What is it?
The Bitter Lesson
An influential idea (from researcher Richard Sutton) stating that, over the long run, general methods that scale with more computing power tend to beat clever hand-built structures that encode human knowledge. Applied to agents: the special scaffolding and tricks you build today to prop up a weak model often become dead weight once a stronger model arrives. The practical advice is to keep your system flexible, test it across model strengths, and be willing to delete structure as models improve.
From the AI agent glossary