In 2026 Microsoft introduced a family of certifications focused on enterprise artificial intelligence: the AB series. It covers several roles, from the Microsoft 365 administrator and business user to the executive leading AI adoption and the architect designing multi-agent solutions. The most important current change affects the expert-level AB-100 Agentic AI Business Solutions Architect exam: Microsoft updated its study guide on October 21, 2026, for a new outline that takes effect for the English exam on November 16, 2026.
This is more than a wording refresh. The three main domains remain, but their weights are substantially rebalanced. Planning becomes the largest area, design increases, and deployment decreases. AB-100 candidates should therefore build their study plan around both the language and date of their scheduled exam.
The four AB-series certifications
Each credential maps to a different responsibility. Choose according to the work you perform, not simply because every option contains “AI” in its title.
| Certification | Exam | Level | Primary audience |
|---|---|---|---|
| Microsoft 365 Certified: Copilot and Agent Administration Fundamentals | AB-900 | Fundamentals | Administrators who configure, secure, and monitor Copilot and agents in Microsoft 365 |
| Microsoft Certified: AI Business Professional | AB-730 | Business | Users who improve business processes with generative AI and Copilot without architecting a complete solution |
| Microsoft Certified: AI Transformation Leader | AB-731 | Leadership | Decision makers who identify value, organize adoption, and manage AI risk |
| Microsoft Certified: Agentic AI Business Solutions Architect Expert | AB-100 | Expert | Architects who plan, design, and govern enterprise AI and multi-agent solutions |
AB-900, AB-730, and AB-731 provide different entry points. AB-100 has a broader architecture scope and an additional requirement: passing AB-100 alone does not grant the Expert certification. Microsoft also requires one eligible Associate-level certification from the list on the official exam page.
What changes for AB-100 on November 16, 2026
The official Microsoft Learn AB-100 study guide (opens in a new tab) includes a change log comparing the previous and updated outlines. The three domain names remain broadly consistent, but the weighting changes significantly.
| Domain | Before November 16, 2026 | From November 16, 2026, English version |
|---|---|---|
| Plan AI-powered business solutions | 25–30% | 35–40% |
| Design AI-powered business solutions | 25–30% | 30–35% |
| Deploy AI-powered business solutions | 40–45% | 20–25% |
A preparation plan centered on the old deployment weighting will be unbalanced after the update. Monitoring, testing, ALM, security, and compliance remain in scope, but requirements analysis, platform selection, AI strategy, and economic justification now deserve more time.
Microsoft says localized versions are typically updated about eight weeks after the English version. That timing is not an unconditional guarantee. Before booking a non-English session, verify the selected language, date, and displayed outline on the official AB-100 exam page (opens in a new tab). Do not assume that November 16 automatically applies to every localized exam.
Domain 1: planning becomes the priority
“Plan AI-powered business solutions” now represents 35–40% of the exam. It begins with requirements analysis: task automation, decision support, data analytics, and the quality of grounding sources. An architect should evaluate accuracy, relevance, timeliness, cleanliness, and availability before proposing an agent.
Strategy then becomes central. The guide expects candidates to choose among Microsoft 365 Copilot, Copilot Studio, and Microsoft Foundry, as well as decide whether to build, buy, or extend a component. It explicitly covers multi-agent design, prebuilt agents, custom models, customized small language models, prompt libraries, and the Microsoft AI Center of Excellence.
The economic dimension is measured too. Candidates must select return-on-investment criteria, account for total cost of ownership, and defend a build, buy, or extend decision. The guide also mentions implementing a model router that sends requests to the most suitable model. A strong architecture therefore does not automatically select the largest model; cost, latency, security, residency, and expected quality should drive the choice.
For broader context on professional AI certification paths, OnLevelUp's AWS AI certification roadmap offers a useful vendor comparison. Keep the exam boundaries clear: AB-100 is built around Microsoft's business applications, agent platforms, and architecture decisions.
Domain 2: design agents that integrate with real systems
Design now accounts for 30–35%. Microsoft expects candidates to select the right agent type, knowledge source, orchestration model, and extension mechanism. The outline names task agents, autonomous agents, prompt-and-response agents, agent flows, prompt actions, and integrations with Dynamics 365 applications.
Several objectives deserve a dedicated lab: proposing Microsoft Foundry Tools for a requirement, designing grounding and data processing, applying the Power Platform Well-Architected Framework, choosing between traditional natural-language processing and generative orchestration in Copilot Studio, and extending an agent through Model Context Protocol.
