Using AI to Learn AI

One of my biggest learnings after almost a year of using AI every day: you should be using AI to learn AI.

Last fall our team did exactly that and our work changed the direction of our Windows 365 management experiences in Intune.

It started with research. Abena Edugyan and our user research team had been running benchmarks across our core Intune experiences. Task success numbers were poor. That led us to go deeper, and what we found confirmed we had a real problem.

We had two years of conversations, studies, and feedback sessions sitting across different places. We used AI to synthesize all of it. It didn't replace the judgment of the researchers who ran the work. It helped us look across everything at once. That's when we made a human decision: this problem is real, and we need to solve it.

Once we committed, we made another call. Instead of spending weeks in Figma building static sketches, we would prototype the actual experience. Something we could send as a link. Something people could feel, not just look at. The goal wasn't to ship code. It was to learn fast enough to influence the people who would.

In about three and a half weeks, through trial and error with AI alongside us: I set up a GitHub repo for the first time in years and made a lot of mistakes getting there (don’t edit yourself out of a YAML access file!). I learned Azure hosting. We locked the prototype down so only Microsoft employees could access it, no engineer required. We iterated using Figma Make, GitHub Copilot, Claude and Codex CLI and eventually Github Copilot in CLI. We built against the Intune design system and partnered with the security org's design team who owned it. The whole team learned pull requests and shipped to one shared codebase.

The speed mattered, but it wasn't the point. We brought the prototype to our product partners and because they could use it, they could feel the problem. That prototype influenced the product direction. The work coming into the product now traces back to what that research made visible.

One additional learning: sit in the room with your team and figure it out together. AI is just an empty text box. Watching someone else use it closes the gap faster than anything else. Big shoutout to March Rogers for building that culture across our team.

That's what "using AI to learn AI" means to me. You don't have to understand the whole system before you begin. You use the work to build the understanding.