# How Long Contexts Fail  URL:: https://www.dbreunig.com/2025/06/22/how-contexts-fail-and-how-to-fix-them.html Author:: Drew Breunig  ## AI-Generated Summary Long context windows in AI models can cause problems like confusion, distraction, and conflicting information. These issues make agents perform worse, even with more data available. To fix this, careful management of context and tools is needed. ## Highlights > Context Poisoning is when a hallucination or other error makes it into the context, where it is repeatedly referenced. ([View Highlight](https://read.readwise.io/read/01m32sw4trz8ghjbx06txf3hbw)) > Context Distraction is when a context grows so long that the model over-focuses on the context, neglecting what it learned during training ([View Highlight](https://read.readwise.io/read/01m32swhhhmw8zj51c39cv185j)) > If models start to misbehave long before their context windows are filled, what’s the point of super large context windows? In a nutshell: summarization[3](https://www.dbreunig.com/2025/06/22/how-contexts-fail-and-how-to-fix-them.html/#fn:summarization) and fact retrieval. If you’re not doing either of those, be wary of your chosen model’s distraction ceiling. ([View Highlight](https://read.readwise.io/read/01m32vt0s663nzan1bfq4bn6k2)) > Context Confusion is when superfluous content in the context is used by the model to generate a low-quality response. ([View Highlight](https://read.readwise.io/read/01m32sx60nx7wgwb0v75cmd04g)) > Context Clash is when you accrue new information and tools in your context that conflicts with other information in the context. ([View Highlight](https://read.readwise.io/read/01m32sxbj85tx68dgbrttp6sfy))