Product Manager @BenTannyhill on why LangSmith Engine routes trace investigation through screener + verifier sub agents instead of letting the main agent read everything.

Listen to the full conversation on your favorite podcasting platform. ⏯️ YouTube: 🎧 Apple: 🟢 Spotify:

@bentannyhill Context engineering is becoming more important than prompt engineering. Routing the right information to the right agent is where reliability improves.

@bentannyhill Multi-agent systems aren’t just about parallelism—they’re about controlling context. Smaller, specialized agents usually outperform one agent trying to reason over everything.

@bentannyhill Screener + verifier is a real pattern. The thing that breaks it at scale is when the screener's understanding of what matters diverges from the verifier's over long runs. Shared persistent context between sub-agents is where it gets interesting.

@bentannyhill Really like the shadow-production approach. Have you thought about running the agent against an isolated DB copy for the write-access problem instead of hand-mocking all the endpoints?

@bentannyhill screener + verifier is the right split. the main agent reading its own traces risks hallucination loops where it reinforces its own biases through query expansion.

@bentannyhill Agent orchestration is becoming just as important as the models themselves. 🚀