PRODUCTIVITY: THE EVIDENCE IS MIXED, BUT THE PATTERN IS CLEAR

Summary from the TLDR newsletter

AI increases productivity on well-scoped tasks where errors are detectable. This breaks down when tasks get harder. When AI agent output is scored algorithmically, it looks moderately capable, but when scored holistically, performance drops substantially. Estimates about AI's effect on productivity drop when task reliability is factored in.