## Microsoft restructured its AI certification track for the agent era Between 2024 and early 2026 Microsoft quietly rebuilt its AI certification ladder. The old path topped out at the associate level with **AI-102 (Azure AI Engineer Associate)** and stopped there. Architects who wanted an expert-level credential had to settle for adjacent exams (Azure Solutions Architect, Power Platform Solution Architect) that did not test agentic AI design. In January 2026 Microsoft closed the gap by launching **AB-100 (Microsoft Certified: Agentic AI Business Solutions Architect)** -- the first expert-level Microsoft credential built specifically for the agentic AI era. AB-100 sits on top of an associate prerequisite and validates exactly the kind of multi-agent, governance-heavy, ROI-aware architecture work that has become standard since GPT-4 and Microsoft Foundry made enterprise agents real. This post is the practical roadmap. It covers the three certs most people will actually sit -- **AI-900, AI-102, AB-100** -- with the timeline, salary impact, and order-mistakes you will see if you ask three architects on LinkedIn. ## Why Microsoft restructured the track in 2025-2026 Three forces drove the rework. First, **the product surface exploded**: in 2023 the answer to "build AI on Microsoft" was largely *call Azure OpenAI Service*; by 2026 it spans Microsoft Foundry, Copilot Studio, the Agent Framework, Microsoft 365 Copilot, Dynamics 365 AI, Power Platform AI Builder, MCP, A2A, and the responsible-AI tooling around them. Second, **architecture became its own discipline** -- multi-agent orchestration, ALM for agents, golden-dataset evaluation gates, prompt-injection defence, and Purview-backed governance are architecture problems, not developer problems. Third, **customers started asking for it**: enterprises rolling out Copilot Studio and Foundry at scale needed an expert-level architect credential, and without one, hiring teams defaulted to less-relevant proxies. AB-100 is Microsoft's answer. ## The three certs that matter: at a glance | Cert | Code | Level | Audience | Focus | Length | Cost | Prereq | |------|------|-------|----------|-------|--------|------|--------| | **Azure AI Fundamentals** | AI-900 | Fundamentals | Anyone | AI concepts, Azure AI services, generative AI basics | 60 min, 40-60 Qs | USD 99 | None | | **Azure AI Engineer Associate** | AI-102 | Associate | Developers, engineers | Building AI solutions with Azure AI services, Foundry, Document Intelligence, Search | 120 min, 40-60 Qs | USD 165 | Strongly recommended: AI-900 + hands-on | | **Agentic AI Business Solutions Architect** | AB-100 | Expert | Solution architects, AI leads | Multi-agent design, ALM, governance, ROI across Foundry, Copilot Studio, Dynamics 365, Power Platform | 100 min, ~60 Qs | USD 165 | Required: one approved associate (e.g. AI-102) | A few notes before we go deeper: - AI-900 has **no prerequisites** and is the cleanest on-ramp for anyone who has not yet sat a Microsoft exam. - AI-102 has no formal prerequisites but expects six to twelve months of hands-on Azure AI work. - AB-100 has a **hard prerequisite gate**: at least one approved associate-level credential from the AI, Dynamics 365, or Power Platform families must be active on your transcript before the AB-100 credential is issued. Official AB-100 prerequisite list: https://learn.microsoft.com/en-us/credentials/certifications/agentic-ai-business-solutions-architect/ ## AI-900: the fundamentals entry point **Azure AI Fundamentals (AI-900)** is the on-ramp. It tests conceptual understanding rather than implementation, and it is the right starting point if you have not sat a Microsoft exam, are coming from a non-AI background, or use AI without writing code. The blueprint covers AI workloads and responsible AI principles, ML fundamentals (classification, regression, clustering), computer vision (Image Analysis, Face, Document Intelligence), NLP (Language, Translator, Speech), and generative AI (Azure OpenAI, prompt engineering basics, Copilot). AI-900 is **not** an architect exam. It is a vocabulary exam -- the value is forcing you to learn the names and shape of every Azure AI service so you are not learning vocabulary and architecture at the same time on AI-102 or AB-100. - **Time investment**: 