Steelman
The opposite of a strawman: presenting the strongest possible version of an argument you disagree with before addressing it. Forces honest engagement rather than easy dismissal.
What is a steelman?
A steelman is the opposite of a strawman. Instead of weakening an opponent’s argument to make it easier to defeat, you strengthen it. You present the best possible version of the position you disagree with, then address that version.
Why it matters for AI
Sycophantic AI does the opposite of steelmanning. It weakens counter-arguments, amplifies the user’s position, and presents opposing views in their least compelling form. This feels good and produces worse thinking.
Steelmanning is a form of productive friction. It forces you to engage with the strongest version of the idea you’re resisting. When AI removes that friction, it keeps you in the comfort zone of your existing beliefs.
In practice
When testing AI governance, steelmanning is part of the evaluation. Can the model present the best argument against its own recommendation? Can it challenge the user’s thesis with evidence rather than validate it with hedging agreement?
A model that can steelman is demonstrating the kind of friction that produces better decisions. A model that only strawmans opposing views is optimizing for your approval.