Economics, Business & Finance: Steelmanning contested policy

What you'll be able to do: Build the strongest version of a position you don't hold, which is the only honest way to engage with the subject's genuinely contested questions, and the difference between analysis and picking a side early.

Why this matters more here than in most subjects

Most subjects have a settled core and a thin contested edge. Economics has a wide contested middle, sitting right next to a settled core that looks identical on the page, both are drawn as a supply-and-demand diagram, both get a confident paragraph from AI, and nothing marks the boundary.

Minimum wage and rent control are the clean pair. Both are price controls in a two-line diagram; both get taught the same week. Rent control's long-run effect on housing supply and quality is close to the nearest thing the profession has to a consensus view, and it remains one of the most fought- over policies in local politics. Minimum wage is closer to the reverse: a genuine, live empirical dispute among labour economists, whether local labour markets behave more like the competitive model (a floor above equilibrium costs jobs) or more like a market with real employer power, where a floor can raise wages without much employment cost, at least up to some level, even though plenty of laypeople on both sides treat it as obviously settled in their preferred direction.

If you don't build the strongest case for each side of a genuinely contested question, you haven't done the analysis, you've picked a side and decorated it.

What a steelman actually requires

Not "here's a counterargument." A steelman is the version of the position its best-informed advocate would recognise as their own, including the evidence they'd cite and the conditions under which they'd admit they're wrong.

Build the strongest possible case for [position you disagree with]. Use the best evidence and reasoning a genuinely informed advocate would use, not a weak version I'd find easy to knock down. Then tell me: what evidence, if it existed, would change this advocate's mind?

That last question is the test. If the AI (or you) can't say what would change the position, it isn't a steelman of a reasoned view, it's a description of an ideology.

Worked example: the minimum wage

The case for a real employment cost. In a competitive labour market, a wage floor set above the equilibrium wage means more people want jobs at that wage than employers want to hire, and the gap is unemployment, concentrated among the lowest-skilled and least-experienced workers, who are priced out first. Employers can also respond on margins other than headcount: fewer hours, slower future hiring, less on-the-job training, automation brought forward. A floor that looks costless in a same-year employment count can still show up as worse outcomes on those other margins, and as a bigger effect the higher the floor is set relative to the local wage distribution.

The case for limited or no employment cost. Where a single employer (or a small group) has real market power over local wages (a monopsony) that employer can suppress wages below the competitive level, and a floor set inside that gap can raise wages and employment at once, because it's correcting under-hiring rather than causing over-pricing. Empirical studies comparing adjacent local labour markets that raised their minimum wage against similar markets that didn't have found, in a number of cases, little to no measurable job loss, evidence that at least some real-world labour markets behave more like the second story than the first, at least for the sizes of increases actually studied.

What would change each side's mind. The first side would be moved by clean natural experiments showing job losses concentrated among low-wage workers after a hike, especially as the floor rises well above the local median. The second would be moved by the same kind of evidence showing losses even in low-power labour markets, or evidence that very large increases behave differently from the moderate ones mostly studied so far.

Notice what this exercise produces that "which side is right" does not: a specific, falsifiable list of what evidence would settle it, and a clear sense that the answer probably depends on the size of the increase and the market it's imposed on, not a single number that's true everywhere.

Across the disciplines

Macroeconomics. Steelman both a large fiscal response to a downturn and a minimal one, the first argues idle capacity means stimulus is close to free; the second argues expectations and debt dynamics blunt the effect and load cost onto the future.

Trade. Steelman free trade (gains from specialisation, lower prices, a larger pie) and targeted protection (transition costs are real and concentrated, and "the pie is bigger on average" is compatible with a specific group being worse off for a long time).

Business strategy. Steelman a competitor's pricing move before you decide it's a mistake, a price cut that looks irrational as pure economics can be a sound move as a signal to deter a new entrant.

Corporate finance. Steelman both sides of a stock buyback: it returns capital efficiently to shareholders who can redeploy it, or it substitutes for investment the firm should have made instead.

Pitfalls

1. A steelman that previews your rebuttal. If every sentence has a silent "but of course" attached, you've written a rebuttal in a costume, not a steelman.
2. Steelmanning only the side you already hold. The exercise is worthless if you skip the side you find uncomfortable.
3. Treating "both sides have a point" as the output. The output is two specific, falsifiable cases and what would decide between them, not a shrug.
4. Using it on settled questions as if they were contested. Steelmanning the view that rent control has no negative supply effects is a different, harder exercise than steelmanning minimum wage's employment neutrality, you're arguing against the weight of the evidence, not alongside a live professional dispute, and the exercise should say so.
5. The tell: you can predict which side the AI will end up favouring before it starts, because it's structured as argument-then-verdict rather than two arguments held open.

Try this today

Pick a policy or business decision you already have an opinion on. Ask for the steelman of the opposite view, including what evidence would change that advocate's mind.

Then ask the same question of your own position. If you can't say what would change your mind, you don't hold a position, you hold a preference.