If you saw a widely shared list of top AI architecture certifications for 2026 and thought you’d finally pick Microsoft’s AI-900 or AI-102 as your next exam, check again. Microsoft retired both exams on June 30, 2026—weeks before the July 18 article that recommended them was even published.

That’s more than a scheduling slip. For anyone building a certification roadmap, drawing up training budgets, or booking a test center appointment, the old exam codes are dead ends. Microsoft has already replaced them with two new credentials: AI-901 and AI-103.

The switch is part of a broader overhaul of Azure’s AI certification ladder, and it changes both the skills you’ll need to prove and the hands-on tools you’ll encounter on exam day. Here’s exactly what changed, who’s affected, and what you should do if those exams were on your list.

What Actually Changed

On June 30, 2026, Microsoft retired the Azure AI Fundamentals exam (AI-900) and the Azure AI Engineer Associate exam (AI-102). Their replacements are live now:

  • AI-900 → AI-901 (Azure AI Fundamentals)
  • AI-102 → AI-103 (Azure AI Apps and Agents Developer Associate)

The new exams aren’t just renumbered reskins. Microsoft rewrote the skill profiles, added prerequisites, and shifted the technical depth notably upward for the fundamentals offering.

AI-901, according to Microsoft Learn, now expects candidates to have “foundational technical skills, familiarity with Azure resources, and knowledge of Python syntax and programming techniques.” That’s a significant departure from the older AI-900, which Microsoft explicitly designed for non-technical professionals—think business analysts, sales teams, or managers who needed AI literacy without deep data-science chops. The AI-900 exam covered AI concepts, machine learning basics, computer vision, natural language processing, and responsible AI, but it didn’t require coding or hands-on platform work.

AI-901 splits its objectives into two broad areas: identifying AI concepts and capabilities, and implementing AI solutions using Microsoft Foundry. Microsoft Foundry, which bundles tools for building, training, and deploying models, is a core piece of the Azure AI stack now, and the exam reflects that practical bent.

For the engineer track, AI-103 replaces AI-102 and leads to the Azure AI Apps and Agents Developer Associate certification. Microsoft says the exam tests your ability to design, develop, and deploy Azure AI solutions with Python and Microsoft Foundry. The subject matter includes generative AI, agentic systems, computer vision, text analysis, and information extraction—a modern workload that covers the kinds of multi-agent, retrieval-augmented generation (RAG) patterns that enterprises are adopting right now. The older AI-102 focused more narrowly on building AI solutions with Azure Cognitive Services and implementing responsible AI practices; agentic systems weren’t part of the picture.

Both new exams are available through Microsoft’s standard testing channels, and Microsoft Learn already hosts updated learning paths. The old exams are no longer offered; any registration for AI-900 or AI-102 after June 30 would have been voided or automatically transferred, though Microsoft hasn’t published a specific transition policy. If you held the old certifications, they may still appear on your transcript but won’t be recognized as current.

What It Means for You

If you’re a Windows administrator, IT generalist, or software developer who uses Azure as part of your stack, these changes aren’t cosmetic. They affect which credential you’ll cite on resumes, which courses your employer will reimburse, and what you actually learn along the way.

For the IT generalist or career switcher

AI-900 was often the “safe” entry point—no coding, no heavy platform work. AI-901 tears up that assumption. You’ll now need at least a working knowledge of Python, plus hands-on familiarity with Azure resources. That doesn’t mean you must be a full-stack developer, but it does mean you can’t walk in cold with just conceptual knowledge. If Python isn’t in your toolkit yet, factor in a few weeks of coding practice before you schedule the exam. Microsoft’s own learning materials for AI-901 include Python fundamentals modules, but they’re not a substitute for real-world coding.

For experienced developers and architects

The move to AI-103 is likely welcome if you’re already building agentic AI applications or working with Microsoft Foundry. The exam’s coverage of generative AI, computer vision, and information extraction aligns with real workloads. But if you had been studying for AI-102 using older materials (which some learners do over months), stop. The domains don’t match. Dump your AI-102 study guide and pivot to the AI-103 content on Microsoft Learn. Check the official exam outline for the exact weighting—Microsoft typically publishes a skills measured document that breaks down each area’s percentage, so you can allocate study time accordingly.