20-40 hours over 2-4 weeks - **Course**: https://www.prepmycert.com/course/ai-900-microsoft-azure-ai-fundamentals - **Official study guide**: https://learn.microsoft.com/en-us/credentials/certifications/azure-ai-fundamentals/ If you already build AI solutions for a living, **skip AI-900 and start at AI-102**. ## AI-102: the associate that opens the AB-100 door **Azure AI Engineer Associate (AI-102)** is where most technical candidates begin. It is also the cleanest prerequisite for AB-100 -- of the 13 associate certs that satisfy the AB-100 gate, AI-102 is the only one in the AI family proper (alongside the Azure AI Apps and Agents Developer Associate, still in beta as of mid-2026). The blueprint covers planning and managing Azure AI solutions (service selection, security, monitoring), generative AI (Azure OpenAI / Foundry, prompt engineering, RAG, content safety), agentic solutions (Foundry Agent Service, tool calling, MCP basics), computer vision (Image Analysis, Face, Document Intelligence), NLP (Language, translation, speech), and knowledge mining (Azure AI Search, vector and hybrid search). AI-102 sits at the *implementation* layer. It tests whether you can wire services together and write working code -- not whether you can architect a multi-agent business solution end-to-end. - **Time investment**: 60-100 hours over 6-10 weeks - **Course**: https://www.prepmycert.com/course/ai-102-microsoft-azure-ai-engineer-associate - **Official study guide**: https://learn.microsoft.com/en-us/credentials/certifications/azure-ai-engineer/ ## AB-100: the expert architect credential **Microsoft Certified: Agentic AI Business Solutions Architect (AB-100)** is the top of the ladder. The exam launched in beta in late 2025 and reached general availability in January 2026. Where AI-102 asks *can you build this?*, AB-100 asks *should you build this, and how would you make it survive in production?* The blueprint is divided into three functional areas, weighted heavier toward deploy than the other two: | Skill area | Weight | |------------|--------| | Plan AI-powered business solutions | 25-30% | | Design AI-powered business solutions | 25-30% | | Deploy AI-powered business solutions | 40-45% | Unlike AI-102, AB-100 covers the full Microsoft AI stack -- not just Azure AI. Expect questions on Copilot Studio, Microsoft Foundry, the Microsoft Agent Framework, Microsoft 365 Copilot, Dynamics 365 AI features, the Power Platform Well-Architected Framework, Microsoft Purview, Azure Key Vault, MCP, A2A, and the OAuth 2.0 OBO flow. The full AB-100 study guide is at https://learn.microsoft.com/en-us/credentials/certifications/resources/study-guides/ab-100. For the deeper study plan and common-mistake list, see our companion guide at https://www.prepmycert.com/blog/ab-100-exam-guide-2026. - **Time investment**: 80-150 hours over 6-12 weeks for candidates with two-plus years of hands-on Microsoft AI architecture experience - **Course**: https://www.prepmycert.com/course/ab-100-microsoft-agentic-ai-business-solutions-architect - **Passing score**: 700 / 1000 ## Timeline: how long does the full path take? A reasonable, realistic schedule for someone working full-time in a related role: | Step | From | To | Effective hours | Calendar time | |------|------|-----|----------------|---------------| | 1 | Cold start | AI-900 pass | 20-40 | 2-4 weeks | | 2 | AI-900 pass | AI-102 pass | 60-100 | 6-10 weeks | | 3 | AI-102 pass | AB-100 pass | 80-150 | 8-16 weeks | | **Total** | Cold start | AB-100 | **160-290** | **4-8 months** | If you skip AI-900 (recommended for anyone already technical), shave one to two months. If you already hold an associate-level cert in Dynamics 365 or the Power Platform, you can sit AB-100 directly without AI-102 -- but expect to backfill the Azure AI plumbing material that AI-102 covers, because AB-100 assumes you know it. ## Salary progression at each step Salary data below is drawn from US-market sources -- Glassdoor, Levels.fyi, the Microsoft 2025 Partner Skilling Report, and 2026 talent surveys from Burning Glass and LinkedIn Talent Insights. Numbers are directional, not contractual. | Stage | Typical role | US median base | US median total | |-------|--------------|----------------|------------------| | Pre-cert | AI-curious