Agentic coding and persistent returns to expertise
…on benchmarks such as METR's time-horizon evaluations , frontier models can now complete software tasks that would take a person hours, autonomously working through obstacles along the way. But what does…