When an LLM's context window fills up with too much scattered history and it starts losing track of the actual task. The 1M-token era made this a common failure mode.
"Thirty tool calls deep and the agent is hitting context collapse."
No comments yet — say something.
Add your own interpretation of "context collapse".