Short, factual notes on GenAI agents in production: evaluation, guardrails, observability, productionisation on Google Cloud.https://alexdahan.deven-gbhttps://alexdahan.dev/en/notes/memoire-agent-deux-niveauxhttps://alexdahan.dev/en/notes/memoire-agent-deux-niveauxA well-injected playbook answers ‘what does the rule say?’, never ‘how does this team work?’. On a contract review agent, I separated each lawyer’s personal memory from the team doctrine, approved by an admin. Two levels, two prompts, and a model that proposes but never writes.Fri, 02 Oct 2026 00:00:00 GMTagentsmemorygovernancelegalcontact@alexdahan.dev (Alexandre Dahan)https://alexdahan.dev/en/notes/observabilite-agentshttps://alexdahan.dev/en/notes/observabilite-agentsAn untraced agent is a black box that sends you a bill. The trace model I use in production on a legal agent: Langfuse for LLM calls, cost calculated alongside, and tracing that can’t bring production down.Fri, 02 Oct 2026 00:00:00 GMTagentsobservabilityLangfusecostscontact@alexdahan.dev (Alexandre Dahan)https://alexdahan.dev/en/notes/tracked-changes-word-mot-a-mothttps://alexdahan.dev/en/notes/tracked-changes-word-mot-a-motA lawyer doesn’t review a rewritten paragraph: they review a diff in their own document. How to produce native, word-level OOXML tracked changes, and why Google Docs suggestions take a detour.Fri, 02 Oct 2026 00:00:00 GMTlegalOOXMLWordadoptioncontact@alexdahan.dev (Alexandre Dahan)https://alexdahan.dev/en/notes/maker-checkerhttps://alexdahan.dev/en/notes/maker-checkerAn LLM on its own makes up references. On a multi-agent system producing scientific briefs, separating the one who writes from the one who verifies — then wiring in programmatic verification — delivered zero hallucinated citations.Fri, 18 Sep 2026 00:00:00 GMTagentsADKguardrailsevaluationcontact@alexdahan.dev (Alexandre Dahan)