Emily Carter

Steelman Both Sides of a Debate Before You Pick One

Get the strongest case for each side of a debate, rated evidence, the real crux of the disagreement and what should change your mind.

At a glance

Best for: Managers making a policy call, students preparing a debate or essay, and anyone who suspects they've only heard one side of an argument.
Tested on: Claude · Opus 5.5
You fill in: questionleaning
You get: Side A: yes, from day one, with guardrails Position: juniors should learn the tools they'll use for their whole career, under supervision, from the… (full result below)

Prompt

I want to understand this debate well enough to argue either side. Steelman both positions: the strongest version each side's smartest advocates would recognise.

Question: {{question}}
My current leaning (so you can push against it): {{leaning}}

For each side:

1. The position in one sentence.

1. The 3 strongest arguments, each with the kind of evidence that supports it and how strong that evidence is (strong / mixed / weak).

1. The best objection to the other side that this side would raise.

1. The concession a thoughtful advocate of this side would make.

Then:

1. The crux: the 1-2 factual or value disagreements that actually drive the split. For each, say whether it's empirical (could be settled by evidence) or about values (can't be).

1. What would change my mind: given my leaning, the specific evidence or argument that should most move me.

Rules: no strawmen, no "both sides have good points" filler, and don't tell me who's right. Label anything you're unsure is factually established.

Fill in

{{question}}: The debate, phrased as a yes/no or either/or questione.g. Should junior developers use AI coding agents from day one?
{{leaning}}: Your current view and why, so the model can push against ite.g. I lean no; juniors should struggle by hand for 6 months first

Example input

question: Should junior developers be allowed to use AI coding agents (such as Claude Code or Cursor) from their first day on the job?
leaning: I run a 12-person dev team and I lean towards no. I think juniors need to struggle with code by hand for 6 months first.

1.8k0

Generated Outputs (1)

2 hours ago

Claude

Opus 5.5

2 hours ago

Generated Output

Side A: yes, from day one, with guardrails

1. Position: juniors should learn the tools they'll use for their whole career, under supervision, from the start.

1. Arguments:

- The job has changed. Reading, directing and reviewing AI-written code is now a core skill, and a ban delays learning it. Evidence: mixed, mostly industry surveys.
- Agents can teach. A junior can ask an agent to explain unfamiliar code instantly instead of waiting for a senior. Evidence: weak to mixed; it depends heavily on how the tool is used.
- Bans don't hold. Juniors will use AI tools privately, so the choice is really between visible, reviewed use and hidden use. Evidence: anecdotal but widely reported.

1. Objection to B: "struggling by hand" assumes the struggle builds the right skills, but much of it is fighting syntax and tooling, not learning design.

1. Concession: unrestricted use in the first months risks juniors shipping code they can't explain.

Side B: no, build the foundations by hand first

1. Position: juniors need a period of unassisted work to build the mental models that make reviewing AI output possible.

1. Arguments:

- Skills come from effort. Research on learning suggests that working things out yourself leads to better retention than being handed answers. Evidence: strong in education research; less direct for programming at work.
- Reviewing needs understanding. You can't catch a subtle bug in code you couldn't have written. Evidence: mixed.
- Early studies suggest AI help can reduce learning. Some recent experiments found learners who used AI assistance understood new material less well afterwards. I'm not certain of the details, so check the studies.

1. Objection to A: "supervised use" assumes seniors have time to review closely, and most teams don't.

1. Concession: six months fully without AI is arbitrary and may leave juniors behind peers elsewhere.

Crux

1. Does using agents early stop juniors building the underlying skills, or just change which skills they build? Empirical. It could be settled with better long-term studies, which barely exist yet.

1. What is the job for: shipping now, or developing engineers for later? Values.

What should move you
If you could show your own juniors explaining and defending agent-written code in review, correctly, after a few months, your case weakens. Try a small trial: give one junior an agent from day one with a rule that they must explain every change in review before it merges, and compare after 3 months.

Why it works

"The strongest version each side's smartest advocates would recognise" sets a higher bar than "give pros and cons". Requiring an evidence strength rating per argument separates research-backed points from plausible-sounding ones. The concession step shows where each side is weakest by its own standards. The crux, labelled empirical or values, is the most useful part: it tells you whether more data could settle the disagreement or whether it comes down to priorities. Giving your leaning lets the model aim the final section at your blind spot.

When not to use it

Not for questions with a settled factual answer (vaccine safety, climate basics), where presenting "two sides" misleads. The model's knowledge of recent studies may be incomplete, so check any evidence it cites. For personal moral dilemmas, a conversation with people affected will teach you more.

Comments (0)

No comments yet. Be the first to share your thoughts!