The PM skills that matter more in the age of AI

AI can produce a convincing answer or research report in seconds. But speed and polish don't tell us whether it solves a real problem—or whether the output is complete, accurate, and trustworthy.

As I build hands-on experience with AI products, I keep returning to questions that have shaped my 14 years in product management.

The technology is new. The need for product judgment is sharper.

• Understand the problem before choosing the feature. “Add a copilot” and “automate it with agents” are solutions, not customer problems. What is the user trying to do? Where are they struggling? What would better look like?
• Make the trade-offs explicit. AI can produce incorrect or incomplete outputs. Which errors matter most? Which can the system catch reliably? What should happen when it can't? These are product decisions, not just technical ones.
• Earn the user's trust. A confident answer can still be wrong. Users need to know what has been checked, what remains uncertain, and where their judgment is needed.
• Measure whether the work improved. Speed and model scores are useful signals. But did the user save time after checking the answer? Did they have to rewrite it? Did it help them make a better decision?

I've faced versions of these questions throughout my career: defining what a product can promise, understanding how it can fail, and aligning teams around difficult trade-offs.

AI adds new skills I'm learning by building: model behavior, evaluations, latency, and cost. But the question I keep coming back to is familiar:

What has to be true for this to help someone in their real work?