AI-103
AI-103 Azure AI Apps and Agents Certification Guide
What Microsoft's AI-103 exam on developing AI apps and agents on Azure covers, how the skills are weighted, who it is for, and how to prepare.
By Code Nexus team · 5 min read · Published · Updated · Reviewed by Code Nexus team
At a glance
- Exam code
- AI-103
- Level
- Role-based (Azure AI engineer)
- Questions
- Check the official page
- Time allowed
- Check the official page
- Passing score
- 700 or greater (scaled score)
- Price
- Check the official page
- Valid for
- Check the official page
- Prerequisites
- None stated; experience developing apps with Python is expected
- Languages
- Check the official page
Last checked 2026-09-25. Prices and exam details change, so confirm them on the official page before you book.
If the earlier AI exams ask whether you understand AI, AI-103, Developing AI Apps and Agents on Azure, asks whether you can build and run it. It is the closest exam on the roadmap to daily AI engineering work: choosing models, grounding them in data, building agents, and keeping the whole system safe and observable. This guide covers what it tests and how to prepare. All facts come from Microsoft's official study guide.
What is the AI-103 certification?
The study guide describes the candidate as an Azure AI engineer who builds, manages and deploys agents and AI solutions that take advantage of Microsoft Foundry. You should have experience developing apps with Python and be familiar with the capabilities of general AI, generative AI and Azure services.
Microsoft lists the responsibilities as planning and managing Azure AI solutions, implementing generative AI and agentic solutions, implementing computer vision solutions, implementing text analysis solutions, and implementing information extraction solutions. In the role, you work with business stakeholders, solution architects, data scientists, DevOps engineers and cloud security engineers.
What the exam looks like
- Passing score: 700 or greater, on a scaled score.
- Skills outline: applies as of April 16, 2026.
- Question count, time and price: not stated in the study guide, so check the exam page.
- Preview features: most questions cover generally available features, but the exam may include commonly used preview features.
- Accommodations and extra time: Microsoft lists options, including an extra 30 minutes if the exam is not offered in your language.
A note on currency: this exam changed with the outline dated April 16, 2026. Use current material and avoid guides written for older Azure AI exams.
What AI-103 teaches
The outline has five areas.
| Area | What it covers | Weight |
|---|---|---|
| Plan and manage an Azure AI solution | Choosing models and Foundry services; setting up deployments; quotas, scaling and cost; monitoring and security (managed identity, private networking, keyless credentials); responsible AI including safety filters, evaluators, auditing and agent governance | 25 to 30% |
| Implement generative AI and agentic solutions | Building generative apps, including RAG; building agents with retrieval, function calling and memory; multi-agent solutions; evaluation; tuning, tracing and orchestration | 30 to 35% |
| Implement computer vision solutions | Image and video generation and editing; multimodal understanding; Content Understanding; responsible AI for images, including defence against indirect prompt injection in embedded text | 10 to 15% |
| Implement text analysis solutions | Extracting entities, topics and structured JSON; sentiment and safety detection; translation; speech to text and text to speech | 10 to 15% |
| Implement information extraction solutions | Building retrieval and grounding pipelines with semantic, hybrid and vector search; extracting content from documents with OCR, layout and Content Understanding | 10 to 15% |
Two features of the weighting matter. First, planning, managing and securing carry 25 to 30%, so this is not only a build exam. Second, responsible AI appears inside several areas, so it cannot be studied as a separate box.
Topics that catch people out
- RAG design. Chunking, indexing and choosing between semantic, hybrid and vector search.
- Agent design. Defining roles, tool schemas, memory and approval steps, and knowing when a workflow needs a human.
- Evaluation. Measuring fabrication, relevance, quality and safety, and analysing errors.
- Security. Managed identity and keyless credentials rather than stored keys.
- Indirect prompt injection. Hidden instructions in documents or images that try to steer a model.
Who AI-103 is for
- Developers building AI features on Azure who want to prove it.
- Engineers moving into AI from software, DevOps or data roles.
- Learners who have completed AI-901 and want the developer-level exam.
- Teams that want a shared standard for building AI responsibly.
If you cannot yet build a small application in Python, do PCEP and PCAP first. The exam expects it.
The benefits
- It mirrors the job. The outline reads like a description of daily AI engineering work.
- It builds the operational side. Quotas, monitoring, security and governance are examined, not just prompts.
- It teaches evaluation. Measuring quality and safety is the skill that separates a demo from a product.
- It is current. Agents, multi-agent workflows and Content Understanding are in scope.
Is there demand for the skill?
We could not find a reliable, current, independent figure for demand for AI-103 specifically, so we do not quote one. The reasoning we can support is about the skills: Python is the language of AI tooling, with 57.9% of respondents to the 2025 Stack Overflow Developer Survey using it in the past year, and the responsibilities Microsoft lists for the role, such as RAG, agents and responsible AI, describe how many teams now build AI features. We have not found evidence that employers require AI-103 by name, so treat it as evidence of applied learning, best paired with projects.
How Code Nexus prepares you for AI-103
Eight episodes cover the exam, taught through a support assistant the fictional company is building and improving.
