Domain 2 · Design AI-powered business solutions
Where Domain 1 decided whether and why, this domain decides how the solution is shaped. It is the broadest domain in terms of products touched — Copilot Studio, Dynamics 365, Power Platform, Microsoft 365, and Foundry all appear — and the connecting thread is choosing the right building block and the right extensibility point for a stated requirement.
Design AI and agents
Much of the design surface is Copilot in Dynamics 365 across the customer-experience, service, and sales apps, including connectors for Copilot in D365 Sales and agents for the D365 Contact Center channels.
Agents themselves come in kinds that the exam expects you to tell apart: task agents that carry out a defined job, autonomous agents that act on their own initiative, and prompt-and-response agents that answer within a turn. Matching the kind to the requirement is a recurring question shape, as is proposing the appropriate Foundry Tools for a requirement. Alongside agents you may be asked to propose code-first generative pages or an agent feed for an app.
Inside Copilot Studio, the constructs are topics — including the fallback topic that catches whatever no other topic matched — agent flows, and prompt actions, together with the data processing that grounds them. The most commonly tested choice in Copilot Studio is between standard NLP, conversational language understanding, and generative orchestration: increasing flexibility against decreasing determinism, so the right answer depends on how tightly the conversation must be controlled.
Two frameworks bound the design work. The Power Platform Well-Architected Framework applies to intelligent workloads much as its Azure counterpart does elsewhere, and AI can be designed into Power Apps canvas apps directly.
Design extensibility
Extensibility is where custom capability meets the packaged products. Solutions can use custom models in Microsoft Foundry and agents in Microsoft 365 Copilot, and Copilot Studio extends through the Model Context Protocol (MCP) and Computer Use, the latter automating applications and websites that expose no API at all.
The open standards in the audience profile are worth knowing by name and by purpose, because they are easy to confuse: MCP connects an agent to tools and knowledge, while Agent2Agent (A2A) provides interoperability between agents. Tools versus peers is the distinction.
Design also covers agent behaviors, including reasoning and voice mode, and optimizing with agents across Microsoft 365 surfaces such as Teams and SharePoint — the places users already work, which is often what determines adoption.
Orchestrate prebuilt agents and apps
The final section returns to the build-versus-extend instinct from Domain 1. AI features are orchestrated within Dynamics 365 finance and supply-chain and customer-experience apps, Microsoft 365 Copilot for Sales and for Service are configured rather than rebuilt, and Power Platform AI features and the AI hub are proposed where they meet the need.
The pattern to carry into the exam: when a prebuilt or configurable capability satisfies the requirement, proposing a custom build is usually the wrong answer.
Quiz · Domain 2
An architect is designing a Copilot Studio agent that must answer multi-intent user queries by combining custom topics, a Power Automate tool, and an indexed knowledge source in a single turn, automatically prompting for any missing inputs. Which orchestration mode should the agent use?