Back in March I wrote that a GitHub repository with three real projects will get more attention than any certification badge on a LinkedIn profile. I stand by that. But a few months later I went and sat a certification exam myself, and it is only fair that I come back and tell you what actually happened, including the parts that do not fit neatly into my original argument.
The exam was AB-100, Microsoft's new one for Agentic AI Business Solutions Architect. It only launched this year, which is part of why I wanted to sit it. Most AI certifications I looked at in March were still catching up to what large language models could already do. This one was built around a problem that is genuinely current: how do you architect a solution where multiple AI agents are doing real work inside a business, not just answering questions in a chat window.
Why I Actually Sat It
I had just come out of two hackathons back to back. One of them, MeetingBuddy, was a Copilot Studio agent that turned a meeting transcript into minutes, tasks, and follow up emails. The other, agent-architect, was an eight agent pipeline built on Azure AI Foundry that turned a raw business idea into a developer ready blueprint. Both of them worked. Both of them taught me things the hard way, through broken agent flows and prompt output that would not pass cleanly into the next step.
What I noticed after both projects was that I understood my own corner of the agentic AI world quite well, and almost nothing outside it. I knew Copilot Studio because I had just spent weeks in it. I knew Azure AI Foundry orchestration because I had just built a pipeline on it. But the wider picture, things like the Agent2Agent protocol, the Model Context Protocol, how you actually govern an agent that is allowed to touch real business data, none of that I had picked up just by building. You only run into those pieces if something forces you to look at the whole map.
AB-100 is that whole map. It expects you to know Dynamics 365, Power Platform, Copilot Studio, Azure AI Foundry, and Azure OpenAI, not as separate products, but as one connected surface you are expected to move across depending on what the business actually needs. It also expects you to think about the parts nobody wants to think about when a demo is going well: security, data residency, audit trails, and what happens when an agent is manipulated into doing something it should not.
Studying For It Was Different From Building
Building MeetingBuddy taught me how one agent behaves under pressure, in one specific tool, for one specific job. Studying for AB-100 taught me how to reason about which tool to reach for in the first place, and why. Those are not the same skill, and I had genuinely underrated how separate they are.
I will be honest about the parts I found tedious. A lot of the exam content is about governance, responsible AI practice, and return on investment analysis, and none of that is exciting to read about at ten at night after a full day of work. But it is exactly the kind of knowledge you never pick up from a weekend hackathon, because a hackathon rewards you for shipping something that works, not for thinking about what happens when it is running in production for two years with real customer data flowing through it.
Did Passing It Change Anything
I passed. And here is the honest answer to the question in the title. It did not change my mind that hands on building teaches you more per hour than studying for an exam does. What it changed is my sense of what a certificate is actually good for.
A certificate is not proof that you can build. My two hackathon projects are the proof that I can build. The certificate is proof that I have looked at the whole system once, deliberately, instead of only the corner of it I happened to be working in that month. On its own it would mean very little to me. Sitting next to the two working projects, it means something closer to: this person has built a real thing, and also understands where that thing fits in a much bigger picture.
I do not think most people need a certification to work in AI, and I said so in March, and I still believe it. But I no longer think certifications and portfolios are competing with each other. They answer different questions. One shows you can do the work. The other shows you understand the map the work sits on. I only really needed both once I had already done enough building to notice the gap between them.