Azure AI Engineer Career Path: From AI-901 to AI-103
Azure AI Engineers design, build and deploy solutions on Azure's AI services: generative AI and agents, computer vision, text analysis and information extraction. The certification path for the role changed in 2026. This guide covers the exams that are current now.
What an Azure AI Engineer does
Day to day responsibilities
Azure AI Engineers build and maintain systems that process natural language, interpret images and video, generate content, and run autonomous or semi-autonomous agents. A typical week includes planning an AI solution's architecture, deploying and configuring models in Microsoft Foundry, wiring retrieval-augmented generation into an application, and monitoring model performance, drift, and safety events once the solution is live.
You work with solution architects to turn requirements into an implementation, with data scientists on model selection and evaluation, and with DevOps and security engineers on deployment, identity and network configuration. Responsible AI is part of the everyday work: content filters, guardrails and audit trails are configured alongside the features themselves.
The services you'll work with
The current Azure AI Engineer exam is built around Microsoft Foundry, the platform for deploying and consuming large language models, small language models, and multimodal models. Around it sit the Foundry Tools: Azure AI Vision for image and video analysis, Azure AI Language for text analysis and translation, Azure AI Speech for speech-to-text and text-to-speech, Azure AI Search for semantic and vector retrieval, Azure OpenAI for generative models, and Azure AI Document Intelligence and Azure Content Understanding for extracting information from documents, images, audio, and video.
Agent work involves defining agent roles and tool schemas, integrating function-calling and conversation memory, and orchestrating multi-agent workflows, with approval steps and oversight controls where a workflow needs them.
Where this role sits
Most Azure AI Engineer work adds AI to an existing product or internal process. Typical projects are customer-facing chat and search, internal document and knowledge-mining tools, and computer vision or speech features added to an existing application. Solution architects and data scientists work on the same projects, each with a separate remit.
The certification path: AI-901 to AI-103
AI-901: Microsoft Azure AI Fundamentals
AI-901 is for candidates "at the beginning of your career in AI solution development." It asks for conceptual knowledge of AI solutions in Azure, foundational technical skills to work with them, and knowledge of Python coding syntax and programming techniques, along with familiarity with Azure resources.
The exam measures two skill areas: identifying AI concepts and capabilities (40 to 45% of the exam) and implementing AI solutions by using Microsoft Foundry (55 to 60%). A score of 700 or greater is required to pass. AI-901 carries no retirement date at the time of writing.
AI-103: Developing AI Apps and Agents on Azure
AI-103 is now the certification for the Azure AI Engineer role. Microsoft describes the audience as "an Azure AI engineer who builds, manages, and deploys agents and AI solutions that take advantage of Microsoft Foundry," and expects experience developing apps in Python plus familiarity with general AI, generative AI, and Azure services. The exam sits at Intermediate level and carries both the AI Engineer and Developer role tags.
The exam lasts 120 minutes and the pass mark is 700. It is offered in English, Chinese (Simplified), Chinese (Traditional), French, German, Japanese, Korean, Italian, Portuguese (Brazil), and Spanish.
Five domains make up the exam:
| Domain | Weight |
|---|---|
| Implement generative AI and agentic solutions | 30 to 35% |
| Plan and manage an Azure AI solution | 25 to 30% |
| Implement computer vision solutions | 10 to 15% |
| Implement text analysis solutions | 10 to 15% |
| Implement information extraction solutions | 10 to 15% |
What changed from AI-102
AI-102 measured six domains: planning and managing a solution (20 to 25%), generative AI (15 to 20%), agentic solutions (5 to 10%), computer vision (10 to 15%), natural language processing (15 to 20%), and knowledge mining and information extraction (15 to 20%). AI-103 folds generative AI and agentic solutions into one domain weighted at 30 to 35%, against 20 to 30% for the two AI-102 domains combined. Text analysis and information extraction replace the natural language processing and knowledge mining domains, and each drops from 15 to 20% on AI-102 to 10 to 15% on AI-103.
The AI-103 skills list also covers building, evaluating and operating generative and agentic systems from start to finish, including tracing, token analytics and error analysis for deployed agents. For the full domain-by-domain breakdown, see our AI-103 study guide.
Adjacent certifications: AI-200 and AI-300
Neither exam is required after AI-103. Both are separate associate certifications, and each suits a different direction.
AI-200: Azure AI Cloud Developer Associate
AI-200 carries the Developer role tag; AI-103 carries the AI Engineer tag. Microsoft's audience profile describes someone "contributing to all phases of implementing AI solutions on Azure, with an emphasis on back-end services and components," proficient in Azure and third-party SDKs, Azure data management services, messaging and eventing, vector databases, Python, and containerised applications. The exam lasts 120 minutes, the pass mark is 700, and it has four domains: developing containerised solutions (20 to 25%), developing AI solutions with Azure data management services (25 to 30%), connecting to and consuming Azure services (20 to 25%), and securing, monitoring, and troubleshooting Azure solutions (20 to 25%). It's offered in English, Arabic (Saudi Arabia), Chinese (Simplified), Chinese (Traditional), French, German, Indonesian (Indonesia), Italian, Japanese, Korean, Portuguese (Brazil), Russian, and Spanish.
AI-200 suits engineers who want more depth on the back-end and data layer under an AI solution, while AI-103 tests the model and agent layer.
