Azure AI Engineer Associate (AI-102): The Complete Exam Guide
Read this before you book anything: exam AI-102 retired on June 30, 2026. The credential it led to, Microsoft Certified: Azure AI Engineer Associate, is still on transcripts for everyone who already passed, but you can no longer sit AI-102. Microsoft replaced it with AI-103 (Azure AI Apps and Agents Developer Associate). The old exam cost $165, needed a 700/1000 to pass, and assumed you could already write Python or C#. If you landed here planning to study for it, the guide below tells you what changed and where to point your effort now.
Quick facts: the Azure AI-102 exam at a glance
| Detail | AI-102 |
|---|---|
| Certification | Microsoft Certified: Azure AI Engineer Associate |
| Exam status | Retired June 30, 2026 (11 PM CST) |
| Replacement | AI-103 → Azure AI Apps and Agents Developer Associate |
| Cost | $165 USD + local tax (Pearson VUE, US) (verify) |
| Passing score | 700 / 1000 |
| Exam time | 100 minutes (Microsoft’s stated duration) |
| Questions | ~40–60, mixed formats incl. case studies |
| Level | Intermediate, role-based associate |
| Prerequisites | None required; Python or C# and REST/SDK experience assumed |
| Renewal | Annually, free online assessment on Microsoft Learn (now retired with the exam) |
| Languages | English + Japanese, Korean, German, French, Spanish, Italian, Portuguese (BR), Chinese (Simplified & Traditional) |
Sources: Microsoft Learn certification and exam pages, plus the official AI-102 study guide (last updated December 23, 2025). Price is the standard US associate-tier fee, confirmed by third-party listings; Microsoft no longer prints it on the cert page.
The retirement, explained (and why it matters right now)
Microsoft ran one of its biggest cert overhauls in years through the first half of 2026. AI-102 was on the list. So was the fundamentals exam it sits above. AI-900 retired the same day and became AI-901. The pattern is the same across both: Microsoft is rebuilding its AI track around Microsoft Foundry (the platform formerly branded Azure AI Foundry) and around agents.
What that means in plain terms:
- The certification name lives on. Microsoft Certified: Azure AI Engineer Associate credentials already earned stay valid and verifiable on your transcript. They don’t get revoked.
- The route changed. New candidates take AI-103: Developing AI Apps and Agents on Azure, which awards the Azure AI Apps and Agents Developer Associate credential. Microsoft’s own training course renumbered too: AI-102T00 became AI-103T00.
- Roughly three-quarters of the material carries over. Redditors mid-study when the news broke pegged it at “75–80% of AI-102 material will still be valid.” The Foundry SDK, Azure OpenAI, vision, language, and search services are the same; the framing shifts hard toward building and orchestrating agents.
If you were weeks from an AI-102 attempt when it closed, you didn’t waste the prep. You just point it at AI-103.
One timing detail worth knowing: Microsoft ran generous free-voucher windows in the run-up to the June 30 cutoff, and community posts flagged limits like two free vouchers per eligible learner before the retirement. Those deadline-driven offers are gone now that the exam is closed, but the same voucher machinery (Virtual Training Days, Cloud Skills Challenges, Exam Replay) typically spins back up for the replacement exam once it reaches general availability. If cost is your blocker, watch those channels before paying full price for AI-103.
Prerequisites: who the AI-102 exam was actually built for
Microsoft never gated AI-102 behind a required exam. But “no prerequisites” was misleading — this was never an entry-level test. The audience profile assumed you already:
- Write Python or C# comfortably, not just read it.
- Call REST APIs and SDKs and know what an endpoint, key, and deployment are.
- Understand the shape of the Azure AI portfolio and its storage options.
- Can apply responsible-AI principles in a real deployment, not just define them.
Most people who passed had touched Azure at work first. A common on-ramp was AI-900 for the vocabulary, then hands-on time in a free Azure account before booking. If you can’t yet deploy a model and hit it from code, the fundamentals exam is the honest starting rung. This azure ai certification rewards building, not memorizing service names.
Exam objectives and weightings (final AI-102 version)
The last published version of the exam (skills measured as of December 23, 2025) broke down like this:
| Domain | Weight |
|---|---|
| Plan and manage an Azure AI solution | 20–25% |
| Implement generative AI solutions | 15–20% |
| Implement natural language processing solutions | 15–20% |
| Implement knowledge mining and information extraction solutions | 15–20% |
| Implement computer vision solutions | 10–15% |
| Implement an agentic solution | 5–10% |
Two things stand out in that final revision. First, the whole exam was rewritten around Microsoft Foundry: service selection, deployment, monitoring, and content safety all reference Foundry rather than the old “Azure AI services” naming. Second, an agentic solution domain appeared: building custom agents with the Foundry Agent Service, wiring multi-agent workflows and orchestration with the Microsoft Agent Framework. That domain is small on paper (5–10%) but it’s the clearest signal of where AI-103 doubles down.
The generative-AI section is where most study hours go: provisioning Azure OpenAI, deploying models, grounding a RAG pattern in your own data, prompt flow, evaluation, fine-tuning, and prompt engineering. Plan and manage is broad but shallow: resource creation, cost management, key protection, authentication, and responsible-AI guardrails like content filters and prompt shields.
The hardest parts, honestly
Passers and third-party trainers agree on where AI-102 bites:
- Azure AI Search (knowledge mining). Indexes, skillsets, custom skills, indexers, semantic and vector search. It’s fiddly, it’s ~15–20% of the exam, and it’s the topic people skip and then fail on.
