edit: it's satire. but likely not too far off from the reality in 6 months.

> Our process is deliberately, provably, almost tediously legal. One set of AI agents analyzes only public documentation: README files, API specifications, type definitions.

since nearly all open source dependencies couple the implementation with type definitions, I'm curious how this could pass the legal bar of the clean room.

Even if they claim to strip the implementation during their clean room process -- their own staff & services have access to the implementation during the stripping process.