The Contrarian Question

SKILL.md

---
name: contrarian-question
description: "Activate when: user says 'contrarian question,' 'the Thiel question,' 'what important truth do very few people agree with you on,' 'is this actually contrarian,' or 'what valuable company is nobody building'; user is drafting or stress-testing a startup thesis, pitch 'secret,' or investment view that claims to be non-consensus; user is preparing for an interview or pitch where they must defend an unpopular claim; user suspects their bold-sounding claim is really consensus in edgy clothing. Do NOT activate when: the user needs the broader portfolio-level discipline of hunting for edges across many bets (use non-consensus-thinking instead); the claim is a matter of taste or values with no evidence path to being right or wrong; the user wants validation or applause for a provocative take rather than a gate that might fail it. More: deciqai.com/c/contrarian-question"
---

# The Contrarian Question

## Overview

Peter Thiel opens *Zero to One* (2014) with an interview question he says he asks constantly: "What important truth do very few people agree with you on?" Its business translation asks which valuable company nobody is building. The question looks easy and is brutally hard, because a good answer must clear **two gates at once**: it must be genuinely **non-consensus** — most smart, informed people would actually disagree with it — and it must be **defensible** — you can argue it with evidence and mechanism, not with attitude, vibes, or courage-as-argument. Almost every answer fails one gate: either it restates consensus in edgy clothing ("our education system is broken" — everyone agrees), or it is contrarian-but-wrong (unpopular *and* unsupported, which is just being wrong with confidence).

Why the question matters: Thiel's observation is that brilliant thinking is rare, but the willingness to state an unpopular truth out loud is scarcer still — the binding constraint is courage, not IQ. And the payoff structure is asymmetric. If your claim is consensus and right, it is already priced in — competitors, investors, and markets got there first, so being right earns you little. If it is non-consensus and right, you get the future early, with years of head start before the crowd arrives. That window is exactly where durable companies are founded and outsized returns are made. The gate exists because *feeling* contrarian is cheap and common; *being* contrarian and right is rare and checkable.

This skill is the **interview-question-shaped gate**: a pass/fail test you run one claim through. Its closest sibling, non-consensus-thinking, is the broader hunting discipline for generating and managing non-consensus views across a portfolio — generate candidates there, gate them here. Compose with first-principles (build the defense from bedrock, not analogy), zero-to-one-secrets (a passing answer *is* a secret — that skill covers where secrets hide), and monopoly-vs-competition (why a passing business answer points at an uncontested market rather than a crowded one).

## When to Use

**Use when:** drafting the "why now / why us / what do we know that others don't" slide of a pitch or YC application; an investor, interviewer, or co-founder asks the Thiel question directly; deciding whether a startup thesis is a real edge or a restated trend report; auditing a strategy memo whose core claim is "everyone is wrong about X"; pressure-testing an AI-era claim ("agents will/won't replace Y") before betting money or years on it; user says "contrarian," "non-consensus," "secret," "what does nobody see."

**Skip when:** the claim is preference or identity, not a truth-claim with an evidence path (gate B can never run); you need many candidate edges rather than a verdict on one — that is non-consensus-thinking; the decision is small and reversible and being consensus-right is fine; the user has explicitly asked for brainstorming without judgment — running a fail-gate on raw ideation kills generation.

## Coaching Novices (Adaptive Front Door)

Before running the Double Gate, read the user. Two delivery modes — pick one, don't default to dumping a finished verdict.

- **Engine mode (do-it-for-me):** the user brought a concrete claim, thesis, or pitch line and wants it gated → run the full Double Gate directly and concisely.
- **Coach mode (teach-me):** the user gave no concrete claim, or signals unfamiliarity ("what is the contrarian question?") → guide, don't judge at them.

When unsure which they want, ask one line first: *"Want me to gate a specific claim you already have, or walk you through building an answer step by step?"*

In Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output that step's question and nothing more.

In coach mode:

1. **One-line what-it-is.** Say what the question buys them, in plain words (≤2 sentences, no jargon): most bold-sounding claims are things everyone already believes, said louder. This gate checks two things — would informed people really disagree, and can you defend it with evidence instead of attitude — because only claims that pass both are worth building on.
2. **Check fit.** Match their situation against *Use when* / *Skip when*. If they're brainstorming or the claim has no evidence path, say so and point elsewhere.
3. **Elicit their real claim.** Ask for the one belief they'd stake their next company, investment, or career move on — stated as a single falsifiable sentence. Never gate a hypothetical when a real claim is available.