The guide also includes Computer Use in Copilot Studio, agent behaviors such as reasoning and voice mode, and designs involving Microsoft 365, Teams, and SharePoint. Do not memorize these as a feature list. For each capability, practice answering three architectural questions: What requirement justifies it? What data and actions can it reach? Which control prevents unintended behavior?
The objective covering Model Context Protocol should be studied as an integration choice rather than general AI trivia. Compare MCP with a native connector or custom API, then document authentication, permission boundaries, failure behavior, observability, and human approval for consequential actions.
Domain 3: a smaller weight does not make operations optional
Deployment falls to 20–25%, but security, governance, and operations remain essential. The domain still covers telemetry analysis, agent monitoring, testing metrics, custom-model validation, end-to-end scenarios, and ALM for data, agents, connectors, actions, and models.
Security should be treated as an architectural property. The study guide asks candidates to design agent security and governance, protect models, analyze vulnerabilities including prompt manipulation, validate data residency and movement, control access to grounding and tuning data, and maintain audit trails for changes.
A practical exercise is to map an agent's complete lifecycle: input data, grounding, model, tools, permissions, output, telemetry, and approval process. Add failure scenarios such as an outdated source, unavailable tool, hostile prompt, or data crossing a prohibited boundary. Your design should explain how the issue is detected, contained, and corrected.
The adjacent OnLevelUp article on advanced AWS AI compute training can strengthen cost and infrastructure reasoning. It is not AB-100 preparation, but it helps candidates separate model and workload choices from the Microsoft-specific business-solution layer.
Exam duration, score, languages, and prerequisites
The Microsoft Learn exam page (opens in a new tab) lists a 100-minute duration and a passing score of 700. AB-100 is offered in multiple languages, including English, French, Arabic, German, Spanish, Italian, Japanese, Korean, and Brazilian Portuguese. Pricing depends on the country or region where the exam is proctored, so use the amount displayed during the official booking flow instead of copying an old price from a blog post.
To earn the Expert certification, candidates must pass AB-100 and hold at least one eligible Associate certification. The current list includes several Power Platform and Dynamics 365 credentials plus AI-oriented options such as AI-300, AI-103, AB-620, AB-210, AB-410, and AB-250. The list can change; check it before committing to a prerequisite path.
A six-week preparation plan
Week 1: lock the correct outline
Choose your target language and date. Save the corresponding study guide and turn every objective into a tracking row. If your English session is on or after November 16, use only the new weights.
Weeks 2 and 3: planning and tradeoffs
Study requirements analysis, grounding-data quality, platform selection, prebuilt versus custom models, and ROI. Build two case studies and justify whether each component should be built, bought, or extended. Include cost, latency, security, data location, and operational ownership.
Week 4: design and extensibility
Create a Copilot Studio prototype with a knowledge source, controlled action, and fallback strategy. Compare a native extension, MCP, and a custom service. Document permissions, failure handling, and testing. Add one multi-agent scenario and define which agent owns each decision.
Week 5: operations, security, and ALM
Define performance, quality, and cost metrics. Prepare a test plan, environment-promotion strategy, and governance model. Add controls for prompt manipulation and noncompliant data movement. Explain how telemetry leads to a tuning or rollback decision.
Week 6: simulate and close gaps
Use the practice assessment available through AI Skills Navigator and explore the Microsoft exam sandbox. Review each error by domain. If planning remains weak, do not use the old weighting to downplay the gap: planning is now the largest portion of the updated exam.
Which AB certification should you choose?
- Choose AB-900 if you administer Microsoft 365 Copilot and agents.
- Choose AB-730 if you improve business workflows with Copilot without owning architecture.
- Choose AB-731 if you lead AI strategy, value realization, and adoption.
- Choose AB-100 if you design multi-service architectures and must balance data, agents, models, security, cost, and operations.
The November 16 update makes AB-100 more strategy-heavy. Planning and design together account for most of the exam. Deployment remains critical, but operational decisions must now sit inside an architecture justified by business requirements, data quality, risk, and measurable value.
Official sources
- Microsoft Learn—AB-100 exam page (opens in a new tab)
- Microsoft Learn—AB-100 study guide, updated October 21, 2026 (opens in a new tab)
- Microsoft Learn—Agentic AI Business Solutions Architect Expert certification (opens in a new tab)
- Microsoft Tech Community—initial AB-100 announcement (opens in a new tab)
- Microsoft Tech Community—AB-900 (opens in a new tab)
- Microsoft Tech Community—AB-730 (opens in a new tab)
- Microsoft Tech Community—AB-731 (opens in a new tab)



