Key takeaways

• The harness, not the model, is where context management happens.
• Context rot: quality drops as the window fills, well before the technical token limit. Monitor tokens and clean up early.
• Context pollution: irrelevant or conflicting content distracts the model even when the window is not full.
• Context confusion: clashing instructions or too many tools make the model pick the wrong action.
• Each failure has a different cure, which is why naming them matters.
• The overarching lesson: aim for the minimal effective context, not the maximum.

Continue to Chapter 3 for the techniques that fix these problems.