AI anxiety reveals deeper problems with how we measure value

In a guest essay on Terence Tao's blog, Karthik C. S., Amit Sahai, and Jie Xue argue that anxiety over AI replacing researchers stems from an error made long before the models arrived: measuring intellectual worth by individual throughput. When academic culture started evaluating mathematicians by publication counts and citation metrics, it treated researchers like production engines. An automated system that generates formal proofs faster naturally feels like an existential threat to anyone whose standing is tied to output volume.

The same dynamic plays out across software engineering. The claim that AI will replace developers falls into that exact trap by treating an engineer's primary job as typing code and closing tickets. I agree with the authors, falling back on vague human advantages like intuition or taste doesn't change the framing. It still pits man versus machine. A working derivation does not gain validity because a human authored it, and a codebase does not become better simply because a person typed every line by hand.

The deeper issue is that choosing the wrong KPIs creates misaligned incentives, and misaligned incentives lead to bad outcomes. When organizations measure value by raw throughput, teams optimize for immediate volume at the expense of system resilience and depth. Offloading every intermediate task to AI might pump up sprint velocity today, but it breaks the apprenticeship pipeline that turns junior engineers into senior problem solvers. Measuring only output confuses the artifact for the work that actually sustains a discipline.