AI-300: Machine Learning Operations Engineer Associate
AI-300 shares the AI Engineer role tag with AI-103. Microsoft's audience profile asks for "subject matter expertise in setting up infrastructure for machine learning operations (MLOps) and generative AI operations (GenAIOps) solutions on Azure," with experience training and deploying models in Azure Machine Learning and deploying, evaluating, monitoring, and optimising generative AI applications and agents in Microsoft Foundry. It also lasts 120 minutes with a pass mark of 700, and is offered in English only. Its five domains are MLOps infrastructure (15 to 20%), machine learning model lifecycle and operations (25 to 30%), GenAIOps infrastructure (20 to 25%), generative AI quality assurance and observability (10 to 15%), and optimising generative AI systems and model performance (10 to 15%).
AI-300 suits an AI Engineer who is moving from building solutions to running them at scale. It concentrates on model lifecycle management, drift detection and production observability.
Skills you'll build
Technical skills
The AI-103 skills measured list is a practical checklist of what to learn: choosing appropriate Foundry services and models for a task, configuring model and agent deployments, implementing retrieval-augmented generation, building agents that integrate function-calling and conversation memory, orchestrating multi-agent workflows, and setting up observability through tracing and token analytics. On the vision and language side, the exam covers image and video generation, multimodal understanding, text analysis and translation, and information extraction from documents, images, audio, and video using Content Understanding and Document Intelligence.
Responsible AI appears in more than one domain: configuring safety filters and content moderation, applying evaluators for fabrication and relevance, and governing agent behaviour with oversight modes and tool-access controls are all named skills, and the computer vision domain adds its own responsible AI section for multimodal content.
Hands-on practice
Read the skills measured first, then build. Before you sit AI-103, deploy a model in Foundry, connect a retrieval-augmented generation pipeline to your own documents, and build one agent with a real tool integration. To keep costs down while you practise, our guide to using Azure's free tier for certification practice explains the setup.
Example AI-103 scenario
Choosing the right Foundry service for a task is one of the core skills in AI-103's Plan and manage domain. The scenario below is typical of that domain.
A financial services firm wants employees to search across ten thousand policy documents and get grounded answers with citations instead of a ranked list of file names. Which capability should the solution use?
- A. Full-text keyword search across the raw documents
- B. Vector search combined with retrieval-augmented generation, so answers are grounded in retrieved passages
- C. A single large language model prompt with as many documents pasted in as fit the context window
- D. Manual tagging of each document by category, with no search layer
B is correct. Vector search finds semantically related passages even when the wording differs from the query, and retrieval-augmented generation grounds the model's answer in those retrieved passages rather than its own unguided output. Option A misses paraphrased questions that share no keywords with the source text. Option C does not scale to ten thousand documents, since most fall outside the context window. Option D removes search entirely and relies on someone tagging every category by hand.
Study strategy
Start with the official study guides for AI-901 and AI-103 on Microsoft Learn, which list every skill measured. Add the Azure documentation pages for whichever Foundry Tools you know least well.
Practice questions show you where the gaps are while there is still time to fix them. AzurePrep has practice questions across current Azure certifications, including AI-901 and AI-103. Read the explanation for every answer, right or wrong, so you learn why one option fits the scenario better than the others.
If you are deciding how much time to set aside, our guide to estimating Azure exam prep time and our spaced repetition guide both apply directly to AI-901 and AI-103 study plans. For the wider Azure certification landscape, see our certification roadmap, and if data science is a closer fit than AI engineering, our Azure Data Scientist career path guide covers that route.
FAQ
Do I need AI-901 before I can take AI-103?
Microsoft does not list AI-901 as a formal prerequisite for AI-103. AI-103 assumes you already have experience developing apps in Python and are familiar with general AI, generative AI, and Azure services, so AI-901 is a useful head start, though Microsoft does not require it.
What happened to AI-900 and AI-102?
Both exams were retired on 30 June 2026. AI-900 was replaced by AI-901, which now leads to Azure AI Fundamentals. The Azure AI Engineer Associate certification and its renewal assessment are also retired; AI-103 covers AI app and agent development on Azure now.
How is AI-103 different from the retired AI-102?
AI-102 measured six domains, including separate weightings for generative AI (15 to 20%) and agentic solutions (5 to 10%). AI-103 measures five domains and combines generative AI and agentic solutions into a single domain weighted at 30 to 35%. AI-103 also groups the old natural language processing and knowledge mining domains under text analysis and information extraction.
What programming and platform skills does AI-103 require?
AI-103 expects experience developing apps in Python and familiarity with Azure services. The skills measured cover building generative applications and agents with Microsoft Foundry, computer vision, text analysis, and information extraction pipelines.
Are AI-200 and AI-300 required after AI-103?
No. AI-200 (Azure AI Cloud Developer Associate) and AI-300 (Machine Learning Operations Engineer Associate) are separate associate certifications you can take independently. AI-300 shares the AI Engineer role tag with AI-103, so it suits engineers moving into MLOps and GenAIOps. AI-200 carries a Developer role tag and concentrates on back-end services.
How often do I need to renew AI-103?
AI-103 is an associate-level certification, and Microsoft's associate, expert and specialty certifications expire annually. You renew by passing a free online assessment on Microsoft Learn.
Start with AI-901 for the foundations, then move to AI-103 when you are ready to build and deploy real solutions. Practise with realistic questions at azureprep.com/exam/ai-103 before you book the exam.