Exam guide for cloud switchers · Associate

AI-103 for AWS engineers: Developing AI Apps and Agents on Azure

AI-103 is the Azure AI Apps and Agents Developer Associate exam and the successor in scope to the retired AI-102.

It covers generative AI apps and agents in Microsoft Foundry, RAG with Azure AI Search, vision, speech, text and extraction, along with responsible AI and operations. If you build on Bedrock, most domains map one to one.

Last verified: Oct 27, 2026 against the official AI-103 study guide · Edit on GitHub · Report an error

Exam at a glance

Exam	AI-103: Developing AI Apps and Agents on Azure
Certification	Microsoft Certified – Azure AI Apps and Agents Developer Associate (Associate)
Status	current Generative AI and agentic solutions with Microsoft Foundry using Python. Successor in scope to the retired AI-102.
Skills measured as of	Apr 16, 2026
Passing score	700 or greater (per the study guide)
Closest AWS cert	Generative AI Developer – Professional

Exam prices vary by country and are set on the exam page, so none are quoted here. Microsoft doesn't publish pass rates.

Who it's for

AWS developers building on Bedrock, Bedrock Agents/AgentCore and Knowledge Bases.
Former AI-102 candidates. AI-102 is retired.
Engineers who will own RAG pipelines, agent tools and content safety in production.

Domains and how they map to AWS

Domain (official)	Weight	Compare pages
Plan and manage an Azure AI solution	25–30%	7
Implement generative AI and agentic solutions	30–35%	8
Implement computer vision solutions	10–15%	4
Implement text analysis solutions	10–15%	6
Implement information extraction solutions	10–15%	5

1. Plan and manage an Azure AI solution 25–30%

Official skill areas: Choose the appropriate Foundry services for generative AI and agents · Set up AI solutions in Foundry · Manage, monitor, and secure AI systems · Implement responsible AI across generative AI and agentic systems

From AWS: Choosing models and services, infrastructure and deployment types, quotas and rate limits, monitoring and private networking, plus responsible AI controls. This maps to Bedrock model choice, provisioned throughput, Guardrails and CloudWatch for Bedrock.

Compare pages to study (7):

2. Implement generative AI and agentic solutions 30–35%

Official skill areas: Build generative applications by using Foundry · Build agents by using Foundry · Optimize and operationalize generative AI systems

From AWS: Bedrock Converse, Flows and Agents/AgentCore map to Foundry SDKs, workflows and Foundry Agent Service. Multi-agent orchestration and tool use map to the Microsoft Agent Framework (≈ Strands Agents). Evaluation maps to Foundry evaluations and observability.

Compare pages to study (8):

3. Implement computer vision solutions 10–15%

Official skill areas: Design and implement image- and video-generation solutions · Design and implement multimodal understanding workflows · Implement responsible AI for multimodal content

From AWS: Image and video generation (≈ Nova Canvas/Reel) and multimodal understanding with Content Understanding (≈ Rekognition plus Bedrock Data Automation), with visual content safety.

Compare pages to study (4):

4. Implement text analysis solutions 10–15%

Official skill areas: Apply language model text analysis · Implement speech solutions

From AWS: Comprehend, Translate, Transcribe and Polly map to language-model text analysis, Translator and Azure Speech in Foundry Tools, including speech as an agent modality.

Compare pages to study (6):

5. Implement information extraction solutions 10–15%

Official skill areas: Build retrieval and grounding pipelines · Extract content from documents

From AWS: Bedrock Knowledge Bases and Kendra map to Azure AI Search (indexers, skillsets, hybrid and vector search) and Foundry IQ. Textract and Bedrock Data Automation map to Content Understanding analyzers.

Compare pages to study (5):

Concepts that trip AWS engineers up

Retrieval is a separate service

Bedrock Knowledge Bases bundle chunking, embedding and the vector store. On Azure you usually design the Azure AI Search index, indexers and skillsets yourself, or use Foundry IQ on top. The exam tests index and query configuration such as semantic ranking and hybrid search.

Deployments and quota

Models are deployed into a Foundry resource with a deployment type (for example standard, global or provisioned) and a tokens-per-minute quota. Rate limits and PTUs are design questions, much like Bedrock on-demand vs provisioned throughput.

Content filters are on by default

Azure OpenAI and Foundry model deployments apply default content filtering. Guardrails are configured rather than added. Bedrock Guardrails are opt-in.

Python-first

Microsoft notes that the exam assumes Python for code-based questions.

Closest AWS certification

AWS Certified Generative AI Developer – Professional: The closest AWS role (generative AI developer). The AWS exam is Professional level, AI-103 is Associate. Exam code not confirmed from the official page at build time; shown by name only.

FAQ

What does AI-103 cover, and how is it weighted?

AI-103 (Developing AI Apps and Agents on Azure) has 5 domains: Plan and manage an Azure AI solution (25–30%); Implement generative AI and agentic solutions (30–35%); Implement computer vision solutions (10–15%); Implement text analysis solutions (10–15%); Implement information extraction solutions (10–15%). Weights are from the official Microsoft Learn study guide (skills measured as of Apr 16, 2026).

What is the closest AWS certification to AI-103?

AWS Certified Generative AI Developer – Professional: The closest AWS role (generative AI developer). The AWS exam is Professional level, AI-103 is Associate.

Is AI-103 a current exam?

As of Oct 27, 2026, Microsoft Learn lists AI-103 as current. Generative AI and agentic solutions with Microsoft Foundry using Python. Successor in scope to the retired AI-102.