- Episode 41, Code Nexus Assist Gets a Real Brief covers planning: choosing services and regions against a brief, deployments and quotas.
- Episode 42, The Assistant Invented a Refund covers retrieval and grounding.
- Episode 43, Creators Want Visual Superpowers covers vision and media.
- Episode 44, Lost in Translation, Found in Tests covers text and speech.
- Episode 45, Receipts, Screenshots and Voice Notes covers content extraction.
- Episode 46, Give the Assistant Tools covers agents and tool contracts.
- Episode 47, The Agent Did What? covers responsible AI and security, including indirect prompt injection.
- Episode 48, The AI Engineering Review covers all areas together and is the readiness check.
The episodes use a simulated Foundry-style environment that names the real Azure services and concepts the exam uses. It is not your own Azure subscription, so build something real in a free Azure account alongside it. See the curriculum.
Code Nexus prepares you for the exam. Microsoft delivers it and issues the certification.
A study plan
- Read the outline on the study guide and mark each bullet you cannot yet do.
- Build one RAG app from a few documents, and measure whether the right passage is retrieved.
- Build one agent with two tools, one that reads and one that changes something, and add an approval step.
- Write an evaluation: ten questions with expected answers, scored for relevance and grounding.
- Practise security: replace a stored key with managed identity and note what changes.
- Do timed practice and review every miss.
What to do next
After AI-103, the roadmap turns to operating what you build: KCNA for containers and Kubernetes, and AZ-400 for DevOps. See the certification roadmap for the whole route, and why we teach by doing for the reasoning behind the episodes.
Ready to start? Create a free account and open Episode 41.
Follow Code Nexus
Frequently asked questions
- Who is AI-103 for?
- Microsoft describes the candidate as an Azure AI engineer who builds, manages and deploys agents and AI solutions that use Microsoft Foundry. You should have experience developing apps with Python and be familiar with general AI, generative AI and Azure services.
- Which area carries the most marks?
- Implementing generative AI and agentic solutions carries 30 to 35%, the largest share. Planning and managing an Azure AI solution follows at 25 to 30%.
- Does the exam cover RAG and agents?
- Yes. The skills outline names implementing retrieval-augmented generation in an application, building agents that integrate retrieval, function calling and conversation memory, and orchestrating multi-agent solutions.
- What score do I need to pass?
- Microsoft states that a score of 700 or greater is required to pass. It is a scaled score, not 70% of questions correct.
- Does this certification expire?
- Microsoft states that its associate, expert and specialty certifications expire annually and can be renewed by passing a free online assessment on Microsoft Learn. Check the AI-103 certification page for how the renewal applies to it.
Related articles
- AI-901 Azure AI Fundamentals Certification Guide
What the AI-901 Microsoft Azure AI Fundamentals exam covers, how it replaced AI-900, who it suits, and how to prepare, with the official study guide.
- PCEP Certification Guide: Python Entry-Level Exam
What the PCEP-30-02 Python certification covers, who it suits, what the exam is like, and how to prepare, with a link to the official exam page.
- PCAP Certification Guide: Python Associate Exam
What the PCAP-31-03 Python associate certification covers, how it differs from PCEP, what the exam involves, and how to prepare for it.
Mentioned in
- AI Engineer vs ML Engineer vs Data Scientist
How AI engineers, machine learning engineers and data scientists differ in daily work, skills and career path, and how to choose which role fits you.
- AI Guardrails and Safety: What Engineers Must Know
AI guardrails explained: the main risks to LLM applications, how prompt injection works, and practical controls, with OWASP and NIST as references.
- AI-901 Azure AI Fundamentals Certification Guide
What the AI-901 Microsoft Azure AI Fundamentals exam covers, how it replaced AI-900, who it suits, and how to prepare, with the official study guide.
- AIF-C01 AWS AI Practitioner Certification Guide
What the AWS Certified AI Practitioner (AIF-C01) exam covers, its five weighted domains, who it suits, and how to prepare, with the official exam guide.
- KCNA Kubernetes and Cloud Native Associate Guide
What the Linux Foundation's KCNA exam covers, its four weighted domains, cost and format, who it suits, and how to prepare, with the official page.
- RAG Explained: Retrieval-Augmented Generation
Retrieval-augmented generation explained: why models need your data, how retrieval and grounding work, what usually goes wrong and how to measure it.
- The Certification Roadmap: Python to Azure AI
A step-by-step certification roadmap from Python and Linux through cloud, AI and DevOps, with what each exam covers and how to sequence them.
- What Are AI Agents and How Do They Work?
AI agents explained: how they differ from simple workflows, how tools, memory and retrieval fit in, when to use one, and the risks to design around.
Sources
- Study guide for Exam AI-103: Developing AI Apps and Agents on Azure (Microsoft Learn) (accessed 2026-09-25)
- 2025 Stack Overflow Developer Survey: Technology (accessed 2026-09-25)
Code Nexus is not affiliated with Microsoft. Developing AI Apps and Agents on Azure and its badge are trademarks of Microsoft Corporation. Code Nexus prepares you for the exam; Microsoft issues the certification.