Why More Context Makes Your Agent Dumber and What to Do About It — Nupur Sharma, Qodo
More context degrades agent performance because LLMs ignore the middle of long contexts in a U-curve pattern.
“This is like a U curve where some of the things from the start, some of the things from the end make sense but whatever you are providing in between that that is not taken up.”
Nupur Sharma of Qodo argues that growing context windows don't improve agent quality because LLMs attend to the start and end of inputs while dropping the middle, a U-curve 'lost-in-the-middle' pattern. She also notes that adding more tools and multi-agent setups compounds failures through conflicting agent understandings. It matters as a practical caution against dumping more context as a fix, but reflects a known limitation rather than a novel finding.