AI has made me faster at product work. It can draft a prototype, summarize research, clean up a Jira ticket, and surface edge cases in minutes.
It can also frame a problem before I have formed my own view. A polished answer arrives with options, language, and implied tradeoffs. It feels easier to edit than to start from an empty page. If I repeat that process, I get better at selecting model output and worse at building judgment.
The execution gains are real. Behind the Craft by Peter Yang and How I AI by Claire Vo document plenty of them. A "think harder" prompt cannot replace the slow work of deciding what you believe.
the trap
You ask the model to brainstorm. It returns six clean approaches. You tune the prompt and receive a smoother list.
The list anchors the problem. You start from its categories, compare its tradeoffs, and adopt its language. The next decision feels strange without the model because you have practised editing answers instead of producing a point of view.
decision quality
PMs earn trust through decision quality: choosing under uncertainty, noticing the missing variable, and carrying context that does not fit into a dashboard.
Experience builds that judgment. You make a call, watch it fail, update a mental model, and recognise a related pattern months later. Twenty context documents can inform a model, but they cannot give you responsibility for the result.
My best ideas tend to arrive away from a prompt box: on a long walk, during a drive, at the gym, or while cooking. Those settings leave enough silence for an unfinished thought to develop. An instant answer removes that time.
Shared access to the same models also reduces prompt skill as an advantage. Your point of view, domain context, and record of decisions remain harder to copy.
use paper first
Try one week without a paid AI assistant for strategic decisions. Take a pen and work through the uncertainty before you open a model.
I use four prompts:
- Problem: Write the decision in one sentence and name the force making it urgent.
- Principles: List the facts, constraints, and assumptions that bound the choice.
- Possibilities: Generate options without grading them.
- Path: Choose one, explain the tradeoff, and set the first deadline.
I call it the 4Ps framework. It also reminds me to book a table at Pizza 4Ps.
Bring AI back after you have a view. Ask it to find missing evidence, attack an assumption, or simulate a counterargument. Keep authorship of the judgment by explaining why you accepted or rejected its advice.