Home › Guides › AI-102 vs AI-103
Career Guide · 2026AI-102 vs AI-103: Which Azure AI Certification Should You Take in 2026?
AI-102 vs AI-103 comes down to one fact: Microsoft retired AI-102 on 30 June 2026 and replaced it with AI-103, the Azure AI Apps and Agents Developer Associate exam built around generative AI, RAG and autonomous agents instead of individual cognitive services. If you are certifying new in 2026, AI-103 is your only option.
- AI-102 retired on 30 June 2026 and can no longer be booked; AI-103 is now the only path to Microsoft's generative AI developer credential.
- AI-103 tests agent orchestration and RAG pipelines, skill areas AI-102 barely touched in four years on the market.
- The list fee hasn't moved, roughly $165 at Associate level, so the entry cost is unchanged even though the syllabus isn't.
- Existing AI-102 badges stay on your transcript but can no longer be renewed by retaking the old exam.
- Generative AI Developer pay in India typically clears 18 to 20 LPA by the five to eight year mark, with Foundry and agent-building experience pushing offers higher.
- Pairing AI-103 with a Claude architecture credential is the stronger 2026 bet, since most production deployments now mix Azure tooling with Claude rather than running Azure AI end to end.
Here's the scenario: it's November 2026, you want an Azure AI certification on your resume, and you type "AI-102" into the Microsoft Learn catalogue the way engineers have for four years. Nothing useful comes back for new bookings. That's not a glitch. Microsoft pulled AI-102 and handed the job to an exam built for different work: agents that call tools and act on retrieved context, not apps that ping a vision API and hand back a result. If you're a Pune-based .NET developer with two years on Azure App Service who kept meaning to book AI-102, that option is gone, and the replacement tests a different set of muscles.
AI-102 vs AI-103: What Changed in 2026
AI-102, officially "Designing and Implementing a Microsoft Azure AI Solution," covered a grab-bag of cognitive services: Computer Vision, Language, Speech, and a thin layer of orchestration logic to stitch them into an app. That syllabus made sense when "AI feature" meant calling a prebuilt model endpoint. It stopped making sense once most real Azure AI work became building agents that plan, call tools, retrieve documents and hand off between steps. Microsoft's own retirement notice set 30 June 2026 as the last day AI-102 could be scheduled, and AI-103 took over as the current path to the credential.
What Is the Azure AI Apps and Agents Developer Associate Credential?
AI-103 is Microsoft's associate-level exam for building generative AI and agent applications on Azure, delivered through Pearson VUE at a test centre or online with a proctor. Microsoft's own exam listing expects candidates to already write Python applications and have working familiarity with general AI concepts, generative AI patterns, and core Azure services before they sit it, which is a step up from AI-102's assumption that you were comfortable calling REST APIs. 360DT's Generative AI Developer course is built directly against this exam, pairing AI-103 prep with Microsoft Azure AI Foundry labs so you are practising on the same platform the exam assumes, not reading about it.
AI-102 vs AI-103 at a Glance
Put side by side, the two exams barely overlap in intent even though they share a credential lineage.
| Aspect | AI-102 (legacy) | AI-103 (current) |
|---|---|---|
| Status in Nov 2026 | Retired, not bookable | Active, the only option |
| Full exam name | Designing and Implementing a Microsoft Azure AI Solution | Developing AI Apps and Agents on Azure |
| Core focus | Calling individual cognitive services (Vision, Language, Speech) from an app | Building and orchestrating agents that plan, call tools and retrieve context |
| Primary dev surface | Azure AI services REST APIs and SDKs | Azure AI Foundry, agent and orchestration SDKs |
| Prerequisite assumption | Comfortable consuming REST APIs | Writes Python apps, understands generative AI concepts |
| Typical candidate | .NET or Python developer adding one AI feature to an existing app | Developer building a standalone agent or copilot-style product |
| List fee | $165 (retired) | Roughly $165, regional pricing and tax apply |
What AI-103 Actually Tests That AI-102 Didn't
The honest version of "AI-103 is harder" is that it isn't harder so much as it's a different exam wearing the same numbering scheme. If you studied AI-102 content in 2025 and assumed AI-103 is an incremental update, you will be surprised by how much of it is genuinely new material rather than a reshuffle.
| Skill area | In AI-102 | In AI-103 |
|---|---|---|
| Individual Azure AI service calls (Vision, Language, Speech) | Core focus, most of the exam | Present but lighter, treated as a building block |
| Prompt engineering and generative AI app design | Minimal, added late in the exam's life | Core focus from the start |
| Agent orchestration and multi-step tool calling | Not covered | Core focus, the exam's centre of gravity |
| Retrieval-augmented generation pipelines | Not covered | Covered as a standard pattern |
| Responsible AI guardrails for autonomous agents | Basic content moderation only | Expanded to cover agent-specific risk |
| Azure AI Foundry as the dev platform | Barely existed in early preview form | The primary environment the exam assumes |
The practical upshot: if your day job is building retrieval-augmented generation pipelines and multi-step agents, AI-103 will feel close to work you already do. If your day job is calling a single Azure Vision endpoint from a mobile backend, you'll spend real study hours on content that has nothing to do with what you build day to day.
