Direct Answer: Take AI-103 — AI-102 Is Already Retired
As of July 2026, the choice is no longer "sit AI-102 before the deadline." Microsoft retired Exam AI-102 and the Azure AI Engineer Associate certification on June 30, 2026, at 11:59 PM Central Time. The current associate path for Azure AI app and agent work is Exam AI-103: Developing AI Apps and Agents on Azure, which earns Microsoft Certified: Azure AI Apps and Agents Developer Associate.
If you still hold Azure AI Engineer Associate, keep it on your résumé until it expires — Microsoft leaves earned credentials on your transcript until normal expiry — but you cannot renew it after the retirement date. If you studied AI-102 material and never sat the exam, do not keep drilling the old outline as if it were still bookable. Migrate the transferable labs into an AI-103 plan centered on Microsoft Foundry, RAG, and agents.
This post is a decision and migration guide, not another full exam walkthrough. For the AI-103 blueprint, cost, and eight-week plan, use the AI-103 exam guide. For the final AI-102 outline and historical context, see the AI-102 exam guide. Official sources: the AI-103 study guide, the AI-102 study guide (retirement notice), and the Azure AI Apps and Agents Developer Associate page.
Snapshot Comparison (Final AI-102 vs Current AI-103)
| Factor | AI-102 (retired) | AI-103 (current path) |
|---|---|---|
| Full exam name | Designing and Implementing a Microsoft Azure AI Solution | Developing AI Apps and Agents on Azure |
| Credential | Azure AI Engineer Associate | Azure AI Apps and Agents Developer Associate |
| Status (July 2026) | Retired June 30, 2026 | Active associate path (verify GA/beta status on Microsoft Learn before booking) |
| Core platform framing | Broad Azure AI services + Foundry generative work | Microsoft Foundry first for apps and agents |
| Languages expected | Python or C# | Python emphasized in the audience profile |
| Duration | 100 minutes | 120 minutes |
| Passing score | 700 / 1000 | 700 / 1000 |
| Languages offered | Multiple (English plus several localizations) | English listed on the certification page |
| Generative + agents weight | Generative 15-20% + agentic 5-10% (separate) | Combined generative AI and agentic solutions 30-35% |
| Planning weight | 20-25% | 25-30% |
| Classical NLP / CLU / Custom QA | Explicit NLP domain 15-20% | Reframed as LLM-driven text analysis 10-15% |
Always recheck price, language availability, and release status on Microsoft Learn before you schedule. Associate exam fees commonly land near $165 USD in the United States, but Microsoft prices by country or region and can change without notice.
What Actually Changed With Microsoft Foundry and Agents
Competitor pages often say "AI-103 is more generative" and stop there. The official outlines show a sharper shift.
Agents moved from a side chapter to the center of gravity. On AI-102's final December 23, 2025 skills measured list, "Implement an agentic solution" was only 5-10%. On AI-103's April 16, 2026 list, agents are folded into Implement generative AI and agentic solutions at 30-35% — the largest domain. That domain expects agent roles and goals, tool schemas, function calling, conversation memory, multi-agent orchestration, autonomous or semi-autonomous workflows with approval controls, tracing, and error analysis (AI-103 study guide).
Foundry is the default workspace, not an optional studio. AI-103 planning asks you to choose Foundry services for generative tasks, grounding, vector search, agent workflows, and multimodal processing; set up projects and deployments; wire CI/CD; manage quotas, drift, safety events, and grounding quality; and apply responsible AI guardrails across generative and agentic systems. AI-102 already used Microsoft Foundry language in its final outline, but AI-103 assumes Foundry as the operating system for almost every scenario.
Vision and extraction stayed — framing changed. Computer vision remains 10-15% on both exams, but AI-103 adds image and video generation, editing workflows, multimodal understanding, captions, visual Q&A, and Content Understanding pipelines, plus image-based prompt-injection defenses. Knowledge mining on AI-102 becomes information extraction on AI-103: ingest documents, images, audio, and video; configure semantic, hybrid, and vector search; connect retrieval to agent tools; and produce grounded representations for RAG.
