How to build agents when the smartest AI isn't smart enough
Benchling's AI agents pair LLMs with structured lab data to roughly halve drug-discovery-to-patient time.
“What we've seen is when you kind of put these models on top of the right data, the quality answers go way up.”
Benchling's head of AI, Nick Stone, explains how the company builds domain-specific scientific agents on 14 years of life-sciences data, relying on SQL with embedded table descriptions and production traces rather than pre-built evals. The interview offers practical engineering insight into grounding LLM agents in proprietary structured data, but it is a vendor podcast with limited industry-wide impact.