François Chollet sees AI agents scaling

What makes this notable is less the exact timeline than the messenger: when a prominent critic of brittle AI systems starts talking about scalable agents in near-term terms, the debate shifts.

• –The underlying claim is about AI systems doing economically useful work at scale, not just demos or benchmark wins
• –Reddit’s “hard takeoff” framing is stronger than the original post, which reads more like a forecast about agent deployment than instant superintelligence
• –For developers, the practical implication is to pay closer attention to tool use, orchestration, reliability, and long-horizon task execution
• –It also shows how fast the conversation is moving from “are agents real?” to “how soon do they become commercially normal?”