How I'm putting AI to work as a product manager
Managing six people and maintaining custom GPTs turned out to be
Most of my work this year has still revolved around the recruiting platform I helped launch in 2023. The product is bigger now, and so is my role. I'm a Senior Product Manager with six people reporting to me, and most of what's changed about my job traces back to that number.
When it was mostly me and the product, I could stay close to every decision. If something was unclear or messy, I'd dig in and work through it myself. I still want to do that. With six people, though, jumping into every hard problem turns me into the place where decisions go to wait. So a lot of my year has gone into giving people enough background to move without me, and I'm still working out the balance. Sometimes I step in too early. Sometimes I find out I left important context sitting in my own head.
What I didn't expect was to learn a version of the same lesson from AI.
What custom GPTs changed for me
Last year I mostly used ChatGPT as a thought partner. I'd bounce ideas around, draft parts of a PRD, prep for a meeting. Useful, but every conversation started from zero, and the quality depended on how much background I was willing to type into the prompt that day.
Custom GPTs changed that. I started building them almost as soon as they launched. Each one gets a specific job, instructions, and uploaded knowledge about the product, so instead of opening a blank chat I'm working with something that already knows the basics. I spend less time explaining and more time actually testing an idea or improving a document.
Keeping the knowledge current is real work
I'm pretty obsessive about keeping my GPTs up to date. If I'm going to rely on one for product work, I want it working from the latest information I can give it. It's a grind. Product context changes constantly. Decisions happen in meetings, constraints shift, new concerns show up, and none of that flows into a GPT on its own. I update knowledge files and rewrite instructions by hand.
Meeting context is the worst part. So much of what a GPT would need to know gets decided in rooms, and I haven't found a good way to get it in there without doing it manually. I keep doing it anyway, because the answers from a GPT that actually knows the product are so much more useful than the generic ones.
Copilot's @workspace pushed the same idea further for me this year. Asking a question with the whole repository in view means I don't have to carry every relevant detail into the conversation myself. That's the direction I hope all of this goes.
The unexpected part is how much this overlaps with managing people. The prep is the same. Before I hand a problem to a GPT or to someone on my team, I need to be clear on the goal, the constraints, what's already been decided, and where there's still room to explore. When I skip that, I don't get good results from either.
I'm still the one directing every AI interaction. I pick the task, keep the knowledge current, and review what comes back. Most of what I've learned this year, from the team and from the tools, has been about getting context out of my head and into a form someone else can use.