Module 1 of 11
Introduction to agentic AI business solutions
Align AI solutions with business goals, know the Microsoft AI stack, and recognise which out-of-the-box agent already solves the problem.
Exam questions Q1–Q10 8 units on Learn
Open module 1 on Microsoft Learn ↗ Watch the episode (21 min)
Why this module matters
This module sets the vocabulary for everything that follows. The exam repeatedly asks you to place a requirement on the Microsoft AI stack — is this a Microsoft 365 Copilot job, a Copilot Studio job, a Dynamics 365 feature that just needs enabling, or genuine pro-code work in Microsoft Foundry?
It also establishes the architect's posture. You are not asked to be the person who builds the agent; you are the person who decides whether an agent is the right answer, which platform it belongs on, and what has to be true about data, governance, and process before it can work.
The one big ideaAn agent is worth building when a task is repetitive, judgement-light, well-grounded in trusted data, and has a clear owner and escalation path. If any of those four is missing, fix that first.
What Microsoft Learn covers
Units in this module, as published on Microsoft Learn:
- Introduction
- Drive AI transformation with architect strategies
- Explore Microsoft AI technologies for business
- Identify Microsoft AI technologies for business solutions
- Identify out-of-box Microsoft AI agent resources for business solutions
- Identify out-of-box Microsoft AI agents for business
- Module assessment
- Summary
Watch this module
Microsoft Learn publishes a free video course for AB-100 on YouTube. This module is covered by one 21 min episode. An instructor walking through the module is often faster than reading it twice.
Key concepts and definitions
| Agentic AI | AI that reasons over a goal, plans multiple steps, and invokes tools or takes actions — as opposed to generative AI, which produces content in response to a single prompt. |
| Microsoft 365 Copilot | The AI experience embedded in Word, Excel, Outlook, Teams and the rest of Microsoft 365. Grounded in your tenant's Microsoft Graph data and enforces existing permissions. |
| Microsoft 365 Copilot Chat | The chat surface for Copilot, available more broadly than full Microsoft 365 Copilot. Useful as the delivery surface for declarative agents. |
| Microsoft Copilot Studio | The low-code environment for authoring, extending, and publishing agents — topics, knowledge, tools/actions, triggers, and agent flows. |
| Microsoft Foundry | The pro-code environment for custom models, custom orchestration, and Foundry tools. Reach for it when platform defaults are genuinely insufficient. |
| Microsoft Power Platform | Power Apps, Power Automate, Power Pages, and Dataverse — the application and automation fabric agents plug into. |
| Out-of-the-box (OOB) agent | A prebuilt agent shipped inside Dynamics 365 or Microsoft 365 that you enable and configure rather than build. |
| Grounding | Anchoring model output in trusted organisational data so responses are accurate, current, and attributable rather than invented. |
| Orchestration | Deciding which knowledge, tool, or child agent handles a given user goal, and composing the result. |
| Human in the loop | A defined checkpoint where a person reviews or approves an agent action before it takes effect. |
Design guidance
Placing a requirement on the stack
- Already exists in Dynamics 365 or Microsoft 365? Enable and configure the OOB capability. This is the cheapest, fastest, most governed path and is very often the intended exam answer.
- Exists but needs your data or your terminology? Extend — add knowledge sources, Graph connectors, custom instructions, or a declarative agent.
- Needs bespoke orchestration, a specific model, or non-standard hosting? Only then build custom, in Copilot Studio for low-code or Microsoft Foundry for pro-code.
- Ask *who owns the outcome* at every level. An agent without a named business owner will not survive contact with production.
Autonomy levels
- Assistive — the user prompts, the agent responds, the user acts. Lowest risk, fastest to approve.
- Semi-autonomous — the agent proposes an action and a human approves it. The right default for anything with financial, legal, or customer-facing consequence.
- Autonomous — an event triggers the agent and it acts without a prompt. Requires strong grounding, tight permissions, monitoring, and a rollback story.
- Autonomy is a spectrum you move along as evidence accumulates, not a setting you choose on day one.
Scaling adoption
- Stand up an AI Center of Excellence before proliferation, not after: standards, reusable patterns, review gates, and a shared backlog.
- Measure adoption with active usage and task completion, not tokens consumed or number of agents deployed.
- Treat prompts, instructions, and knowledge curation as maintained assets with owners — they decay just like code.
Good practice vs. anti-pattern
Do
- Start from the business process and its pain, then choose technology.
- Prefer configuring an existing agent over building a new one.
- Define escalation and fallback behaviour before go-live.
- Name a business owner for every agent.
Avoid
- Choose a platform because it is new or interesting.
- Assume a bigger model fixes a grounding or data-quality problem.
- Let each team build its own overlapping agent unsupervised.
- Give an agent autonomy over irreversible decisions on day one.
Exam traps to watch for
Trap"Agentic" does not mean "large model." The distinguishing feature is goal-directed reasoning plus tool use and action-taking, not model size.
TrapWhen a question offers *build a custom agent* alongside *configure the existing capability*, and the existing capability plausibly fits, the configure answer is correct.
TrapBeware answers that measure success in technical units — tokens, models deployed, environments provisioned. Business-outcome metrics win.
TrapA scenario describing an irreversible, legally binding, or unreviewable decision is signalling that full autonomy is the wrong answer.
Readiness checklist
You are ready to move on when you can do all of these without notes.
- Name the four main Microsoft surfaces for AI work and say what each is for.
- Given a requirement, choose between enable, configure, extend, and build — and justify it.
- Explain the difference between generative AI and agentic AI to a non-technical sponsor.
- Describe the three autonomy levels and the risk controls each one needs.
- List two leading indicators of adoption success that a business sponsor would accept.