Note on Better Models: Worse Tools via Simon Willison’s Weblog

Armin theorizes that this is because more recent Anthropic models have been specifically trained (presumably via Reinforcement Learning) to better use the edit tools that are baked into Claude Code. This has the unfortunate effect that other coding harnesses, such as Pi, may find that their own custom edit tools are more likely to be used incorrectly. FROM: Simon Willison's Weblog Better Models: Worse Tools Source Oh man, does this mean we’re in a golden age of generalist/open-source model harnesses that will eventuallly deteriorate as the models are more specifically trained for their own proprietary environments? Or is the discrepancy and difference in harnesses just a temporary path finding our way to the best harness patterns, at which point they converge and become interchangeable like the models beneath them?