AI-103 Replaces AI-102: What Changed for AI Engineers

AI-103 replaces AI-102 as the Microsoft exam for Azure AI engineers, and it is a rewrite, not a refresh. According to Microsoft Learn, AI-102 retired on June 30, 2026, and the AI-103 skills outline is dated April 16, 2026.

AI-103 is called Developing AI Apps and Agents on Azure. It earns a new credential, Microsoft Certified: Azure AI Apps and Agents Developer Associate. This guide compares the two outlines and shows where your study time should go.

Does AI-103 replace AI-102?

For the job it measures, yes. Microsoft retired AI-102 and the Azure AI Engineer Associate certification on June 30, 2026, and AI-103 now describes its candidate as “an Azure AI engineer who builds, manages, and deploys agents and AI solutions”.

The credential is different, though. Passing AI-103 earns the Azure AI Apps and Agents Developer Associate, not the old Azure AI Engineer Associate.

That matters on a CV. If you hold the old certification, keep it, and name the new one separately once you pass AI-103.

What changed between AI-102 and AI-103?

The outline moved from six domains to five, and agents moved from a side topic to the centre. In AI-102, agentic work was a separate 5-10% domain. In AI-103 it shares a 30-35% domain with generative AI.

Microsoft Foundry is Microsoft’s Azure platform for deploying models and building AI apps and agents. Both outlines use it, but AI-103 builds almost every objective around it.

AI-102 (retired June 30, 2026)	Weight	AI-103 (skills as of April 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 the domain above)
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 solutions	15-20%	Implement information extraction solutions	10-15%

Planning and managing grew to 25-30%. It now includes monitoring drift and grounding quality, managing quotas and cost, and security through managed identity, private networking and keyless credentials.

Which AI-102 topics are not in the AI-103 outline?

Several build-and-train tasks from AI-102 do not appear in the AI-103 skills list. These include training and publishing custom vision models, creating intents and utterances for language understanding, and building custom question answering knowledge bases.

Custom Document Intelligence models, Speech Synthesis Markup Language, Azure AI Video Indexer, prompt flow and fine-tuning a generative model are also absent. Microsoft notes that related topics may still appear, so treat this as a shift in emphasis.

What is new in AI-103?

The new material is agents, multimodal generation and operations. According to the AI-103 study guide, candidates should be able to:

• build agents that integrate retrieval, function calling and conversation memory;
• implement orchestrated multi-agent solutions with safeguards and approval flow controls;
• generate and edit images and videos from prompts, including inpainting and mask-based edits;
• detect and mitigate indirect prompt injection carried by text embedded in images;
• set up tracing, token analytics, safety signals and latency breakdowns;
• integrate Foundry projects with CI/CD pipelines.

Retrieval-augmented generation (RAG) means grounding a model’s answer in documents you retrieve at query time, usually from a search index. AI-103 tests it in both the generative domain and the information extraction domain.

How long is the AI-103 exam, and how many questions does it have?

Microsoft gives AI-103 120 minutes. The retired AI-102 had 100 minutes, so the sitting is longer.

Microsoft does not publish a question count for AI-103. Its exam-experience page says most exams have 40-60 questions and that associate exams which may contain labs run 120 minutes, with 140 minutes of seat time.

A score of 700 or greater is required to pass. You can retake 24 hours after a first failed attempt, and later retakes have longer waits. Practise the pacing with timed AI-103 practice tests before the day.

Which skills carry the most weight on AI-103?

Generative AI and agentic solutions carry the most, at 30-35%. Together with planning and managing at 25-30%, those two domains make up more than half of the exam.

Vision, text analysis and information extraction share the rest at 10-15% each. Do not skip them: content safety for images and Content Understanding pipelines appear there, and they are easy points if you have built one.

Who is AI-103 for, and who is it not for?

AI-103 is for developers who build AI features with Python on Azure. Microsoft’s profile asks for experience developing apps by using Python and familiarity with generative AI and Azure services.

It suits:

• AI-102 holders who need a current, renewable associate credential;
• application developers moving into agent and RAG work on Microsoft Foundry;
• engineers responsible for monitoring, securing and costing AI workloads in production.

It is a weaker fit for:

• beginners with no Python, who should start with AI-901 (see how AI-901 replaces AI-900);
• back-end developers whose work is containers, databases and messaging rather than models, for whom AI-200 maps more closely.

How should you study for AI-103? A 6-week plan

Study in the order of the weights, and build each piece once in Foundry before you practise questions on it. The plan assumes six to eight hours a week.

1. Week 1: plan and secure. Choose models and services for sample scenarios, configure a deployment, and set up managed identity and keyless access.
2. Week 2: generative apps and RAG. Build a grounded chat app over an Azure AI Search index with vector and hybrid search, then evaluate it for relevance and fabrications.
3. Week 3: agents. Build an agent with tools and memory, add a second agent with orchestration, and add an approval step before a risky action.
4. Week 4: operate. Add tracing and token analytics, configure guardrails and content filters, and wire the project into a CI/CD pipeline.
5. Week 5: vision, text and extraction. Run image generation and editing, multimodal question answering, speech as an agent modality, and Content Understanding analyzers.
6. Week 6: timed practice. Sit full timed tests with a case study, review every miss by domain, and rebuild the weakest task.

The method behind this order is explained in our post on what blueprint-aligned practice tests actually mean.

What happens to your Azure AI Engineer Associate certification?

It stays on your Microsoft Learn transcript but can no longer be renewed. Microsoft’s certification page states that the certification and the renewal assessment are retired.

Microsoft associate certifications normally expire annually and are renewed through a free online assessment. With the renewal assessment gone, AI-103 is the path to a current associate credential in this area.

Should you take AI-103 or wait for the outline to settle?

Take it when you are ready rather than waiting. The current AI-103 study guide lists skills measured as of April 16, 2026, and Microsoft updates English exams first and localised versions about eight weeks later.

Before you book, check the AI-103 study guide for a newer skills date. If you sit in another language, allow for that eight-week lag.

Related reading

• AI-901 replaces AI-900: what changed in 2026
• AI-200 replaces AZ-204: what changed for Azure developers
• Are AI and data certifications worth taking in 2026?
• All CertCrafter articles