developer or analyst | $85,000-$110,000 | $95,000-$125,000 | | AI-900 | Same role, slight uplift | $90,000-$115,000 | $100,000-$130,000 | | AI-102 | Azure AI Engineer | $115,000-$150,000 | $135,000-$180,000 | | AB-100 | AI Solutions Architect / Agentic AI Architect | $160,000-$215,000 | $190,000-$285,000 | The largest single jump is AI-102 to AB-100 -- the seniority and scope gap between an implementer and an architect. The smallest jump is pre-cert to AI-900: fundamentals alone do not move base pay much, but they open doors to AI-adjacent projects that build the experience for bigger steps. For partner-side consultants and pre-sales architects, the AB-100 multiplier on day-rate billing tends to exceed the in-house salary multiplier, because Microsoft Partner Programme competencies and customer-facing trust both lean on expert credentials. ## How each cert's content appears in a real architecture To make this concrete: a global retailer wants a Copilot Studio agent that helps store associates resolve customer service tickets. The agent searches the knowledge base, calls SAP for order data, drafts a reply for the associate to approve, and -- when policy allows -- issues a refund autonomously. Here is where each cert shows up: | Layer | What's there | Cert that covers it | |-------|--------------|---------------------| | Vocabulary: "what is a generative AI agent" | Conceptual understanding | AI-900 | | Wire Copilot Studio to Azure AI Search; use Document Intelligence for legacy invoices | Service implementation | AI-102 | | Decide between Copilot Studio, Foundry Agent Service, or Agent Framework SDK | Architecture decision | AB-100 | | Choose between an MCP server vs a Power Platform connector for the SAP call | Architecture decision | AB-100 | | Add OAuth 2.0 OBO flow so the agent acts as the store associate | Identity architecture | AB-100 | | Add a golden-dataset evaluation gate in CI before each promotion | ALM architecture | AB-100 | | Add Purview sensitivity labels and audit trails to the refund path | Governance | AB-100 | | Choose GPT-4o-mini for drafting and o3 for the policy decision | Model selection | AI-102 + AB-100 | | Calculate the ROI vs the cost of a model router | Business case | AB-100 | A single project touches all three certs. Fundamentals gives you the words. Associate gives you the wiring. Expert gives you the judgement to choose what *not* to wire. ## Common order-mistakes that cost people six months From watching candidates progress through this track, the same mistakes show up repeatedly. ### 1. Skipping AI-102 and jumping to AB-100 cold AB-100 assumes you know the Azure AI plumbing. Candidates who skip AI-102 without equivalent hands-on experience tend to over-index on architecture diagrams and under-index on the implementation details that case studies bury in paragraph three. The fix is to either sit AI-102 first or build at least one production-grade Foundry deployment before booking AB-100. ### 2. Sitting AB-100 without the associate prerequisite active You can sit AB-100 without the prerequisite, but Microsoft will not issue the credential until an approved associate is on your transcript. Candidates pass AB-100, wait for the certificate, and discover they need to also pass AI-102 (or equivalent) before the credential is awarded. Sequence the associate first. ### 3. Treating AI-900 as mandatory when you are already technical Fundamentals exams are valuable for non-technical learners. For working developers, they consume two weekends without moving the needle on hiring or salary. Skip AI-900 if you already build software. ### 4. Choosing the wrong associate prereq for your background AB-100's associate list spans AI, Dynamics 365, and Power Platform families. If you live in Dynamics 365 day-to-day, an MB-series associate is faster for you than AI-102. If you live in Azure development, AI-102 is faster. Choose the one closest to your actual job, not the one with the most prestige. ### 5. Studying from the November 2024 Azure AI Foundry material Microsoft renamed Azure AI Foundry to **Microsoft Foundry** in January 2026 and pushed three services to GA at the same time. The AB-100 blueprint uses the new vocabulary; late-2024 material misses small but scoring-relevant updates. ### 6. Underestimating