נלחמים באי-סדר עם אי-סדר

תקציר מקורי באנגליתYou cannot tell great engineers what to do, and you increasingly cannot tell what an agent did either, so Vaibhav Gupta's answer is to fight slop with slop. At Boundary the team turns the same cheap, sloppy generation loose as a tool: agents that run constantly over the transcripts of other agents, flagging hallucinations, spotting which tool calls produced errors, and comparing which approaches produced fewer. He pairs that with hard invariants, the design docs, rules, and CLI checks that do not change for months and tell you exactly where a codebase stops converging, so the messy detection layer sits on top of something stable. The deeper move is to attack the foundational layer from first principles. Instead of trusting generated code, he leans on type systems that make whole classes of