Who Should Take AI-103, and Who Can Skip It
Here's where I'll commit to an opinion instead of hedging: skip the AI-102 study material entirely, even the PDF you downloaded last year. The syllabus overlap is maybe a third of the exam, and relearning around an old mental model costs you more study hours than starting fresh with AI-103's own guide. People who try to "adapt" their AI-102 notes consistently take longer to prep than people who start clean.
That said, here's the caveat worth sitting with before you book anything: if your actual job is writing prompts against a single hosted model with no Azure deployment responsibility, no tool-calling, and no infrastructure ownership, AI-103 will not move your resume much. It's an integration-and-infrastructure exam, not a prompting badge, and recruiters who know the difference will see through a credential that doesn't match your actual work.
AI-103 is worth the hours for three groups specifically: Azure developers moving from traditional app features into agent products, the data or ML engineers our MLOps Engineer course trains who now own a generative AI service in production, and anyone targeting a Generative AI Developer title where the job description mentions Azure AI Foundry, Semantic Kernel or agent orchestration by name.
What Is a Generative AI Developer in 2026?
A Generative AI Developer builds the application layer around a foundation model: retrieval pipelines that fetch the right documents, orchestration logic that decides which tool an agent calls next, guardrails that stop an agent from doing something it shouldn't, and the Azure or Claude plumbing that keeps all of it running in production. It's a step up from "I called the OpenAI API in a Flask app," and if you want proof you can do this before you even sit an exam, our generative AI project ideas guide has eight worth building. That gap between calling an API and running an agent in production is exactly what 360DT's Generative AI Developer Course: Azure AI-103 + Claude CCAR-F is built to close, with 50+ hours of live weekend instruction over 8 weeks covering both the Azure AI-103 syllabus and the Claude Certified Architect Foundations (CCAR-F) credential in the same cohort.
That CCAR-F pairing isn't filler. Most production generative AI systems we see built in Indian enterprises in 2026 aren't pure Azure or pure Claude, they're both: Azure Foundry for the infrastructure and governance layer, Claude models for the reasoning and tool-use layer where teams want stronger agentic behaviour. A developer who can speak both certification vocabularies reads as more deployable than one who only knows Azure's framing of the problem. If your path runs through data rather than app code, the same logic holds elsewhere in Microsoft's ladder: our Microsoft Fabric Data Engineer course maps to DP-700 the way this one maps to AI-103, and our Data Analyst course does the same for PL-300.
Generative AI Developer Salary in India 2026
Treat the numbers below as typically advertised ranges on Indian job postings that mention generative AI, Azure AI Foundry, or agent-building skills, not a guaranteed offer; actual pay still depends heavily on city, company type and whether the role is GCC, product, or services.
Generative AI Developer pay by experience, India 2026
Typically advertised ranges; GCCs and product companies usually sit at the top of each band
Typical advertised ranges on Indian job boards for generative AI and Azure AI roles, checked 2 November 2026. Not a placement guarantee.
The ceiling moves fastest for people who can show agent orchestration work, not just API integration. If you're coming from a role like Azure AI Cloud Developer (AI-200) work and want to climb into generative AI specifically, AI-103 plus a working RAG project on your resume is the more direct lever than a generic cloud certification stack.
Is AI-103 Worth It in 2026? Cost and ROI
AI-103 at a glance
The facts that actually change your prep decision
Figures reflect Microsoft's standard Associate-level exam structure as checked 2 November 2026; confirm the exact fee at Pearson VUE scheduling since pricing updates without notice.
$165 plus tax is not the real cost. The real cost is the retake fee if you go in underprepared, and the opportunity cost of studying the wrong material because you assumed AI-103 was "AI-102 with a new agent chapter bolted on." Worth it, in my view, if you're actually building or planning to build generative AI products on Azure. Not worth it as a resume decoration if you have no intention of touching Azure AI Foundry in your next role, in which case a different certification from 360DT's full lineup will serve you better.
How to Actually Prepare for AI-103
The fastest way to waste six weeks is studying AI-102 material "updated" with a few AI-103 search results bolted on. Here's the timeline of how the exam got here, which also tells you which resources are stale.
The AI-102 to AI-103 transition, start to finish
Where the confusion in forums and old study guides comes from
AI-103 goes live
Microsoft publishes the AI-103 skills-measured guide and opens bookings; AI-102 is still bookable in parallel for a short overlap window.
AI-102 retires
Last day to book or sit AI-102; no new scheduling after this date, and training content built against it stops getting updated.
