The short version
Take AB-620 if you build agents mostly in Microsoft Copilot Studio and connect them to other systems. Take AI-103 if you write Python and build AI apps and agents on Azure with Microsoft Foundry. Both are intermediate, 120-minute associate exams, and neither lists a prerequisite exam. If your job covers both, take the one that matches your daily work first, then the other.
The two exams in one table
| AB-620 | AI-103 | |
|---|---|---|
| Full name | Designing and Building Integrated AI Agent Solutions in Copilot Studio | Developing AI Apps and Agents on Azure |
| Credential | AI Agent Builder Associate | Azure AI Apps and Agents Developer Associate |
| Level (Microsoft) | Intermediate | Intermediate |
| Audience | Professional developer or advanced builder of custom agents | Azure AI engineer building AI apps and agents with Foundry |
| Main platform | Microsoft Copilot Studio, Power Platform | Microsoft Foundry, Azure AI services |
| Code focus | Power Fx, connectors, flows, REST APIs | Python is expected |
| Time | 120 minutes | 120 minutes |
| Passing score | 700 of 1,000 | 700 of 1,000 |
| Delivery | Proctored, Pearson VUE | Proctored, Pearson VUE |
| Languages | 13 | 10 |
| Domains | 3 | 5 |
| Prerequisite exam | None listed | None listed |
| Renewal | Free yearly assessment | Free yearly assessment |
Microsoft does not publish the number of questions for either exam. The general statement is that most exams have 40 to 60.
Domains side by side
AB-620
- Plan and configure agent solutions (30-35%)
- Integrate and extend agents in Copilot Studio (40-45%)
- Test and manage agents (20-25%)
AI-103 (skills measured as of April 16, 2026)
- 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%)
What overlaps
Both exams deal with agents, multi-agent designs, retrieval and grounding, monitoring, and responsible AI. Both mention Microsoft Foundry. AB-620 includes connecting a Foundry agent and using the Foundry model catalog inside Copilot Studio. AI-103 builds agents directly in Foundry. Both include some form of evaluating agent behavior and operating the solution after it is built.
If you study one, you will meet the vocabulary of the other: retrieval-augmented generation, tools, orchestration, safety filters, and evaluation.
What differs
- Where you work. AB-620 lives in the Copilot Studio maker experience and the Power Platform around it: topics, agent flows, connectors, environments, solutions. AI-103 lives in Azure: model deployments, quotas, networking, SDKs, and AI services.
- How much code. AI-103 expects Python experience. AB-620 expects Power Fx, Dataverse, REST integration, and prompt engineering, but its main surface is a low-code studio.
- What else is tested. AI-103 spends a large share on vision, text analysis, speech, and information extraction. AB-620 has none of those as separate domains. It spends more on application lifecycle management and Power Platform data policies.
- How focused it is. AB-620 puts 40-45% on one area: integrating and extending. AI-103 spreads weight across five areas, so you cannot skip any.
How to choose
Ask yourself three questions.
- What do I build with today? Copilot Studio and Power Platform point to AB-620. Python and Azure point to AI-103.
- Who do I work with? Business teams, Power Platform admins, and Copilot admins sit around AB-620. Data scientists, DevOps, and cloud security engineers sit around AI-103, according to Microsoft’s audience profiles.
- What does my next job or client need? Match the credential to the platform in the job description.
If you cannot decide, compare the skills lists with the self-check on each prerequisites page: AB-620 and AI-103.
Taking both
Taking both is reasonable if you design agents across the low-code and code sides of Microsoft’s stack. There is no required order. A common-sense order is to start with the one closer to your current work, because the extra familiarity shortens preparation. Expect some savings: the shared agent and grounding concepts carry over, but the platform details do not. Do not skip the second exam’s own study guide.
Frequently asked questions
Is one harder than the other?
It depends on your background. Python developers often find AB-620’s studio-based items unfamiliar. Power Platform builders often find AI-103’s code and Azure items unfamiliar. See is AB-620 hard and is AI-103 hard.
Do I need AI-103 before AB-620, or the other way around?
No. Microsoft lists no prerequisite exam for either one.
Which has the longer language list?
AB-620 lists 13 languages and AI-103 lists 10, according to Microsoft’s certification pages.
More guides for AB-620
- AB-620 Exam Format, Scoring, and Retakes
- AB-620 Study Plan: 4-Week and 8-Week
- AB-620 Domains Explained With Weightings
- AB-620 Sample Questions With Answers
- AB-620 Renewal and Retakes Explained
- Is AB-620 Worth It? An Honest Look
- AB-620 Cheat Sheet: Key Facts to Review
- Agentforce Specialist vs AB-620: Which First?
- AB-100 vs AB-620: Which Exam First?
Next step
Read both overviews, AB-620 and AI-103, then try the AB-620 practice exams or the AI-103 practice exams. Both are original practice questions.