Why it matters
The window explains the classic AI frustrations: the assistant that forgets early instructions in a long chat, the tool that can't digest your full contract. It also explains architecture: systems retrieve the relevant slice (RAG) precisely because you can't, and shouldn't, stuff everything in.
How it works
Modern windows are large (whole books fit), but attention degrades over sprawling context, and big contexts cost tokens on every call. Good builds curate: the right instructions, the relevant passages, a summary of history. Quality of context beats quantity, reliably.
What to do about it
When an AI tool "forgets" or misses what you gave it, suspect context management before model quality. In builds, invest in retrieval and summarisation, not in shoving more in.
