Retired exam
AI-102 was retired on June 30, 2026
Microsoft retired AI-102 and the Azure AI Engineer Associate certification on June 30, 2026. Microsoft identifies AI-103 and the Azure AI Apps and Agents Developer Associate certification as the replacement path for new candidates.
Prepare for AI-103: Developing AI Apps and Agents on AzureHistorical AI-102 Scope
AI-102 was Microsoft's Azure AI Engineer Associate exam until its retirement on June 30, 2026. Its final blueprint covered service planning, generative AI, agents, computer vision, natural language processing, speech, knowledge mining, and information extraction. The historical outline remains useful for understanding the predecessor credential and for identifying transferable skills, but it is no longer a schedulable exam or the correct blueprint for new candidates.
Who AI-102 Was For
Nobody should begin a new certification plan around AI-102. This historical page is for people who previously studied or earned the retired credential, training teams updating older material, and candidates deciding how much prior Azure AI preparation transfers to AI-103.
Skills You Should Be Ready to Demonstrate
- Carry forward Foundry service-selection, deployment, security, and monitoring skills
- Reuse generative AI, retrieval-augmented generation, evaluation, and orchestration experience
- Reframe agent skills around tools, memory, safeguards, observability, and multi-agent workflows
- Update vision preparation for generation, editing, multimodal understanding, and visual safety
- Map language and speech experience to AI-103 text-analysis and agent-modality objectives
- Map Azure AI Search and extraction experience to current retrieval and grounding pipelines
How to Reuse Your Preparation
Do not continue an AI-102 checklist unchanged. First mark every completed AI-102 objective as transferable, changed, or absent after comparing it with the dated AI-103 guide. Keep hands-on work in Foundry, generative AI, agents, vision, language, search, and information extraction, but rebuild the plan around AI-103's heavier combined generative-and-agentic domain, explicit Python expectation, multimodal generation and editing, agent governance, and retrieval pipelines. Retest only against the current AI-103 blueprint.
What changed from AI-102 to AI-103
| Historical AI-102 | Current AI-103 | Practical impact |
|---|
Plan and manage an Azure AI solution
20-25% | Plan and manage an Azure AI solution
25-30% | The current blueprint adds explicit infrastructure design, quotas, drift, grounding quality, private networking, keyless credentials, audit provenance, and agent oversight. |
Generative AI plus agentic solutions
20-30% combined | Generative AI and agentic solutions
30-35% | AI-103 combines and increases this area, emphasizing tool schemas, memory, guarded autonomy, multi-agent orchestration, error analysis, token analytics, and hybrid model-and-rules systems. |
Computer vision solutions
10-15% | Computer vision solutions
10-15% | The weight is unchanged, but the focus moves toward image and video generation, editing, multimodal understanding, accessibility output, and visual safety controls. |
Natural language processing solutions
15-20% | Text analysis solutions
10-15% | Text and speech remain, while AI-103 frames them through generative prompting, structured output, agent modalities, sensitive-content handling, and domain-specific extraction. |
Knowledge mining and information extraction
15-20% | Information extraction solutions
10-15% | AI-103 centers this area on multimodal ingestion, semantic and vector grounding, RAG pipelines, Content Understanding, and clean representations for agents. |
Migration checklist
- Stop using AI-102 as a schedulable target; use the active AI-103 certification page for booking and current exam details.
- Retain hands-on Foundry, RAG, agent, vision, speech, search, and extraction work, but remap it to the AI-103 objectives.
- Prioritize the combined generative AI and agentic domain, which carries 30-35% of AI-103.
- Add AI-103-specific practice in Python, multimodal generation and editing, agent governance, observability, and retrieval pipelines.
- Recheck the dated AI-103 study guide before booking because Microsoft updates role-based exams periodically.
Open the official AI-103 study guide
Historical Domain Guide
Plan and manage an Azure AI solution: Historical Scope
Planning and management provide the decision framework for every implementation domain. Study service selection, resource creation, authentication, responsible AI, cost management, monitoring, and deployment together. You should be able to move from a stated requirement to an appropriate service and an operable, secured solution design.
- Compare Microsoft Foundry and Azure AI services by workload and operational requirement
- Practice secure resource creation, authentication, monitoring, and cost control
- Review responsible AI capabilities and the controls used throughout a solution lifecycle
Implement generative AI solutions: Historical Scope
Generative AI preparation should cover more than calling a model endpoint. Work through model selection and deployment, prompt construction, retrieval-augmented generation, evaluation, content filtering, and application integration. Pay attention to how grounding, safety, latency, and cost requirements change the design.
