Amy Mitchell: Why AI initiatives break normal product manager instincts

Published on May 12, 2026, Amy Mitchell’s essay argues that the product management skills developed in stable, well-defined contexts actively interfere with AI transformation work.

The central claim is that PMs trained in traditional environments develop an instinct toward stabilization: they map multiple steps ahead, sequence work to reduce risk, and try to understand the full system before committing resources. That instinct serves them well when building features for a predictable product. It becomes a liability when the product is an AI system whose behavior is probabilistic and context-sensitive.

Mitchell distinguishes between two leadership modes. Stabilization leadership optimizes for multiple future steps. Transformation leadership enables the next immediate step and uses that step’s results to decide what follows. In AI product contexts, the goal is “learning enough for controlled risk in a bounded pilot” — not full system stability before action. Waiting for clarity that won’t arrive is one of the patterns she identifies as a learned PM behavior that causes AI initiatives to stall.

She also points to how AI systems expose organizational inconsistencies earlier than conventional feature work does. Traditional products behave deterministically: a form validates what the schema requires, a button triggers the expected function. AI systems interact with workflows dynamically, which means they surface upstream inconsistencies — in how different teams define a term, how a process is documented versus how it is actually done — before those inconsistencies would appear in a conventional release cycle. PMs who aren’t expecting this often try to eliminate the inconsistency before shipping, rather than treating the exposure as signal and scoping accordingly.

The article uses a hypothetical PRD scenario to illustrate the sequencing principle: rather than writing comprehensive requirements and waiting for alignment, the guidance is to define the tightest possible slice where success is measurable, ship to a narrow audience, and use the results to decide the next scope boundary.

The piece is most useful for mid-to-senior PMs who are being asked to lead AI features or transformation initiatives for the first time, particularly those who are bringing strong prior product experience and finding that their instincts are generating friction rather than momentum.