Harrison Chase questions eval-driven development for autonomous agents

LangChain's Harrison Chase argues that eval-driven development, which iterates on software by measuring it against evaluations, works for narrowly scoped tasks but breaks down for more autonomous agents.

Citing Goodhart's Law, the idea that a measure stops being useful once it becomes a target, he questions how such agents can be systematically improved through hill-climbing. According to his post, the question is left unanswered.