A few years ago, when this all started, we had major breakthroughs fairly often that would throw out all the existing tooling. Now, most of the breakthroughs have to do with improvements to the tools instead. All the gains I see personally have to do with agent upgrades, system prompt tuning, skill tuning, context optimization, etc. Didn’t the last Anthropic release come with some system prompt updates too? Plateauing of the LLMs themselves has long since been the case but that means they’re stable enough to build on which isn’t bad.