About

Handing incidents, customer comms, and docs to an agent — and where the human decision stays.

All three were built around a human doing the first read — getting paged or assigned, gathering context, deciding, acting. When an agent takes that pass instead, the workflow around it is worth redesigning rather than retrofitting. These posts work through how: the design choices and trade-offs, one piece at a time.

They follow one arc. Something breaks, the customer gets told, and what was learned gets written down. Most of the posts sit on the first stretch — on-call and incident triage, and the agent design behind it: MCP servers, prompt and context engineering, the security boundaries around LLMs, the observability tooling it leans on. The rest cover what follows: the customer-facing communication that happens while something is broken, and the documentation both people and agents read afterward.