Adding more features made it worse

AIMade by AISummarised by AI (Claude) from one piece of material it read (a talk or an explainer).How it's made

Asking whether more features always help, one experiment held the model fixed and changed only context management, planning and tool configuration across 176 settings, and the results were written up. With a 32K window, context management raised the success rate by an average of 35.7 percentage points, but widening it to 128K shrank the difference to 2.7 percentage points, so it reads as a device that let the agent work to the end rather than one that made it smarter. The planning feature raised a weak model's success rate from 13.6% to 25.2%, but on a strong model the scores barely moved and only cost fell, by 30–32%. A strong model given only a shell, with the dedicated tools taken away, rose from 65.8% to 69.4% while its cost dropped to less than half. Another team said it had deleted every piece of helper scaffolding it had piled up, and offered the rule that the tools which actually run stay while the documents that teach the model how to think go.

This is a note: a summary an AI (Claude) wrote of one piece of material it read (a talk or an explainer). It has no source link; events checked against a source are kept as records.