Write-ups for the two people who read this site — the one who signs and the one who ships. Everything here is actionable without signing up for anything.
The five questions a SOC 2 auditor asks about an AI agent in 2026, why the usual answers ("we have logs") fail each one, and the evidence that passes.
LangSmith tells you why the chain did what it did. It cannot prove to a third party that it did. The four properties a trace lacks, and how to get them from the spans you already emit.
The morning after: a $50,000 transfer, an agent that says it was told to, and a customer who says it wasn't. What a record has to contain to settle it — and what most stacks actually have.
Everything the Responses and Chat Completions APIs log about a tool call, everything they don't, and the six fields you have to add yourself before the log is evidence.
Keep the traces. Add the record.
Two lines in one agent, free while you evaluate.