What's this about?
OrangeAI argues AI coding, despite producing some slop, is protected by years of engineering QA infrastructure (tests, CI, code review) and real-user feedback that catch and constrain the slop. AI knowledge-work ("AI work" — documents, analysis, office tasks) has no equivalent quality-assurance standard yet, so without one it risks being slop end-to-end.
What exactly is "slop" here?
Plausible-looking but low-quality or subtly wrong AI output. In coding, tests and reviewers catch a lot of it before it ships; in open-ended knowledge work, there's often no test to fail, no reviewer, no user bug report — so bad output can pass undetected.
Why does it matter?
31 likes. It's a sharp diagnosis of why the "AI office" race (China's giants, Cursor, ChatGPT Work) is harder than it looks: the missing piece isn't the model, it's a QA layer for non-code work. Without one, scaling AI knowledge-work scales the slop too.