CertifyDeckAI-103 is Microsoft's associate certification for people who build generative AI apps and agents on Azure with Python and Microsoft Foundry. It takes over the Azure AI Engineer path that AI-102 used to represent and shifts the focus: fewer standalone cognitive APIs, more RAG, agents, evaluation and operations. This guide answers, outline in hand, who it pays off for, what it unlocks and when you should prepare something else.
What AI-103 actually proves
The official profile is an Azure AI engineer who builds, manages and deploys agents and AI solutions with Microsoft Foundry, with experience developing in Python. The exam splits its weight between planning and managing the solution (25–30%), implementing generative AI and agents (30–35%), vision (10–15%), text analysis and speech (10–15%) and information extraction (10–15%).
In real work: choosing models and services, building RAG with Azure AI Search, building agents with memory and tools, orchestrating several agents, evaluating fabrications and safety, configuring managed identity and private networking, and leaving traces and cost metrics behind. It certifies building and operating, not concepts. The AI-103 outline breakdown splits it into hours per block.
Who it pays off for
Backend or full-stack developers who already use Azure and want to show they can take a model into an application safely. Data or ML engineers moving from training models to integrating foundation models. Consultants at Microsoft partners, where certifications count towards partner programme designations. And anyone who had AI-102 on their plan: AI-103 is the current version of that path.
It also pays off as a stepping stone. AB-100, the expert certification for agentic AI solutions, requires an associate certification first, and AI-103 is one of those it accepts, along with AB-620, AI-300 and the Power Platform developer path. If architecture is your goal, this is the most technical of the four.
Who it is not for (and what to prepare instead)
If you do not code, AI-103 is an uphill climb: the outline talks about SDKs, function calling, CI/CD pipelines and deployments. To build agents with low-code tools, AB-620 certifies Copilot Studio. For Copilot and agent fundamentals from the admin side, AB-900. For advanced everyday use of Copilot, AB-730.
It is not the best first certification either if you have never touched Azure: you will lack the identity, networking and resource basics the exam assumes. In that case a few weeks of practice with a free subscription, or a fundamentals certification, saves time later. And if your job is training, versioning and deploying your own models, AI-300 is a closer match.
Cost, effort and a realistic return
The direct cost is that of a Microsoft role-based exam (USD 165 as a reference; the price varies by country and is confirmed when you book). Typical effort for someone with an Azure and Python background is six to eight weeks of evenings; the eight-week plan spells it out. Renewal is annual and free, through an unproctored online assessment.
On salary: a certification does not set a salary and any specific figure would be made up. What it does is give verifiable proof of a skill set that is in high demand right now and has few certified candidates, because the exam is recent. Measure your starting point with the AI-103 practice questions, free to start, before you decide.
Frequently asked questions
Does AI-103 replace AI-102?
AI-103 is the evolution of the Azure AI Engineer path towards generative AI and agents with Microsoft Foundry. If you are starting now, prepare AI-103.
Do I need to know how to code for AI-103?
Yes. Microsoft states experience developing apps with Python and familiarity with generative AI and Azure services.
Does AI-103 count as a prerequisite for AB-100?
Yes, it is one of the associate certifications Microsoft accepts as a prerequisite for the AB-100 expert certification.
How often does AI-103 expire?
Every year. You renew it for free by passing an online assessment on Microsoft Learn before the expiry date.