The article is about how OpenAI and Anthropic each build data agents differently, and what that reveals about the challenge of making AI useful on real enterprise data. It shows that raw file access alone is not enough - agents need metadata, schemas, lineage, and other context to work reliably with data stored in systems like S3: We read OpenAI's and Anthropic's data-agent posts - DataChain OpenAI’s internal system is described as working well because it sits on top of a rich warehouse e...