- Deploy and consume generative models through supported SDK and REST patterns
- Build grounded solutions with prompts, data sources, and retrieval
- Evaluate outputs and apply responsible AI and content-safety controls
Implement an agentic solution: Historical Scope
The agentic-solutions domain focuses on systems that plan or coordinate work through instructions, knowledge, tools, and multi-step execution. Practice creating and testing agents, connecting approved tools and data, managing state, and observing behavior so failures can be diagnosed instead of hidden behind a conversational interface.
- Configure agent instructions, knowledge, tools, and orchestration
- Test agent behavior against expected tasks and safety boundaries
- Monitor, troubleshoot, and improve an agentic solution
Implement computer vision solutions: Historical Scope
Computer vision work includes choosing the appropriate image or video capability and integrating its results into an application. Prepare by implementing common analysis workflows, handling returned structures, and considering confidence, privacy, and responsible-use requirements rather than memorizing service names alone.
- Select image and video services for the stated analysis requirement
- Implement analysis through an SDK or REST API and interpret the response
- Apply confidence, privacy, and responsible-use considerations to vision output
Implement natural language processing solutions: Historical Scope
Natural language processing covers text analysis as well as speech and translation scenarios. Build small applications that submit content, process service responses, and handle language or audio constraints. Compare available capabilities so you can justify the service and configuration used for a particular workload.
- Implement text analysis and language-understanding capabilities
- Build speech recognition, synthesis, and translation workflows
- Handle language, audio, and application-integration constraints
Implement knowledge mining and information extraction solutions: Historical Scope
Knowledge mining and information extraction combine ingestion, enrichment, indexing, and retrieval. Practice designing an Azure AI Search solution, selecting enrichment and extraction capabilities, and validating what enters the index. Include document, image, audio, and video processing where the published objectives call for them.
- Design indexes, indexers, skillsets, and retrieval behavior in Azure AI Search
- Extract structured information with Document Intelligence and related services
- Process and ingest multimodal content while checking extraction quality
Sources and Verification
Verified 2026-08-26
How this page was made
Cert Atlas compared Microsoft's final AI-102 study guide with the current AI-103 study guide at the domain and objective level, then checked Microsoft's retirement and replacement announcement and the active AI-103 certification page. OpenAI Codex assisted with extraction, normalization, comparison, and drafting. Claims on this page were reviewed against those linked official sources; no exam questions, answers, choices, or explanations were used.
Historical AI-102 Domains
1.0 Plan and manage an Azure AI solution
20-25%
- 1.1Select the appropriate Microsoft Foundry Services
- 1.2Select the appropriate service for a generative AI solution
- 1.3Select the appropriate service for a computer vision solution
- 1.4Select the appropriate service for a natural language processing solution
- 1.5Select the appropriate service for a speech solution
- 1.6Select the appropriate service for an information extraction solution
- 1.7Select the appropriate service for a knowledge mining solution
- 1.8Plan, create and deploy a Microsoft Foundry Service
- 1.9Plan for a solution that meets Responsible AI principles
- 1.10Create an Azure AI resource
- 1.11Choose the appropriate AI models for your solution
- 1.12Deploy AI models using the appropriate deployment options
- 1.13Install and utilize the appropriate SDKs and APIs
- 1.14Determine a default endpoint for a service
- 1.15Integrate Microsoft Foundry Services into a continuous integration and continuous delivery (CI/CD) pipeline
- 1.16Plan and implement a container deployment
- 1.17Manage, monitor, and secure a Microsoft Foundry Service
- 1.18Monitor an Azure AI resource
- 1.19Manage costs for Microsoft Foundry Services
- 1.20Manage and protect account keys
- 1.21Manage authentication for a Microsoft Foundry Service resource
2.0 Implement generative AI solutions
15-20%
- 2.1Build generative AI solutions with Microsoft Foundry
- 2.2Plan and prepare for a generative AI solution
- 2.3Deploy a hub, project, and necessary resources with Microsoft Foundry
- 2.4Deploy the appropriate generative AI model for your use case
- 2.5Implement a prompt flow solution
- 2.6Implement a RAG pattern by grounding a model in your data
- 2.7Evaluate models and flows
- 2.8Integrate your project into an application with Microsoft Foundry SDK
- 2.9Utilize prompt templates in your generative AI solution
- 2.10Use Azure OpenAI in Foundry Models to generate content
- 2.11Provision an Azure OpenAI in Foundry Models resource