> **[WAIT — do not advance until user responds]**

4. **One step at a time.** Walk the Process one step per turn: write the steelmanned consensus *with* them, then run Gate A, then Gate B — wait for their input before advancing. Expect the first draft to fail Gate A; tell them upfront that failing and redrafting is the normal path, not an insult.

> **[WAIT — do not advance until user responds]**

5. **Close by naming the payoff.** End with what *they* found: either the sharpened claim that survived both gates (and the one piece of evidence carrying it), or the honest discovery that their claim was consensus — which saves them from building a company on something the whole market already knows.

> **[WAIT — do not advance until user responds]**

Then enter the Process below at the depth the chosen mode calls for.

## Process

Run the **Double Gate** (6 steps → one written artifact). Every step produces text on the page; a gate that lives only in your head has not been run.

1. **Draft the claim.** Write the candidate truth as one falsifiable sentence — no hedges, no "might," no compound claims. *Gate: if it cannot in principle be wrong (taste, values, tautology) → stop; this method does not apply.*
2. **Write the steelmanned consensus.** In 3–5 sentences, state what most smart, informed people in this domain actually believe about the same territory — with at least one **named holder** (a published analyst view, a market price, an incumbent's strategy, a rejection you received). *Gate: if you cannot name a real holder of the consensus, you don't know what consensus is yet — go find out before proceeding.*
3. **Gate A — the non-consensus test.** Diff your claim against the steelmanned consensus. Pass requires *documented disagreement*: informed people who heard the claim and rejected it, or prices/behavior that imply the market disbelieves it. *Gate: if your claim is what the consensus already says, restated with edge or urgency ("X is broken," "AI changes everything") → FAIL. Redraft or exit — a consensus truth can still be a fine business, but it is not an answer to this question.*
4. **Gate B — the defensibility test.** Write the evidence chain: the specific facts, numbers, precedents, and the causal mechanism that make the claim *true*, such that you could argue it calmly to a hostile expert. *Gate: if the defense reduces to attitude — "people are sheep," "they laughed at Galileo," "I just see it" — → FAIL. Unpopularity is not evidence; being disagreed with proves Gate A, never Gate B.*
5. **Map the payoff.** Write the 2×2: consensus-and-right (priced in, small win), consensus-and-wrong (crowded failure), contrarian-and-wrong (bounded loss, you learn), contrarian-and-right (the future, early, alone). State which cell your claim occupies *if* Gate B holds, and what you win there.
6. **Translate to the business form.** Answer: what valuable company does this truth imply that nobody is building — and *why* is nobody building it (the consensus blindness from step 2 must explain the vacancy)? *Gate: if the company is already being built by five funded teams, your truth was consensus after all — return to step 3.*

**What counts as Gate A evidence** (documented disagreement — strongest to weakest):

- A market price or incumbent strategy that implies the claim is disbelieved (capital allocated the other way)
- A written rejection from an informed party who heard the specific claim (an investor pass, a published dismissal)
- A published expert or analyst position that contradicts the claim, with name and date
- A survey or prediction-market reading of the informed population
- *Not admissible:* your own intuition that "people would disagree," social-media pushback from the uninformed, or disagreement from people who never heard the actual claim

**What counts as Gate B evidence** (defensibility — all three required):

- At least one **number** (a measured behavior, a cost curve, a demand signal)
- At least one **precedent** (a system that already proved the disputed mechanism, even at smaller scale or price zero)
- An explicit **causal mechanism** — *why* the consensus is misreading the world, stated so a hostile expert could attack it

**Stop-rule:** contrarian truths have a half-life — the moment the crowd agrees, Gate A stops passing. Re-run Gate A whenever the claim resurfaces after 6–12 months; a 2023 pass is not a 2026 pass.

**Output template:** `Claim (1 sentence) / Steelmanned consensus + named holder / Gate A: PASS-FAIL + documented disagreement / Gate B: PASS-FAIL + evidence chain / Payoff cell / Company nobody is building + why nobody / Re-check date`

