Writing

Putting guardrails around a LangChain agent

Four ways to keep an agent in check, redacting PII on input, pausing for approval before a risky tool runs, blocking bad requests up front, and checking the model's answer before it goes out.