Domain 1 · Plan AI-powered business solutions
AB-100 is an architect's exam, and this domain sets the tone for it. The questions are not about how to call an API; they are about whether a capability should be built at all, what it will cost, and how it fits an organisation's wider AI strategy. Answers that are technically correct but ignore reuse, governance, or total cost tend to be the distractors here.
Analyze requirements
The first judgement is where agents genuinely fit — task automation, data analytics, and decision-making are the named categories, and part of the skill is recognising a process that does not need an agent at all.
Next comes the grounding data, assessed against accuracy, relevance, timeliness, cleanliness, and availability. Each of those can independently sink a solution: data that is accurate but stale misleads just as effectively as data that was wrong to begin with, and data that cannot be reached at request time is not grounding data no matter how good it is.
Underpinning both is organising business-solution data so it is available for reuse by other AI systems. This is the architectural instinct the exam is testing — the second and third agent should not each need their own bespoke data pipeline.
Design the overall AI strategy
Strategy work is anchored in the AI adoption process from the Cloud Adoption Framework for Azure, which gives the exam a shared vocabulary for how an organisation moves from experiment to production.
Solutions are then designed as multi-agent across the Microsoft estate — Microsoft 365 Copilot, Copilot Studio, and Microsoft Foundry — with prebuilt agents considered before anything custom is proposed. That leads directly to the recurring decision of whether to build custom or extend Microsoft 365 Copilot, and the narrower question of when a custom AI model or a customized small language model is justified. The default posture is extend before build; a custom model needs a reason that a prompt and grounding data cannot satisfy.
Two organisational elements complete the strategy. Prompt-library and prompt-engineering guidelines make quality repeatable across teams rather than dependent on individuals, and the AI Center of Excellence is the structure that owns those standards. The domain also expects solutions spanning multiple Dynamics 365 apps, which is where a single-app answer usually falls short.
Evaluate costs and benefits
Financial reasoning is examinable here in its own right. ROI criteria must include total cost of ownership — not just inference spend, but the data preparation, integration, monitoring, and human review the solution requires — and you should be able to construct an ROI analysis for a specific business process.
That analysis feeds the build, buy, or extend decision for each AI component, evaluated per component rather than for the solution as a whole.
One concrete technique sits in this section rather than in a technical domain: implementing a model router that directs each request to the most suitable model. It belongs here because it is fundamentally a cost-and-benefit mechanism — routing routine requests to a cheaper model and reserving the expensive one for requests that need it.
Quiz · Domain 1
An enterprise architect is structuring an AI adoption program using the Cloud Adoption Framework. Which phase of the CAF AI adoption process is responsible for identifying AI use cases, defining the AI technology strategy, developing a data strategy, and developing a responsible AI strategy?