Notes from a product manager learning to work with AI
I've been a product manager for about six years, and for most of that time my job looked the same week to week. Write specs. Talk to engineers. Sit through user calls. Argue about prioritization. Update the roadmap. Repeat.
The last eighteen months have been different. Not because the job changed in name — I'm still doing the same things — but because the way I do almost every one of those things has shifted under me. AI tools have crept into my PRD drafts, my user research synthesis, my competitive analysis, my standups, even my Slack replies. Some of those changes have been useful. Some have been hype. A few have been quiet failures I only noticed weeks later.
I've been wanting to write some of this down for a while, partly to clarify my own thinking and partly because there isn't a ton of practical writing on this from PMs working in the trenches. Most of what I read is either breathless ("AI changes everything!") or dismissive ("just glorified autocomplete"). The truth I've found, working in mobile games and now consumer tech, is more interesting and more boring at the same time.
So this is my attempt at notes from the field. Not predictions, not frameworks lifted from a Substack, not anything I'd put in a deck. Just what I've actually tried, what's working, and what's not.
What I want to write about
How I actually use AI in my PM workflow. Specifically what I use it for and — more importantly — what I don't. There are tasks where AI saves me hours every week. There are also tasks where I tried to use it and it cost me hours of cleanup. The line between those isn't where I expected it to be. I'll start with PRDs in the next post.
Evaluating AI features from the inside. I work on consumer products that increasingly have AI features in them. Most of the conversations I have with engineering and design about these features happen at a very different level than the public discourse suggests. Things like latency budgets, fallback behavior, hallucination tolerance, and eval cost dominate. Marketing talks about "AI-powered." We talk about "what does this thing do when the model is wrong, and how often is that?"
User research in the AI era. Synthesis was always the bottleneck for me — turning a stack of interview transcripts into something I could actually use. AI has changed that workflow more than almost any other. But it's also introduced new failure modes I had to learn to spot.
Tools I actually open. Not in a sponsored-roundup kind of way. Just what's on my dock and what I open them for. This list has changed a lot in the last year and I expect it'll keep changing.
What I'm not going to write about
Predictions about AGI. Hot takes on which model is "best." Philosophical takes on whether AI is good or bad for society. Plenty of people writing about that. I'd rather write about whether using Claude to draft my PRDs actually makes them better, and how I'd know.
Who I am, briefly
I'm a Senior Product Manager at Yahoo. Before that I spent three years at Product Madness, the mobile games studio, leading new user experience and the in-game shop on a portfolio of titles that collectively did over $170 million a year. Before that I was at Big Fish Games. I have an MBA from Georgetown. I'm based in Maryland, I'm an avid video gamer (I'm writing this with a Switch in arm's reach), and I've been shipping consumer products for long enough to have strong opinions but not so long that my mental model is calcified.
What that means for these posts: I write from a consumer / mobile background. I haven't shipped enterprise B2B SaaS, I haven't run a big-corp data team, I haven't built infrastructure. So if you came here looking for "how to use AI to refactor your data warehouse," I am not the right person. But if you ship products that real people use on their phones, the things I've learned might be useful.
One more thing
I'm going to be wrong about things in these posts. Not in a hand-wavy false-modesty way — actually wrong. The space is moving too fast for anyone to be consistently right, and the temptation to publish something polished with confident claims is a trap. I'd rather get something useful out and update it than wait until I'm sure.
If you want to follow along, the posts will go up here roughly monthly. You can find me on LinkedIn, Medium, or X if you want to argue with anything I write — I read everything, even the unhinged DMs.
Next post will be about PRDs. Specifically: how I went from "AI is useless for PRDs" to "AI does the first 60% of every PRD I write" in about a quarter, and what that has and hasn't changed about the document.
Thanks for reading.