Microsoft retired AI-102 and the Azure AI Engineer Associate title on 30 June 2026. AI-103, Developing AI Apps and Agents on Azure, is the replacement, and it earns a different credential: Microsoft Certified: Azure AI Apps and Agents Developer Associate. Searches for ai-103 went from around 200 a month to nearly 3,000 over the spring, so a lot of people are now trying to work out whether it is worth doing. Short answer: yes, if Azure is your platform. The exam is 120 minutes, proctored through Pearson VUE, pass mark 700, and priced by country (about $165 in the US).
The skills split is planning and managing an Azure AI solution (25 to 30%), generative AI and agentic solutions (30 to 35%), and then computer vision, text analysis and information extraction at 10 to 15% each. The candidate profile says Python, plainly. That alone is an improvement on the old exam, which let you flip between C# and Python and tested SDK trivia in both. What I like is the agent section. You are expected to define agent roles and tool schemas, wire up retrieval and function calling with memory, orchestrate multiple agents, build semi-autonomous workflows with approval controls, and then monitor and error-analyse what the agents did.
The planning section adds quota and cost management, drift and grounding-quality monitoring, managed identity and keyless auth, and CI/CD for Foundry projects. That is a fair description of the boring work that decides whether an agent survives contact with production. The responsible AI objectives are also better than usual. Instead of asking you to recite principles, they ask you to configure safety filters, run safety evaluations, log provenance, and restrict which tools an agent can call. There is even an objective on catching prompt injection hidden in text inside images.
I have reviewed a lot of AI courses and very few mention that attack at all. The weaker parts are inherited. The vision, language and extraction sections are still partly a naming exercise: which Foundry Tool, which Content Understanding mode, which search type. Foundry also renames things at a pace that the Learn content cannot keep up with, so expect to read three names for the same service. And you will need real Azure spend to practise agents properly.
A free trial helps, but multi-agent experiments chew through tokens quickly. The bigger complaint is about AI-102 holders. There is no bridge exam and the old renewal assessment is gone, so anyone who earned AI-102 in early 2026 has a credential that expires with nowhere to go but a full new exam. That is a poor way to treat people who paid for it months ago. I give it 4.2. It is the most current mainstream cloud AI developer exam I have looked at, with good weight on evaluation and safeguards.
It loses points for vendor lock-in, product churn and the clumsy cut-over from AI-102.