AI

What is context engineering for LLMs?

Context engineering explained: what an LLM's context window holds, why context rot makes overfilled contexts worse, and how to choose what the model sees.

  • 8 min read
  • 4 questions
  • +30 XP
Transcript

Context Engineering, explained by cats. Kitty gives an LLM instructions. Chat history. files. and maybe a giant spec. Everything the model can currently see is the context window. It is the model’s working memory. Not forever memory. Just what fits on the desk right now.

Small window? Less to work with. Big window? More room for rules, examples, documents, and tool results. But bigger is not always better. As the window fills up, the model can get sloppy. It may forget the first instruction. Miss details buried in the middle.

Or act very confident while losing the plot. That is context rot. Context rot is when performance gets worse as more stuff gets packed into the context window. So Kitty does context engineering. That means. Carefully choosing what the model sees before it answers.

Not just one clever prompt. But the whole setup. Instructions. Relevant history. Useful files. Tool outputs. Summaries. Examples. Keep the important stuff. and remove the junk. Because if the context is messy, the answer gets messy. The model is the brain. The context window is the desk. And context engineering is how Kitty keeps that desk clean enough to think.