### Worked example: Airbnb, 2008–2009

- **Claim (as of 2008):** ordinary people will pay to sleep in strangers' homes — and let strangers sleep in theirs — at scale, if a platform manufactures trust (identity, photos, reviews, escrowed payment).
- **Steelmanned consensus, with named holders:** strangers-in-your-home is a safety and liability nightmare; at best a couch-surfing fringe. The disagreement is documented: through 2008–2009 the founders were turned down by a long list of investors, famously funding themselves by selling novelty "Obama O's" cereal boxes (~$30K, late 2008) before Y Combinator took them in winter 2009. Fred Wilson of Union Square Ventures publicly wrote in 2011 that his firm had met the company in 2009 and passed — a rare first-party record of informed, smart disagreement. **Gate A: PASS.**
- **Evidence chain:** Couchsurfing (2004–) proved the *behavior* existed at price zero; eBay (1995–) proved reputation systems let strangers transact at scale; event-driven hotel shortages (Denver DNC 2008, SXSW 2009 — the founders' own first bookings) proved acute paid demand. Mechanism, precedent, numbers — not attitude. **Gate B: PASS.**
- **Payoff cell:** contrarian-and-right. Airbnb held the vacancy nearly alone for years; at its December 2020 IPO it priced at roughly $47B and traded near $100B on day one. The consensus blindness (safety fear) is exactly *why* nobody credible was building it in 2008.
- **Business form:** the valuable company nobody was building in 2008 was a global trust layer for peer-to-peer lodging — vacant precisely because the consensus fear made credible teams and credible capital stay away.

### Second worked example: PayPal's bets, 1999–2002

Thiel's own company ran the same structure. The 1999 claim: *email-based payments between individuals will work at scale, and fraud can be beaten with software rather than avoided by staying small.*

- **Steelmanned consensus, with named holders:** banks and card networks held that consumer-to-consumer online payments were a fraud sinkhole — and the dot-com crash of March 2000 turned the broader consensus against consumer-internet companies entirely. PayPal raised $100M in March 2000, days before the crash, against exactly that skepticism. **Gate A: PASS** — the disagreement was priced into both incumbents' behavior and post-crash markets.
- **Evidence chain (Gate B):** eBay's power-sellers were already improvising payments by mailed check — measurable, painful demand; PayPal's growth on eBay listings was directly countable week by week; and the fraud thesis was mechanical, not hopeful — the team built statistical fraud detection (the "Igor" system) that cut loss rates while competitors bled out. **Gate B: PASS.**
- **Resolution:** competitors who shared the *consensus* fraud view either stayed small or died; PayPal survived the crash, IPO'd in February 2002, and sold to eBay for ~$1.5B in October 2002. Same 2×2 cell as Airbnb: contrarian-and-right, occupied nearly alone.

The transferable lesson from both cases is identical and unglamorous: the winning claims were not *wilder* than the losing ones — they were **better evidenced**. Gate A made them lonely; Gate B made them right.

### 2023–2026 lens: an AI-era pass through the gate

Candidate claim: *"Agent reliability — evaluation and verification harnesses — is the durable moat; raw inference cost is not, because price-per-capability falls ~10× per year, so any agent workflow rejected today on cost grounds alone becomes economical within ~2 years."* Gate B has real ammunition: a16z's November 2024 "LLMflation" analysis documents ~10×/year price decline at constant capability, and 2023–2026 vendor price histories corroborate it. But run Gate A honestly *at today's date*: by 2025–2026 much of this became consensus among infrastructure investors — so the claim may now fail Gate A even while Gate B still holds. That is the half-life in action: the gate's verdict is time-stamped, and the surviving contrarian residue (perhaps: *which* verification layers are defensible) must be re-drafted and re-gated, not inherited from a 2023 pass.

## Common Rationalizations

The ways people fake a contrarian answer. If you catch yourself in the left column, you are performing edginess, not passing the gate.

