Ask HN: How is your company managing internal AI agents?
2 comments
can treat each internal agent like a small product: one owner, one budget, and one event ID that bundles retries and tool calls into the true cost per successful outcome.
If you only track provider spend, nobody knows what one useful outcome actually costs. Behavior changes should go through the same path as product changes (review/audit trail/rollback)
If you only track provider spend, nobody knows what one useful outcome actually costs. Behavior changes should go through the same path as product changes (review/audit trail/rollback)
How many agents are running internally that you know of? Who manages them day-to-day -- engineering or the business team that uses them? How do you track what they cost (LLM API fees, compute)? If the business team wants to change the agent's behavior, what's the process?
Genuinely trying to understand the landscape, not selling anything.