AI 第三时代：持久型 AI 同事、从划船到掌舵，以及野心成为新瓶颈

The third era of AI products may be defined by persistent coworkers that understand goals and keep working over time.

As AI takes on more execution, people shift from rowing every stroke to steering direction and judging outcomes.

When execution becomes cheaper, ambition becomes a bottleneck because teams must imagine outcomes larger than their old capacity.

A central product leadership task is to elevate other people's ambition, not merely coordinate their existing plans.

OpenAI product teams often build for model capabilities expected two or three months ahead rather than only for today's behavior.

Fast experiments and tight user loops matter more than pretending there is a secret room with a complete AGI master plan.

Dogfooding a product intensively reveals friction and enables faster iteration than relying only on external reports.

Users should not need to understand every model and harness; the product should route tasks to the right capability.

Writing should remain a way to develop thought, while AI can research, challenge, and enrich ideas in the middle of the process.

Knowledge work is harder for agents than coding because quality is often subjective and lacks an executable test.