Product Fieldnotes

I’ve seen teams celebrate a product rollout too early. The feature shipped. Access was granted. Training happened. The dashboard showed “adoption.” But people still did the real work the old way. That changed how I think about AI products too. Deployment is not adoption. Adoption is not behavior change. Behavior change is not value. The hard part starts after the launch.

AI agents change the PM job in one important way: You’re not shipping software that only answers. You’re shipping software that can act. So “is the model accurate?” is no longer enough. 5 questions PMs should ask before giving an agent authority. 🧵 cc @pingali @carveragents

One of the most expensive product mistakes I’ve seen started with a sentence everyone thought they understood. The requirement was written. The meeting happened. Everyone said “aligned.” Then delivery started. Product meant one thing. Engineering assumed another. Operations had a third interpretation. Customer-facing teams were preparing for a fourth. Nothing was technically wrong. The problem was that nobody had forced the ambiguity into the open. Since then, I’ve stopped asking: “Is the requirement clear?” I ask: “What could two smart people interpret differently here?” That question has saved me more time than any prioritization framework.