AB-100 · Agentic AI Business Solutions Architect
Level: Expert · Passed: March 2026 · Length: 100 minutes · Pass mark: 700 Official: AB-100 study guide
AB-100 is an architect exam covering Copilot Studio, Azure AI Foundry, Dynamics 365 and Power Platform governance. The questions are long business scenarios: a company's constraints, and the best combination of services, controls or steps. Deployment and governance carry the most weight (40–45%). As the prerequisite, you need one of several associate certifications, such as AI-102.
Pages in this section
- Governance controls: DLP, Managed Environments, Entra ID, approval gates and RBAC, plus which control fits which incident.
- Platform selection: Copilot Studio vs. AI Foundry vs. custom Azure, and A2A vs. MCP.
- ALM and components: promotion paths by platform, Copilot Studio components, pipelines.
- Dynamics 365 integration: AI capabilities by app, Dataverse, security modes.
- Open-book experiment: which questions a Microsoft Learn lookup can actually answer.
Five ordering principles
Sequence questions are common and can't be looked up. Rather than memorizing specific sequences, learn the principles that produce them:
| # | Principle | Why |
|---|---|---|
| 1 | Governance first | Nothing moves without guardrails: DLP, RBAC and approval gates come before data or users |
| 2 | Evaluation last | Measuring real outcomes needs production traffic |
| 3 | Contain before you investigate | In an incident, stop the damage first, then find the cause |
| 4 | Automated before human | Scalable automated checks come before expensive human review |
| 5 | Monitor after deploy | Telemetry needs live traffic; set it up during deployment, read it after |
A typical deployment sequence follows from these: governance → grounding (knowledge + access controls) → test → promote → monitor → evaluate.
Elimination rules
| Rule | Kills which trap |
|---|---|
| Temperature or confidence is never a security fix | Lowering temperature for an auth or injection problem |
| Rollback fixes deployment errors, not knowledge gaps | Rolling back for a content or glossary problem |
| Permission problems get permission fixes; content problems get content fixes | RBAC for prompt injection; content filters for over-permissioned agents |
| Governance goes first; monitoring goes after promotion | Telemetry too early, governance too late |
| Tenant settings ≠ environment DLP ≠ row-level security | Tenant settings for user-level data filtering |
Rule of thumb: Microsoft prefers the highest-level managed platform that meets the constraints. Only drop down to custom code when the scenario explicitly requires it.