Chapter 2 The traits that survive AI
AI will write your first draft. It will query your database. It will generate a passable Figma mock and a prioritized backlog and a competitive analysis by Tuesday morning.
It will not tell you you’re wrong. Not in the way that matters.
What’s left for you is not a set of skills. Skills got cheaper. What’s left is a set of traits — dispositions that were always the real job and are now, with the scaffolding stripped away, impossible to hide behind.
2.1 1. Judgment
Anyone can generate twenty options. Judgment picks one and kills the other nineteen.
AI expands the solution space. Your job is to collapse it. That requires taste — which comes from reps, from failure, from arguing with people who are smarter than you in specific ways and learning what made them right. Taste is not a prompt.
The risk AI introduced here is not that it replaces judgment. It’s that it validates any judgment you bring to it. Ask the model to defend option seven and it will produce a coherent case. Ask it to defend option twelve and it will do the same. The model is not calibrated to your users, your org’s debt, or what failed last quarter. You are.
Test: Give two PMs the same prompt, the same data, the same context. If their recommendations are identical, one of them isn’t doing the thinking.
By context:
Startup. Judgment is the company’s primary asset and yours is unproven. Ship fast enough to test it. You will be wrong often. The goal is to be wrong in ways that teach you something.
Scale-up. You now have enough data to rationalize almost anything. Bad judgment hiding behind good charts is the dominant failure mode. The question isn’t “what does the data say” — it’s “what does the data say that I haven’t already decided to hear.”
Mega-corp. Judgment operates in a system designed to diffuse it. Committees, review processes, legal holds. Your job is to carry a point of view through that system without either losing it or becoming so attached to it that you miss when it’s wrong.
2.2 2. Systems thinking
Features are easy. Systems are hard.
AI is good at the first. It is mediocre at the second. It doesn’t know your org’s scar tissue. It doesn’t know that Sales promised X to an enterprise customer, that the infra team is three weeks from a major migration, that the data model you’re about to build against is being deprecated in Q2.
Systems thinking means understanding what happens after the feature ships. What behavior changes? What new support tickets arrive? What gets slower? What gets stressed at scale? What is the second-order effect of this working too well?
Test: Can you draw the loop? Trace what happens after the user clicks. Then trace what happens at the end of the quarter. Then trace what happens if this succeeds beyond your projections.
By context:
Startup. The system is small enough to hold in your head, which makes it tempting to skip the drawing. Don’t. Externalize the loop. It will look different on paper than it does in your head, and the difference is usually where the problem lives.
Scale-up. The system is now too big to hold in your head, and parts of it are owned by people who don’t report to your manager’s manager. Systems thinking here means mapping dependencies before they become blockers — not after.
Mega-corp. The system includes regulatory constraints, platform policies, and commitments made to partners you’ve never met. Systems thinking at this scale requires institutional memory AI cannot replicate. The people who have it are worth more than their titles suggest.
2.4 4. Narrative gravity
Teams don’t follow the best-written PRD. They follow the story they can retell at 11pm when something breaks.
AI gives you words. Unlimited, grammatically correct, professionally organized words. It cannot give you conviction. It does not know why this product for these users in this moment is worth the cost. You have to know that, and you have to be able to make others feel it — not through rhetoric but through the clarity of your reasoning.
Narrative gravity is the force that pulls a team toward a shared understanding of what they’re building and why. It’s the difference between a team that executes because they were told to and a team that executes because they believe it.
Test: Explain the product to your most skeptical non-technical friend in two minutes. No jargon. No “we’re building a platform.” Do they ask a follow-up question, or do they change the subject?
By context:
Startup. The narrative is existential. Why does this company exist? Why now? Why us? If you can’t answer that in two sentences, you’re building without a foundation.
Scale-up. The narrative is strategic. Why this feature instead of the ten other things we could be doing? The story needs to survive a quarterly planning meeting where everyone has competing priorities and your feature is someone else’s deprioritization.
Mega-corp. The narrative is political. Why should the org absorb the cost — the eng investment, the legal review, the policy risk — of shipping this? The story must reach people who will never be in a room with you and must survive being retold by someone who barely understood it.
2.5 What doesn’t matter as much anymore
This is as important as the list above.
Typing speed. Your throughput is not the bottleneck.
Formatting docs. AI formats. Spend zero time on this.
Being the only person who knows SQL. Everyone has an LLM-to-SQL tool now. The edge was never the SQL. It was knowing which question was worth asking.
Knowing every framework. Jobs-to-be-done, RICE, HEART, OKRs — these were proxies for judgment. AI broke the proxies. Now we test the judgment itself.
Producing the most artifacts. Volume of output is the thing most legible on a performance review and the thing least correlated with actual product impact. AI inflated output. Impact got harder to fake.
2.6 Try this
Exercise: Take a product decision from the last two weeks — yours, your team’s, or someone else’s.
Run it through an LLM. Ask for three alternatives and the risks of each.
Now answer: Which would you ship? Why? What would change your mind before launch?
If you answered “whatever the model recommended,” you didn’t do the exercise. If you had a view before you saw the model’s output, and the model didn’t change it — ask yourself whether that’s confidence or brittleness. The difference matters.
Chapter 3 goes deeper into how the decisions actually happen — the mental moves, the sequencing, the calibrated conviction that separates a PM from a well-organized note-taker.