Notes on using AI in product work
Generative AI did not only change what product managers are responsible for, it fundamentally changed how much they can own and how much judgment the role now requires.
The biggest shift for me is that AI expanded what I can own end to end.
Before, my role was mainly bound to framing problems, prioritization, and alignment. Execution across design, engineering, and research mostly happened in coordination with others.
Today, those boundaries are looser. I still own the core product questions, but I also prototype, explore technical directions, and test ideas directly. AI did not just make me faster. It changed where I spend my time.
Research became continuous
The clearest change for me is in research.
Understanding markets, industry dynamics, and incentives used to take days of manual work. Now it happens in minutes. I still decide what questions matter and how to interpret signals, but the mechanical work is largely gone.
Research is no longer a phase. It became continuous and mostly automated.
Solution exploration shifted
The second major shift for me is in solution exploration.
I use Figma less and write more. I prototype directly in code using tools like Cursor and Claude Code to explore directions and options early.
This is powerful but also risky. When building is cheap, volume increases, not necessarily quality. Judgment and taste become the constraint.
What did not change
One area has not changed much for me. Qualitative research.
Talking to real users remains irreplaceable. If anything, it has become more important, because AI increases the risk of making fast decisions without real understanding.
What I hope changes for the PM profession
Fewer PMs will be mainly focused on orchestration. The role should become more selective, not more diluted, as it has been in recent years.
There should be more focus on strategy, judgment, and long-term direction. When information and execution are cheap, decision quality becomes the main constraint.
This raises the bar for accountability. AI removes friction, but it does not remove responsibility. Someone still has to decide what matters, what to ignore, and what to commit to. Those decisions compound over time.
I also expect fewer PMs per product. Not because the work disappears, but because leverage per PM increases. One strong PM, supported by AI, can own more surface area than before.
PMs will need to be more generalist. Not by doing everything, but by being literate across disciplines. Understanding incentives, user psychology, and financial mechanics matters more when AI amplifies both good and bad decisions.
In that world, activity is no longer a good proxy for impact. The profession matures when PMs are evaluated on decision quality and long-term outcomes, not throughput.