Early AI-103 candidates sit a moving target
Third-party study guides lag the official content, which is why forum advice from this window is often already outdated.
AI-103 is the only live path
Study materials have caught up; AI-102 badges already earned stay on your transcript but can't be renewed via the old exam.
Timeline reconstructed from Microsoft's published retirement date and exam listing, checked 2 November 2026.
For the hour-by-hour plan, our dedicated AI-103 exam prep guide walks through a full six-week schedule. Here's the shorter version of how to split your prep hours if you only have the summary version:
Recommended AI-103 prep effort split
Our recommendation for where to spend study hours, not Microsoft's official exam weighting
A practical hour allocation we recommend for a six-week plan, checked 2 November 2026.
This is the structure 360DT's live cohorts follow: Foundry labs first, orchestration patterns second, legacy service fundamentals as a refresher, practice tests in the final two weeks. To spot your own gaps before committing six weeks, our generative AI developer interview questions guide doubles as a decent diagnostic.
What Usually Goes Wrong Here
A few patterns show up often enough with candidates that they're worth naming before you start, the way a colleague would warn you over a desk rather than the way documentation warns you.
- Studying old AI-102 dumps and assuming they transfer. They don't, not for the agent orchestration and RAG sections, which is most of the new exam.
- Skipping hands-on Azure AI Foundry practice. Reading about agent tool-calling and actually wiring one up in Foundry are different skills, and the exam leans toward the latter.
- Not checking the exact fee before scheduling. Microsoft adjusts pricing by country and currency, and the number you saw on a blog post six months ago may not match what you pay at checkout.
- Treating AI-103 as a standalone credential. On its own it signals Azure competence; paired with a Claude credential like CCAR-F, it signals you can work across the two stacks most real agent deployments actually use.
Related guides
- What Is an AI Agent Harness in 2026? useful background on the orchestration frameworks AI-103's agent sections assume you understand.
- Retrieval Augmented Generation Explained in 2026 walks through the RAG pipeline mechanics that now sit inside the AI-103 syllabus.
- Azure AI Developer Roadmap 2026 for readers comparing the AI-200 cloud developer path against the AI-103 generative AI path.
- What Are Claude Agent Skills in 2026? for the Claude-side equivalent of the agent patterns AI-103 tests on Azure.
- MLOps vs DevOps in 2026 if you're deciding between a generative AI developer path and an operations-heavy one.
- CCAR-F Interview Questions 2026 for the Claude-side interview prep that pairs with the AI-103 track.
Learn AI-103's agent and RAG patterns the way the exam tests them, hands-on
8 weeks of live, instructor-led sessions covering the Azure AI-103 syllabus and Claude Certified Architect Foundations (CCAR-F) in one cohort, with hands-on Azure AI Foundry and agent-building labs plus mentor support.
Explore the course
Frequently asked questions
AI-102 vs AI-103: which one should I prepare for right now?
AI-103, with no real choice involved. Microsoft retired AI-102 on 30 June 2026, so there is nothing left to compare if you haven't already sat the old exam.
Is my existing AI-102 certification still valid?
Yes, it stays on your Microsoft Learn transcript as earned. What changes is renewal: you can no longer extend it by retaking AI-102, so most badge holders move to AI-103 instead.
What is the AI-103 exam fee in India?
The list price sits around $165, Microsoft's standard Associate-level fee, with regional pricing and GST applied at checkout through Pearson VUE. Confirm the exact figure when you schedule, since Microsoft updates pricing without notice.
Is AI-103 worth it in 2026?
Worth it if you're building generative AI apps or agents on Azure, where it signals real platform familiarity. Weaker if your work never touches Azure AI Foundry, in which case a certification closer to your actual stack serves you better.
What is a Generative AI Developer's salary in India?
Postings mentioning generative AI, Azure AI Foundry or agent-building skills typically advertise 6 to 9 LPA for freshers, rising to 18 to 28 LPA at the five to eight year mark and 30 LPA or higher at lead and architect level.
Do I need to know Python for AI-103?
Microsoft's exam listing assumes you can already write Python applications, plus general familiarity with AI and generative AI concepts and core Azure services. Not an exam to walk into cold.
Should I pair AI-103 with a Claude certification?
Reasonable, given how production agent systems get built in 2026: Azure infrastructure mixed with Claude or another frontier model for reasoning and tool use. CCAR-F signals you can work across both stacks, not just one.
How long does it take to prepare for AI-103?
Six weeks is realistic for a working developer with existing Azure and Python experience, at roughly 8 to 10 hours a week split between Foundry labs, orchestration practice and review tests.
About this guide. 360 Digital Transformation is an Anthropic Partner and a Microsoft Authorized Training Partner. Other certification bodies, vendors and employers named here are not affiliated with us. Figures cited were checked on 2 November 2026.
Microsoft Foundry
GPT and Foundry Models
Python