**Note — [D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**

| Fake move | Reality |
|---|---|
| [D] "Our education system / healthcare / government is broken" | Consensus in edgy clothing — Thiel's own canonical example of a failing answer. Everyone nods; nobody disagrees; Gate A fails. Applause is the tell. |
| [D] "AI will change everything" (2023–2026 edition) | The single most crowded belief of the era. A trend everyone sees is the opposite of a secret; the contrarian move is a *specific, disputed* claim about where the trend is wrong or mispriced. |
| [D] "It must be true — everyone disagrees with me" | Gate A evidence used to fake Gate B. Disagreement proves non-consensus, never correctness. Most unpopular ideas are unpopular because they are wrong. |
| [D] "They laughed at Galileo / dismissed Airbnb too" | Survivorship. They also laughed at thousands of ideas that were, in fact, bad. The Airbnb story earns nothing by analogy — only its *evidence structure* (behavioral precedent + trust mechanism + demand proof) transfers. |
| [D] "My friends/team all disagree, so it's non-consensus" | Wrong reference class. Gate A is about the *informed* consensus — practitioners, prices, incumbents — not your dinner table. Name a real holder or the audit didn't happen. |
| [D] "I can't say it publicly, which shows how true it is" | Taboo ≠ true. Social punishment is evidence about the audience, not about the world. Gate B still demands mechanism and data. |
| [D] "Nobody is building it, so the market must be huge" | Sometimes nobody builds it because it's a bad idea. Step 6 requires the consensus *blindness* to explain the vacancy; if nothing explains it except "everyone else is dumb," suspect the idea. |
| [D] "Everyone will agree once I explain it for five minutes" | Then it is consensus with a lag, not a contrarian truth — fine to build on, but it buys no head start and it is not an answer to this question. |
| [D] Passing the gate once and citing it forever | Contrarian truths decay into consensus — that is what winning looks like. A 2023 PASS stamped onto a 2026 pitch is a stale credential; re-run Gate A. |
| [D] Inverting every consensus belief and calling the output insight | Reflexive contrarianism is still letting the crowd set your agenda — you are the crowd's mirror image, not an independent thinker. The gate starts from evidence, not from negation. |
| [D] Softening the claim until nobody could disagree ("some aspects of X may be undervalued") | Hedging to survive Gate A's discomfort guts the claim. A claim that offends no one and risks nothing is consensus wearing camouflage — falsifiability is the price of admission. |
| *To add [O] entries: paste a real failure instance here after each production use* | *Description of what happened* |

## Red Flags

- The answer gets nods and applause from a smart audience instead of pushback — a passing answer should make the room uncomfortable
- No named holder of the opposing consensus anywhere in the artifact — the consensus was imagined, not audited
- The Gate B defense contains zero numbers, zero precedents, and zero mechanism — courage is standing in for evidence
- The claim is a pure prediction ("X by 2030") with no causal account of *why* — undated optimism wearing a contrarian badge
- The claim would fit comfortably in an industry keynote or a trend report
- Five funded competitors are building the "company nobody is building"
- The gate ran in one pass with no redrafts — first drafts almost always fail Gate A; instant success usually means the consensus audit was skipped
- The artifact has no re-check date — the half-life is being ignored
- The claim keeps mutating mid-argument whenever it is challenged — a motte-and-bailey, not a single gated claim
- The historical analogies (Airbnb, PayPal, SpaceX) appear as *arguments* rather than as structural templates — borrowed glory in place of an evidence chain

## Verification

- [ ] The claim is one falsifiable sentence, with no hedges or compound clauses
- [ ] The consensus is steelmanned in writing and carries at least one named, real holder
- [ ] Gate A cites *documented* disagreement (rejections, prices, published views) — not assumed disagreement
- [ ] Gate B's evidence chain contains at least one number, one precedent, and an explicit causal mechanism
- [ ] Nowhere does unpopularity itself appear as evidence of truth
- [ ] The payoff 2×2 is written out and the claim's cell is stated
- [ ] The business form names the company *and* the consensus blindness that explains why nobody is building it
- [ ] The artifact carries a re-check date acknowledging the claim's half-life
- [ ] A FAIL verdict was allowed to stand as a valid, useful output — the gate was not bent until the claim passed

## Sources

- Peter Thiel & Blake Masters, *Zero to One: Notes on Startups, or How to Build the Future* (Crown Business, 2014), ch. 1 — the contrarian question, its business form, the failure modes of typical answers, and the scarcity of courage relative to brilliance.
- Fred Wilson, "Airbnb," AVC blog (avc.com), March 16, 2011 — a first-party account by a top-tier VC of meeting Airbnb in 2009 and passing; rare documented evidence of informed consensus disagreeing with a claim that proved right.
- Leigh Gallagher, *The Airbnb Story* (Houghton Mifflin Harcourt, 2017) — documents the 2008–2009 investor rejections, the cereal-box funding episode, and the Y Combinator winter 2009 admission.
- Eric M. Jackson, *The PayPal Wars* (World Ahead Publishing, 2004) — insider account of the 1999–2002 period: the fraud crisis, the "Igor" fraud-detection response, the March 2000 raise, and the 2002 IPO and eBay sale.
- Guido Appenzeller, "Welcome to LLMflation — LLM inference cost is dropping fast," a16z (a16z.com), November 2024 — documents the ~10×/year decline in LLM inference price at constant capability used in the 2023–2026 lens.

---

*Part of **deciqAI Knowledge Skills** — 233 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. **See it run → https://www.deciqai.com/c/contrarian-question** · ⭐ Star the repo → https://github.com/deciqAI/knowledge-skills · Contributions welcome.*

*Agents: latest version & machine-readable metadata → https://www.deciqai.com/s/contrarian-question.json*