the deploy domain on AB-100 The deploy section is 40-45% of the exam and covers ALM, testing, monitoring, responsible AI, security, and governance. Developer-trained candidates under-prepare here and over-prepare on model selection. Fix: spend a week building a CI-promoted Copilot Studio agent with a golden-dataset evaluation gate. ### 7. Booking the exam too early Booking AB-100 within a month of passing AI-102 leads to first-attempt failures. Build a real Foundry + Copilot Studio + Agent Framework project first, then sit two full-length mocks under timed conditions, then book. ## Recommended sequence by background A quick lookup for the most common starting points. | Background | Recommended sequence | |------------|----------------------| | Brand-new to Microsoft and AI | AI-900 -> AI-102 -> AB-100 | | Working Azure developer, no AI | AI-102 -> AB-100 | | Working Azure AI developer | AI-102 (if not already held) -> AB-100 | | Power Platform / Copilot Studio maker | AB-620 (or PL-200) -> AB-100 | | Dynamics 365 functional / technical consultant | MB-series associate (you likely already hold one) -> AB-100 | | Senior cloud architect with no AI exam | AI-102 -> AB-100 | | Senior data scientist with Azure exposure | AI-102 -> AB-100 | When you are ready for realistic, exam-format questions on AB-100, sit a full-length mock at https://www.prepmycert.com/course/ab-100-microsoft-agentic-ai-business-solutions-architect. ## Frequently asked questions ### Do I need AI-900 before AI-102? No. AI-900 is not a formal prerequisite for AI-102. If you are already technical, skip it. If you are new to Microsoft or to AI, AI-900 saves you time on AI-102 by front-loading vocabulary. ### Do I need AI-102 before AB-100? Not specifically AI-102 -- but you do need at least one approved associate-level credential from the AI, Dynamics 365, or Power Platform families. AI-102 is the cleanest fit for candidates with an Azure AI background. ### Can I sit AB-100 without any prerequisite cert? You can sit the exam and even pass it, but Microsoft will not issue the AB-100 credential until an approved associate is active on your transcript. Sequence the associate first. ### How long does the full path take? Four to eight months of part-time study, depending on starting point. Candidates who already work as Azure AI developers can compress to three to four months. ### Is AB-100 worth the effort compared to AI-102? If you architect AI solutions, yes. AB-100 is the only expert-level Microsoft AI credential focused on agentic AI and is increasingly named in job descriptions for AI architect roles. The salary delta between AI-102 and AB-100 typically exceeds the cost and time investment within the first year. ### What is the difference between AB-100 and the Azure Solutions Architect Expert (AZ-305)? AZ-305 is a general cloud architecture credential. AB-100 is specialised in AI and agent design. They are complementary, not substitutes -- a senior AI architect on Azure often holds both. ### How does Microsoft Foundry change the cert path? Microsoft Foundry (renamed from Azure AI Foundry in January 2026) is the platform tested in both AI-102 and AB-100. The rename does not change the path -- it changes the vocabulary in the question stems. Make sure your study material uses the 2026 terms. ### How often do these certs need renewal? All three renew every 12 months through free Microsoft Learn assessments. AB-100 renewal is a single online assessment, not a re-sit of the full proctored exam. ### What if I already hold an MB-series Dynamics 365 associate? You already satisfy the AB-100 prerequisite. Plan to backfill the Azure AI plumbing material that AI-102 covers, because AB-100 assumes familiarity with Foundry, Azure AI Search, and Document Intelligence. ## Next step The restructured Microsoft AI track is the clearest career ladder in enterprise AI today: fundamentals -> implementer -> architect. Each step compounds the next, and the salary numbers reflect it. When you are ready for the top of that ladder, the AB-100 course at https://www.prepmycert.com/course/ab-100-microsoft-agentic-ai-business-solutions-architect gives you 599 questions, six full-length mock exams, and detailed explanations mapped to Microsoft's official skill areas. Good luck on the path.