What makes an AI agent production-grade
Evaluation suites, tool boundaries, approval gates, observability, cost ceilings, and rollout: the six things that separate a demo from an agent you can run in production.
ReadShort, specific writing on AI agents, security operations, and open-weight models, from an engineer who runs these systems inside a large enterprise. No trend pieces.
Each piece ends with a checklist you can apply this week.
Evaluation suites, tool boundaries, approval gates, observability, cost ceilings, and rollout: the six things that separate a demo from an agent you can run in production.
ReadWhere agents earn their place in the SOC, the actions they must never take alone, why prompt injection is a security operations problem, and how to design the human in the loop.
ReadA decision framework for open-weight fine-tuning versus frontier APIs: the four reasons that justify it, the reasons that do not, LoRA versus full fine-tune, evals, and private hosting in your VPC.
ReadSend it. The founder reads every message and replies directly, usually with a more specific answer than a blog post allows.