The Steel Man Bot — Refactor If the Counter-Argument Wins
Take every opinion baked into this codebase — comments, commit messages, design docs. Find the strongest counter-argument to each. If it wins, refactor accordingly.
Why this exists
Design opinions calcify. The tradeoff that justified a decision in month one gets forgotten by month twelve, but the convention keeps shipping anyway — encoded in folder layouts, repeated in code review, defended on reflex. The codebase ends up enforcing a position nobody on the team could still argue for. The Steel Man Bot is the antidote: it forces the strongest possible counter-argument into the room, on a regular cadence, before the original rationale fully decays.
What you get back
- A ranked list of the load-bearing opinions in your codebase — extracted from comments, commit messages, design docs, and folder structure.
- The strongest counter-argument for each one, written as if by an engineer who genuinely disagrees with you and is better at this than you expected.
- A verdict on which opinions survive the challenge and which ones lose on the merits.
- A refactored version of the code where the counter-argument won, with a written diff of what changed in your thinking and why.
When to reach for this pattern
Run it as pre-rewrite reflection before you commit a quarter to a new direction. Run it before you double down on a contested architecture that keeps generating friction in code review. Run it when a convention has outlived its rationale — when you catch yourself defending a structure because it is the structure, not because you can still name the tradeoff it was solving. The point is not to flip every position. The point is to make sure the ones that survive are the ones that earned it.
Take any of my opinions baked into this codebase — comments, commit
messages, design doc paragraphs, decisions encoded in folder layout.
Find the strongest counter-argument to each. Steelman it: argue it
better than the engineer who actually disagrees with me would. If
the counter-argument wins on its merits, refactor the relevant code
accordingly and explain what changed in my thinking.
Paste this into Claude, Cursor, or Copilot. Change one thing that matters to you.
What I learned shipping it
- The strongest critic of your code is the version of you that argued the other side and lost.
- Most architectural debates die because nobody bothers to steelman the loser — the winner just keeps shipping.
- A convention you can no longer defend out loud is a convention that has already expired.