Why your AI agent stops following your rules
Long tasks make AI agents forget your rules. It is not random. It is space: everything the agent works with sits in its context window, and your rule ends up buried under the pile. Four context engineering habits keep the window clear: 1. Put away used recipes: clear old tool results once they have been used. 2. Squash the pile into a summary: compaction. Summarise the history, then carry on from it. 3. Keep notes in the recipe tin: a memory file, kept outside the window. 4. Send a helper: a sub-agent works in its own clean context and returns a short summary. Chroma tested 18 models in July 2
Discussion
Rejoignez la communauté pour réagir et poser vos questions.
Aucun message pour l'instant. Lancez la discussion !