For training managers and budget planners

If your team’s certification plan listed AI-900 or AI-102 for the latter half of 2026, those line items are now invalid. Update the plan to AI-901 and AI-103, but also audit the team’s readiness. Because AI-901 demands Python, you might need to add a preparatory workshop before the certification track. Vouchers purchased for the old exams might be redeemable for the new ones, but you’ll need to check with Microsoft’s certification support or your training partner. Don’t assume automatic transfer.

For students and educators

Many university and bootcamp programs aligned their intro-to-Azure-AI modules with AI-900. Those curricula are now out of sync. If you’re enrolled in a course titled “Azure AI Fundamentals” that still references AI-900, ask the instructor when the syllabus will update. Using deprecated exam content to study is a waste of time and money.

How We Got Here

Microsoft’s AI certifications have been in constant motion for several years. The AI-900 exam, introduced in 2020, served as a gentle on-ramp during a period when “democratizing AI knowledge” was the buzz. It remained stable through multiple minor updates, but the ground shifted fast. Generative AI, large language models, and multi-agent systems exploded, and Microsoft responded with a wave of new services: Azure OpenAI Service, Semantic Kernel, prompt flow, and eventually the rebranded Microsoft Foundry.

By early 2026, it was clear the old certification ladder no longer matched the technology stack. Microsoft began signaling at industry events that a certification refresh was coming, though the company didn’t announce a specific retirement date until close to the deadline. The June 30, 2026 retirement gave learners only a few months’ notice, catching many off guard.

The timing explains why Analytics Insight’s July 18 article ended up with stale recommendations. The piece was likely researched and drafted weeks earlier, before the retirement date, and published without a final fact-check against the current Microsoft Learn catalog. It’s a reminder that the certification landscape, especially in AI, can shift underneath you while a blog post is still in editing.

Other cloud providers aren’t standing still, either. AWS and Google Cloud have also introduced or refreshed AI certifications recently, but their timelines haven’t collided as sharply with a single popular guide. Microsoft’s overhaul stands out because it touched both the entry-level and intermediate tracks at once.

What to Do Now

Step one: stop using any study materials that reference AI-900 or AI-102 as current exams. That includes the Analytics Insight article, third-party books published before July 2026, and old practice tests.

Step two: visit the official Microsoft certification page for AI-901 and AI-103. Microsoft Learn is the authoritative source; every other guide should be cross-checked against the skills measured documents there. Bookmark these pages and share them with your team.

Step three: assess your skill gaps. For AI-901, can you write simple Python scripts? Are you comfortable navigating the Azure portal and provisioning a resource? If not, Microsoft offers a free learning path that covers these basics. Don’t skip the Python piece—an AI fundamentals credential that requires programming is a new reality, and Microsoft isn’t likely to roll that back.

Step four: for AI-103, you’ll need more than just Python. Familiarize yourself with Microsoft Foundry, review the concept of agentic AI (AI systems that can plan, reason, and take action), and get hands-on with Azure AI services. The free tier of Azure is enough to experiment with many of these tools. Build a small project: maybe a RAG-based Q&A system or a multi-step agent workflow. Practical experience will matter both for the exam and for the career it’s meant to advance.

Step five: if you already earned AI-900 or AI-102, consider whether adding the new credential makes sense. Microsoft hasn’t announced a direct upgrade path, but holding the previous version might still carry some historical value on your transcript. For most professionals, though, employers will gravitate toward candidates who hold the current certification. Weigh the time investment against your job requirements.

Step six: if your organization relies on a different cloud platform, this might be a moment to compare. The Analytics Insight article also recommended AWS Certified AI Practitioner and AWS Certified Machine Learning Engineer – Associate, both of which remain valid and current as of mid-2026. Google Cloud’s Generative AI Leader and Professional Machine Learning Engineer certs are similarly unaffected. However, the same rule applies: always verify a certification’s status directly with the vendor before committing time or money.

Outlook

Microsoft’s AI certification track will likely see more rapid iterations as the tools evolve. The integration of Microsoft Foundry into the exam blueprints signals that the company wants certified professionals to prove competency in its latest platform, not just theoretical concepts. Expect additional specialized credentials down the line, possibly covering advanced agentic systems or specific industries.

For now, the message is clear: the AI-900 and AI-102 era is over. Update your plans, learn some Python, and point your career at the tools that matter in 2026 and beyond.