearly half the exam lives after the design document is signed. Four clusters:
- Monitor and tune — the monitoring process and tools for agents, analyzing usage backlogs and user feedback, interpreting telemetry for performance and model tuning.
- Testing — test processes and metrics for agents, validation criteria for custom models, prompt best-practice validation, end-to-end scenarios across multiple Dynamics 365 apps, and using Copilot itself to build test cases.
- ALM — lifecycle processes for the data behind models and agents, for Copilot Studio agents, connectors and actions, for the Microsoft Foundry Agents service, and for AI features in Dynamics 365 apps.
- Responsible AI, security, governance and compliance — agent and model security, vulnerability analysis including prompt manipulation, data residency, access controls on grounding data and model tuning, and audit trails for changes to models and data.
If you allocate study time by the published weights — as our practice forms do — Deploy gets almost half. Most candidates give it a quarter, because design is the part that feels like architecture. The weighting says Microsoft disagrees: an architect who cannot govern, test and operate an agent fleet is not certified to build one.
Source: Microsoft's official AB-100 study guide (skills measured as of July 22, 2026). Wording above condenses the outline; the live guide is authoritative and can change without notice.