Guess what.... agents, like humans, also get confused with too much context, all at once. "Minimum relevant context injected at the right time" is how you assure quality artifacts and actions from your agentic workflows.

It's just sooo hard to reconstruct the "minimum relevant context" from running infrastructure state matched against a moving target (i.e. the next version you should be upgrading to).