Classical NLP shrinks. AI-102's NLP domain covered Language features, Translator, Speech, CLU, and Custom Question Answering at 15-20%. AI-103's text analysis domain is 10-15% and leans on generative prompting, structured JSON outputs, sentiment and safety, Translator or LLM translation, and speech as an agent modality. If your AI-102 notes are full of CLU intents and Custom QA knowledge bases, treat that as background — not the AI-103 scoring core.
Who Should Take Which (Post-Retirement Decision Tree)
| Your situation in mid-2026 | What to do |
|---|---|
| Starting Azure AI associate prep from scratch | Prepare for AI-103 only |
| Studied AI-102 for weeks but never sat it | Migrate to AI-103; keep overlapping labs, rebuild agents/Foundry |
| Already hold Azure AI Engineer Associate and it is still active | Keep the credential until expiry; plan AI-103 before it lapses if you need a current associate signal |
| Renewal window for Azure AI Engineer opened before June 30, 2026 and you renewed | You bought time until the new expiry; still plan AI-103 for the next cycle |
| Job is building Foundry agents, RAG apps, tool calling, evaluation | AI-103 is the matching exam |
| Job is mostly classical Vision/Language/Speech APIs with little generative work | AI-103 still covers those areas, but you must add Foundry generative and agent labs to pass |
| Need fundamentals first | Start with Azure AI fundamentals (AI-900 / successor path), then AI-103 — do not treat fundamentals as a substitute |
The honest nuance Microsoft's own pages support: AI-103 is the current Azure AI engineering-style associate exam for Foundry apps and agents, not a renamed AI-102. Credentials differ. Skills overlap. Study transfer is partial.
Migrate an AI-102 Study Plan Without Starting Over
Use this keep / drop / add map so you do not throw away useful work.
Keep (still high value on AI-103)
- Azure AI Search vector, hybrid, and semantic retrieval patterns
- Document Intelligence / Content Understanding extraction pipelines
- Responsible AI: content filters, Prompt Shields, groundedness, safety monitoring
- Basic generative patterns: prompt parameters, structured outputs, RAG grounding
- Identity and security habits: managed identity, private networking, keyless credentials, cost and quota awareness
Drop or deprioritize (low AI-103 yield)
- Deep CLU / Custom QA authoring workflows as primary study time
- AI-102-only timing strategies for a 100-minute, multi-language exam seat
- Memorizing service silos without a Foundry project that ties them to an agent or RAG app
- Treating agents as a 5-10% "nice to have" chapter
Add (the migration gap most candidates underestimate)
- A Foundry project with model deployment, managed identity, and cost/quota controls
- A RAG app that grounds answers in your own index and measures relevance and groundedness
- An agent with at least two tools (for example search plus a custom function), memory strategy, and an approval or safeguard path
- Multi-agent or orchestrated workflow practice, even at a small scale
- Observability: tracing, token analytics, safety signals, latency breakdowns
- Multimodal generation or understanding labs beyond classic OCR/classification
Two- to four-week bridge plan (if you already finished an AI-102 outline)
| Week | Focus | Exit criteria |
|---|---|---|
| 1 | Diff the official AI-103 outline against your AI-102 notes | Every AI-103 bullet mapped to keep/add/drop |
| 2 | Foundry planning + generative app | Deployed model, RAG path, evaluation metrics recorded |
| 3 | Agents | One agent with tools, memory, monitoring, and a failure analysis note |
| 4 | Timed AI-103 practice + weak-domain labs | Consistent reasoning on Foundry/agent scenarios; vision/text/extraction refresh only where you miss |
If you are brand new, do not use a compressed bridge. Follow a full AI-103 plan such as the eight-week path in the AI-103 exam guide, and use AI-102 material only as optional background reading.
