From Columns to Rewards: Automating the Two Pillars That Drive Modern AI

Summary

Former Google engineers automated feature engineering with AutoML—turning manual spreadsheet column tinkering into processes that added tens to hundreds of millions in revenue—and the author argues reinforcement learning (RL) is now at that same inflection point. Today RL can generate plans but humans still craft reward functions, keeping it a craft rather than a scalable, automated pipeline like AutoML. While AutoRL is not imminent given open technical debates, its development could unlock a new wave of agentic systems and materially accelerate business applications and revenues tied to autonomous decision-making. The stakes: whoever automates reward design first could capture outsized market and product advantages in next‑generation AI services.