- 2.12Select and deploy an Azure OpenAI model
- 2.13Submit prompts to generate code and natural language responses
- 2.14Use the DALL-E model to generate images
- 2.15Integrate Azure OpenAI into your own application
- 2.16Use large multimodal models in Azure OpenAI
- 2.17Optimize and operationalize a generative AI solution
- 2.18Configure parameters to control generative behavior
- 2.19Configure model monitoring and diagnostic settings, including performance and resource consumption
- 2.20Optimize and manage resources for deployment, including scalability and foundational model updates
- 2.21Enable tracing and collect feedback
- 2.22Implement model reflection
- 2.23Deploy containers for use on local and edge devices
- 2.24Implement orchestration of multiple generative AI models
- 2.25Apply prompt engineering techniques to improve responses
- 2.26Fine-tune a generative model
3.0 Implement an agentic solution
5-10%
- 3.1Create custom agents
- 3.2Understand the role and use cases of an agent
- 3.3Configure the necessary resources to build an agent
- 3.4Create an agent with the Microsoft Foundry Agent Service
- 3.5Implement complex agents with Microsoft Agent Framework
- 3.6Implement complex workflows including orchestration for a multi-agent solution, multiple users, and autonomous capabilities
- 3.7Test, optimize and deploy an agent
4.0 Implement computer vision solutions
10-15%
- 4.1Analyze images
- 4.2Select visual features to meet image processing requirements
- 4.3Detect objects in images and generate image tags
- 4.4Include image analysis features in an image processing request
- 4.5Interpret image processing responses
- 4.6Extract text from images using Azure Vision in Foundry Tools
- 4.7Convert handwritten text using Azure Vision in Foundry Tools
- 4.8Implement custom vision models
- 4.9Choose between image classification and object detection models
- 4.10Label images
- 4.11Train a custom image model, including image classification and object detection
- 4.12Evaluate custom vision model metrics
- 4.13Publish a custom vision model
- 4.14Consume a custom vision model
- 4.15Build a custom vision model code first
- 4.16Analyze videos
- 4.17Use Azure AI Video Indexer to extract insights from a video or live stream
- 4.18Use Azure Vision in Foundry Tools Spatial Analysis to detect presence and movement of people in video
5.0 Implement natural language processing solutions
15-20%
- 5.1Analyze and translate text
- 5.2Extract key phrases and entities
- 5.3Determine sentiment of text
- 5.4Detect the language used in text
- 5.5Detect personally identifiable information (PII) in text
- 5.6Translate text and documents by using the Azure Translator in Foundry Tools service
- 5.7Process and translate speech
- 5.8Integrate generative AI speaking capabilities in an application
- 5.9Implement text-to-speech and speech-to-text using Azure Speech in Foundry Tools
- 5.10Improve text-to-speech by using Speech Synthesis Markup Language (SSML)
- 5.11Implement custom speech solutions with Azure Speech in Foundry Tools
- 5.12Implement intent and keyword recognition with Azure Speech in Foundry Tools
- 5.13Translate speech-to-speech and speech-to-text by using the Azure Speech in Foundry Tools service
- 5.14Implement custom language models
- 5.15Create intents, entities, and add utterances
- 5.16Train, evaluate, deploy, and test a language understanding model
- 5.17Optimize, backup, and recover language understanding model
- 5.18Consume a language model from a client application
- 5.19Create a custom question answering project
- 5.20Add question-and-answer pairs and import sources for question answering
- 5.21Train, test, and publish a knowledge base
- 5.22Create a multi-turn conversation
- 5.23Add alternate phrasing and chit-chat to a knowledge base
- 5.24Export a knowledge base
- 5.25Create a multi-language question answering solution
6.0 Implement knowledge mining and information extraction solutions
15-20%
- 6.1Implement an Azure AI Search solution
- 6.2Provision an Azure AI Search resource, create an index, and define a skillset
- 6.3Create data sources and indexers
- 6.4Implement custom skills and include them in a skillset
- 6.5Create and run an indexer
- 6.6Query an index, including syntax, sorting, filtering, and wildcards
- 6.7Manage Knowledge Store projections, including file, object, and table projections
- 6.8Implement semantic and vector store solutions
- 6.9Implement an Azure Document Intelligence in Foundry Tools solution
- 6.10Provision a Document Intelligence resource
- 6.11Use prebuilt models to extract data from documents
- 6.12Implement a custom document intelligence model
- 6.13Train, test, and publish a custom document intelligence model
- 6.14Create a composed document intelligence model
- 6.15Extract information with Azure Content Understanding in Foundry Tools
- 6.16Create an OCR pipeline to extract text from images and documents
- 6.17Summarize, classify, and detect attributes of documents
- 6.18Extract entities, tables, and images from documents
- 6.19Process and ingest documents, images, videos, and audio with Azure Content Understanding in Foundry Tools