What's this about?
A lawyer describes reading a memo produced by a client's legal department with extensive AI help. It looks very pretty: beautiful font, nice formatting, even a handsome — though unnecessary — organisational chart. It reaches a completely incorrect conclusion about whether regulatory approval is required. And worse, it misses two separate reasons the transaction is actually statutorily *prohibited*. His note: this is becoming increasingly common, and will have a multitude of consequences before in-house lawyers learn to use good models instead of cheap slop.
Why does presentation quality make this dangerous rather than merely bad?
Because polish is the heuristic people use for care. A memo with clean formatting and a diagram reads as the product of someone who spent time, and the reader's scrutiny drops accordingly. Generation decoupled the two: presentation is now free and correctness is not, so the signal that used to correlate with diligence now correlates with nothing. The reviewer's instincts are calibrated to a world that no longer exists.
Why does this matter to you?
- Missing that a transaction is *prohibited*, not merely unapproved, is the kind of error that ends deals and careers.
- His diagnosis is specifically about cheap models, which makes it the same problem as the next entry.
- The organisational chart being unnecessary is the tell — generated work adds artifacts because it can, not because they help.