Everyone keeps asking me how product work is changing with AI. I've always believed that to teach it well, I have to walk the walk. So that's what we did. We scoped out a project to use AI as a team at Product Institute, for two reasons. The first: we had to. Product Institute used to be one of the first results when you searched for product management courses. Then the business took off. We were signing clients, updating our courses, launching new ones, growing the team. But while we focused on upgrading our core product, the marketing site just waited. And to be fair, that was because our SEO was good. We were getting so much inbound. There was no reason to fix it. Until there was. Our SEO started to slide. We went from the top of the results to buried pages back. And it happened faster than I expected, especially once people started using AI to search instead of Google. We're a small team. A full rebuild was a big lift. The last time we redesigned the site, it took forever, and we wanted to cut down on that time. The second reason: I wanted to learn. Moving faster was part of it. But I wanted two things more than speed. I wanted us to learn how to actually work as a team with AI in the loop. And I wanted to see for myself how product work changes when AI moves inside the company, and how much of it doesn't change at all. Because I can't help other leaders through this if I haven't lived it. The first real decision was buy or build. We looked hard at a content management system versus building the front end ourselves in Claude Code. We chose to build. We wanted our own front end, not a site boxed in by someone else's platform. And with Claude Code, we could now edit it far faster than we ever could before. By the end of this experiment, all four of us were pushing code to Git. Two of us for the first time ever. I'm going to share what we learned over the next few posts. The wins, the messy parts, and the lessons that turned out to have nothing to do with AI at all. The first one starts with the mistake almost everyone makes on day one. Follow me so you don't miss a post. New site launches next week. We're excited it's finally here. Any guesses on what our major lessons learned are? Tell me in the comments!
I'm guessing that your output is only as good as your human inputs into Claude, and human expertise, experience, and decision-making are still critical to shaping the platform to ensure success.
You forgot the Prd and atarted prompting features to build. Use AI to build the PRD and the little details you might forget are added in. Then tell it to build using the PRD. Next, the first few iterations are good. Now you have a small change. It completely changed the look or took a different approach and you have to keep telling it to revert. Then you start arguing with it to ONLY do this one small change and start to miss being able to tell the developer instead. Haha. Have fun!
The "we wanted our own front end, not a site boxed in by someone else's platform" instinct is the one I'd protect. Building with AI in the loop, the leverage isn't the speed — it's owning the surface and the source of truth the model works from, so you can change direction without asking permission. My guess at your big lesson: the real shift wasn't "AI writes code faster," it was that product decisions moved upstream. Someone still has to define what "correct" looks like before AI can move fast against it. Speed is downstream of clarity.
As a product leader, this is the part I care about most. AI can put code in more hands, which changes who can participate in building. But once that happens, the questions get more important, not less: Who owns quality? Who makes the call? What gets shipped? What technical debt are we quietly creating while everyone is excited about speed?
The faster AI makes it to build, the more valuable product discovery becomes. Speed only matters if teams have a shared understanding of the problem they're trying to solve.
Buy or build being the first real decision says a lot. The tool got cheaper, but the judgment about what you actually wanted did not. The teams that win with AI in the loop are usually the ones who already knew how to work as a team without it.
AI is a catalyst that speeds up processes. Having good team communication, process, and orientation around the problem to solve becomes more important when you're moving faster.
In no particular order. 1. Get the design system solid. 2. Know the spec and what you are trying to do, the value you are creating. 3. Get your hands on and play.
tried something similar with our internal dashboards last month. what was the day-one mistake you're teasing?
Nailing the design system up front saves a lot of head and heartache later when iterating?