Side-by-Side Domain Weights (Official Outlines)
| AI-102 final domains (Dec 23, 2025) | Weight | AI-103 domains (Apr 16, 2026) | Weight |
|---|---|---|---|
| Plan and manage an Azure AI solution | 20-25% | Plan and manage an Azure AI solution | 25-30% |
| Implement generative AI solutions | 15-20% | Implement generative AI and agentic solutions | 30-35% |
| Implement an agentic solution | 5-10% | (merged into generative + agentic) | — |
| Implement computer vision solutions | 10-15% | Implement computer vision solutions | 10-15% |
| Implement natural language processing solutions | 15-20% | Implement text analysis solutions | 10-15% |
| Implement knowledge mining and information extraction | 15-20% | Implement information extraction solutions | 10-15% |
Sources: AI-102 study guide and AI-103 study guide.
The arithmetic is the study strategy: generative plus agents jumped from roughly 20-30% combined on AI-102 to 30-35% alone on AI-103, while planning also rose. That is why "I already know Cognitive Services" is not an AI-103 pass plan.
Common Mistakes After the Retirement Date
Pretending AI-102 is still schedulable. Pearson VUE bookings for AI-102 ended with the retirement cut-off. New seats are AI-103.
Assuming the credentials are interchangeable on a résumé. Azure AI Engineer Associate and Azure AI Apps and Agents Developer Associate are different names. Be precise with employers and on LinkedIn.
Reusing AI-102 dumps and calling it AI-103 prep. Overlap exists, but the highest-weighted AI-103 scenarios are Foundry agents, RAG evaluation, tool schemas, and production safety — areas many AI-102 banks under-sample.
Skipping hands-on agents because "I understand the concept." AI-103 tests whether you can define roles, wire tools, manage memory, orchestrate, and debug with traces — not whether you can define an agent in one sentence.
Ignoring holders' renewal reality. If your Azure AI Engineer Associate is still active, use the remaining validity window to build Foundry portfolio work and schedule AI-103 before the credential expires. Microsoft's credential retirement guidance is the policy source of truth.
FREE Practice Path on OpenExamPrep
Turn the comparison into hours on the right question bank:
- Take a mixed set of FREE AI-103 practice questions first — that is the exam you can still earn.
- Use FREE AI-102 practice questions only to refresh overlapping vision, search, extraction, and responsible AI judgment while you migrate.
- After misses, rebuild the matching lab (Foundry project, RAG index, agent tools, Content Understanding) instead of only rereading notes.
- For skill-by-skill AI-103 drilling, pair practice with the AI-103 practice questions by skills post.
- Use the AI-103 study guide and AI-102 study guide pages when you need structured topic navigation alongside practice.
OpenExamPrep also includes an AI tutor with 10 free AI interactions per day for explanations and study-plan prompts — enough to turn a weak domain into a short remediation block without paying for a bootcamp.
Official Sources to Recheck Before You Book
- AI-103 study guide (skills measured as of April 16, 2026): https://learn.microsoft.com/en-us/credentials/certifications/resources/study-guides/ai-103
- AI-103 certification page (120 minutes, English, Foundry/Python audience): https://learn.microsoft.com/en-us/credentials/certifications/azure-ai-apps-and-agents-developer-associate/
- AI-102 study guide (retired June 30, 2026 notice + final Dec 23, 2025 outline): https://learn.microsoft.com/en-us/credentials/certifications/resources/study-guides/ai-102
- Azure AI Engineer Associate page (retired certification warning): https://learn.microsoft.com/en-us/credentials/certifications/exams/ai-102
- Credential retirement policy: https://learn.microsoft.com/en-us/credentials/support/credential-retirement
- Exam scoring (700 to pass): https://learn.microsoft.com/en-us/credentials/certifications/exam-scoring-reports
Bottom Line
AI-102 vs AI-103 in 2026 is no longer a toss-up. AI-102 is retired. AI-103 is the associate exam that matches how Microsoft now expects Azure AI engineers to build: Foundry projects, generative apps, RAG grounding, and agents with tools, memory, evaluation, and safety. Keep any active Azure AI Engineer Associate credential until it expires, migrate transferable AI-102 labs, and spend your remaining study hours on the 30-35% generative-and-agentic domain that decides most AI-103 scores.
Buy on Amazon
Take courses on Udemy
Learning path on Pluralsight