Writing from the Sentrial team
Concepts, tutorials, and product updates on building reliable AI agents in production.
What is agent drift, and why does it matter?
Your evals pass. Your demos are clean. But three weeks into production, users are reporting weird responses. We explain the mechanics of agent drift and how to catch it early.
Instrumenting the OpenAI Agents SDK with Sentrial in 10 minutes
A step-by-step guide to adding production monitoring to an agent built with the OpenAI Agents SDK. We cover session tracing, tool call capture, and drift alerts.
Introducing baseline snapshotting
You can now pin any session as a verified baseline. Sentrial will compare every future session against it and alert you when behavior diverges beyond your defined threshold.
Silent regressions: the bug you never see in staging
Unlike traditional software bugs, silent regressions in LLM-based agents often pass every automated check. They only surface in production, through user feedback or degraded outcomes.
Why we built Sentrial
We were building an AI agent that passed every eval. Then we deployed it. This is the story of what happened next — and why we decided to build the observability tool we wished we had.