The Prompt Lab — Steelman Prompting

Steelman Prompting

The Technique

Steelman prompting asks the model to construct the strongest possible version of an argument, position, or counterpoint — not a strawman, but the most charitable, rigorous interpretation. It works because models default to surface-level synthesis when asked to “consider both sides,” but explicit steelmanning forces engagement with the best opposing evidence, not just the most obvious.

The Naive Prompt

What are the pros and cons of remote work for software engineering teams?

Why It Falls Short

This produces the Wikipedia version of the debate: “pros include flexibility and talent access; cons include communication overhead and isolation.” It’s balanced in the worst way — symmetrically shallow. You learn nothing you couldn’t have guessed, and the “cons” are usually softened versions of real objections.

The Improved Prompt

I believe remote-first engineering teams consistently outperform co-located ones 
when properly managed. Steelman the strongest possible case AGAINST this position. 

Requirements:
- Draw on the most credible research, not anecdotes
- Identify the failure modes of "properly managed" as a caveat
- Explain why a thoughtful, experienced engineering leader might find 
  this belief dangerously naive
- Do not soften the critique to be polite

After the steelman, briefly note which part of the argument you personally 
find most difficult to rebut.

Why It Works

The explicit framing — “strongest possible case,” “dangerously naive,” “don’t soften it” — overrides the model’s tendency toward diplomatic hedging. Asking it to identify the hardest-to-rebut point at the end forces a second pass of genuine evaluation rather than a laundry list. You now have something worth arguing with.

When to Use This

Before committing to a technical decision — feed your architecture choice or vendor selection to Claude Opus 4.7 or GPT-5.4, ask it to steelman the case for the alternative you didn’t pick. If the steelman is weak, you have more confidence. If it’s strong, you caught something.
When writing persuasive content — steelman your audience’s likely objections before drafting. The exercise surfaces the two or three arguments you’ll actually need to address, not the ones that are easy to dismiss.
When you suspect you’re in an echo chamber — if everyone on your team agrees, steelman prompting on Claude Sonnet 4.6 (slightly less sycophancy pressure at lower cost) can surface the structural critique your groupthink has been quietly ignoring.

The Prompt Lab runs weekly in Stochastic Sandbox. Have a technique you’d like covered? Reply to this edition.