When Agents Meet Physical Data: The Other Physics of Agent Harnesses - Dmitry Petrov, DataChain
Anthropic found agents achieve only 21% accuracy on data projects without specialized data harnesses.
“Anthropic published that accuracy for data projects on their agents is only 21% until you add specific data harnesses to them and provide context.”
Dmitry Petrov of DataChain highlights a critical gap in agentic AI: both Anthropic (21% accuracy) and OpenAI (requiring six context layers) have published findings showing agents struggle badly with data tasks without specialized harnesses. For physical/unstructured data like video, sensor telemetry, and robot data, the complexity explodes further—2,000 video files can generate millions of nested objects. The talk argues agents need purpose-built data harnesses that let them see, touch, verify, and remember physical-world datasets.