AI-era Product Management

Gokul Rajaram argues that AI changes product management because software is becoming faster, cheaper, and less deterministic to build. The durable human role is judgement: choosing what matters, defining the customer behaviour change, and evaluating whether AI-generated output is good enough.

PM as keeper of the why

Rajaram's core product definition is balancing customer needs and business needs. Every feature should have a hypothesis stated as a customer behaviour change: users go from doing X to doing Y, from spending X minutes to Y minutes, or from one customer state to another. The PM protects this why while engineers, researchers, designers, and PMs increasingly build prototypes together.

PMs must become hands-on

Rajaram says PMs are starting to check code into production repositories with tools like Claude Code and Codex, subject to review. Product and design roles are also converging as AI can work within established design systems. The product manager of the AI era needs enough hands-on fluency to prototype, evaluate, and challenge the capability frontier rather than simply write specs.

Non-deterministic products need evals

AI software can respond differently to small input variations. That makes evaluation a core product responsibility. PMs and researchers need to design evals, including AI-assisted evals, to test whether outputs are useful, safe, and shippable across important use cases.

Future-proof skill: judgement

When a thousand AI agents can write code, the scarce question is not whether something can be built but whether it should be built. Judgement applies to product priorities, code review, design coherence, and customer-value trade-offs.