- Breadth over depth. Vision, speech, language, document intelligence, search, and generative AI all in one exam. You can’t cram one service and coast; the questions punish gaps.
- Foundry rebrand churn. After December 2025, a lot of prep material (courses, practice tests, blog walkthroughs) still used the old Azure AI Foundry / Azure AI services wording. Candidates lost time reconciling names that had quietly changed.
- Case studies and code. This isn’t multiple-choice trivia. Expect drag-and-drop ordering, code completion, and scenario questions where you pick the right service and the right SDK call.
A prep plan that works (HowTo)
A realistic path from “I know some Python” to a pass, whether you aim it at the retired AI-102 material or the live AI-103:
- Get the vocabulary first. If AI concepts feel shaky, clear AI-900 / AI-901 fundamentals first. It’s cheaper, it’s easier, and it never expires. See our AI-900 guide.
- Set up a free Azure account. You learn this exam by provisioning resources, not reading about them. A free tier plus a few dollars of pay-as-you-go covers the practice.
- Work the Microsoft Learn paths. The official self-paced modules map directly to the objective domains and are free. Do the labs, don’t just read them.
- Build one end-to-end thing. A small app that does OCR, sentiment, or a grounded RAG chatbot against your own docs will teach you more than 200 flashcards. Add a simple agent to cover the agentic domain.
- Take the free practice assessment. Microsoft’s own practice test mirrors the exam’s style and difficulty. Aim for 80%+ before you book; the pass bar is 700/1000, and comfort margin matters under a timer.
- Book smart. Register through Pearson VUE. Watch for Microsoft’s Exam Replay deals and free-voucher campaigns (Virtual Training Days, Cloud Skills Challenges) that regularly knock the $165 down or to zero.
Community-reported prep time ran from a couple of focused weeks for working Azure developers to a couple of months for people coming in cold. There’s no shortcut past the hands-on hours.
AI-102 vs AI-900 vs AWS: where it sits
People conflate the two Microsoft exams constantly, so here’s the split:
- AI-900 (fundamentals) is a no-code literacy test: $99, never expires, passable in 8–10 hours of study, built for managers and non-engineers. It proves you understand AI on Azure. It does not prove you can build it.
- AI-102 (associate) was the builder’s exam — $165, code required, and it expired annually until it retired. It proved you could ship an Azure AI solution.
That renewal difference caught people out. Unlike fundamentals certs, role-based associate credentials expired every year and had to be renewed through a free online assessment on Microsoft Learn. The assessment was shorter than the exam, took place in your browser, and cost nothing. But you had to remember to do it inside the renewal window, and letting a credential lapse meant re-sitting the full exam. Real credential, real upkeep. AI-103 inherits the same annual renewal model, so factor that into the “is a badge worth it” math before you start: this is a commitment that renews, not a trophy you earn once.
Against AWS, the closest analogue to AI-102 is AWS’s machine-learning engineering track rather than the foundational AWS Certified AI Practitioner. If you’re deciding which cloud’s AI badge to chase for your career, we break down cost, hiring signal, and difficulty in AWS vs Azure AI certification. And for where this azure ai engineer associate certification ranks against every other option (Google, IBM, NVIDIA, AWS), see our AI certifications ranking.
Is the Azure AI Engineer Associate worth it? The honest take
The Azure AI Engineer Associate was one of the more respected mid-tier cloud AI credentials, precisely because you couldn’t pass it by memorizing. It signaled real Azure development ability, and it mapped to jobs that exist.
The honest negatives: it never got anyone hired on its own. Recruiters read it as a filter, not a finish line. The annual renewal was a genuine chore for a busy engineer. And the December 2025 Foundry rebrand, landing months before retirement, left a trail of stale study material that wasted candidates’ time. Chasing the retired AI-102 now would be a mistake — there’s nothing left to sit.
The move that makes sense in mid-2026: if you want the Azure AI engineer credential, aim at AI-103 and its agent-first syllabus. Most of your AI-102 prep transfers. If you mainly need to prove you can build with modern AI, a portfolio of shipped projects still outargues any badge in an interview.
FAQ
Is the AI-102 exam still available in 2026?
No. Microsoft retired AI-102 on June 30, 2026, at 11
PM Central Standard Time. You can no longer schedule or sit it. The Azure AI Engineer Associate certification is now earned through the replacement exam, AI-103 (Azure AI Apps and Agents Developer Associate).What replaced the AI-102 exam?
AI-103: Developing AI Apps and Agents on Azure replaced it, awarding the Azure AI Apps and Agents Developer Associate credential. Microsoft reports about 75–80% of AI-102 content carries over; the new exam leans harder into building and orchestrating agents with Microsoft Foundry.
How much did the AI-102 exam cost?
The AI-102 exam fee was $165 USD plus applicable local tax, the standard price for Microsoft’s role-based associate exams in the US. Microsoft frequently discounted or fully waived it through Exam Replay deals, Virtual Training Days vouchers, and Cloud Skills Challenge campaigns.
What was the passing score for AI-102?
A score of 700 out of 1000 was required to pass, scaled — not a flat 70% of questions. The exam ran 100 minutes per Microsoft, with roughly 40 to 60 questions across multiple-choice, drag-and-drop, code, and case-study formats.
Do I need AI-900 before taking the AI-102 or AI-103 exam?
No, it was never required. But AI-900 (now AI-901) gives you the AI vocabulary cheaply and is a sensible first rung if the concepts feel new. The engineer-level exam assumes you can already write Python or C# and call Azure services from code.
Certificates open doors; they were never the room. Point your hours at AI-103 and something real to show, and the badge takes care of itself.