In 1976, the CIA had a problem. Their analysts kept reaching the same conclusions about Soviet capability — and those conclusions kept being wrong. So the agency tried something radical: they assembled an external group, gave them the same intelligence, and told them to argue the opposite case as aggressively as possible. They called it Team B. The technique it spawned is now called red-teaming, and it's how serious organisations stress-test serious decisions.
Today, red teams operate inside the Pentagon, OpenAI, Anthropic, Amazon, every major bank, and basically every cybersecurity team on Earth. They don't operate inside most startups. That's a mistake.
Amazon, OpenAI, and the US military all use red teams before they ship. Your startup doesn't. Guess which one is most likely to blow up.
What red-teaming actually means (for founders)
A red team is a group of people whose explicit job is to attack the plan. Not constructively critique it. Not "yes-and" it. Attack it. Find the failure modes the people who built it can't see precisely because they built it.
Applied to a startup, red-teaming is the discipline of treating your own idea as if a very smart, very motivated competitor had been given a week to take it apart. What would they say? Where would they push? Which assumption is the whole thing balanced on?
Why your co-founder is the worst possible red-teamer
Founders default to two people for stress-testing: their co-founder and their early investors. Both are useless for this, for the same reason — shared incentives. Your co-founder needs the company to work as much as you do. Your investors have already bought in. They cannot, in good faith, run the strongest possible attack on the thing they're financially exposed to.
A real red team has two properties:
- No skin in the game on the upside. They don't benefit if the company succeeds.
- Permission to be wrong, harsh, and rude. Politeness is the death of useful critique.
Most founders have access to neither. That's the gap.
The 6 archetypal red-teamers every idea needs
1. The Competitor
"I have $20m, a team, and a head start. I see your launch on Product Hunt. What do I do this week to crush you?" This lens forces you to confront the asymmetry that's actually most likely to kill you.
2. The Regulator
"I read your landing page. Which of these claims is going to put you in front of a regulator, and in which jurisdiction?" Especially critical for fintech, health, AI, data, and anything touching minors.
3. The Defector
"I just left your company after 14 months. What did I tell my next boss about why?" Forces honesty about culture, equity, founder behaviour, and the gap between the pitch and the day-to-day.
4. The Journalist
"I'm writing a 2,000-word feature on why your company failed. What's my headline, and what are my three damning quotes?" If you can write this story plausibly, the failure modes are real.
5. The Auditor
"Show me your numbers. Now show me how you got to those numbers. Now show me the assumptions underneath." The Auditor finds the moment your model stops being arithmetic and starts being wishful thinking.
6. The Future Self
"It's three years from now. The company is dead. Write the post-mortem in the past tense." This is the classic pre-mortem, and it's the single most empirically validated technique in this entire article — Gary Klein's research shows it surfaces 30% more risks than standard planning.
A worked example: red-teaming a "Calendly for X" idea
Suppose you're building "Calendly for therapists." Here's how the six lenses tear it apart in five minutes:
- Competitor: SimplePractice already does scheduling + EHR + billing. You're a feature, not a product.
- Regulator: HIPAA. Notes, intake forms, cancellation reasons — all PHI. Your stack needs a BAA with every vendor.
- Defector: "The founders kept pivoting toward enterprise; I joined to build for solo therapists and they forgot us."
- Journalist: "How a $5/month scheduling tool accidentally exposed 4,000 therapy clients' identities."
- Auditor: "Your CAC assumption is $40. Your LinkedIn ad cost per click on 'therapist' is $11. Walk me through the funnel."
- Future Self: "We died because we were 30% better than incumbents in a category where switching cost is enormous."
None of those mean don't build it. They mean: build it knowing which fights you're walking into.
How to build your own red team in a box
The challenge with the framework above is the same as with consulting: you can know the questions and still struggle to play the parts. A red team works because the people in it are different people — with different incentives, biases, and blind spots from yours.
That's the gap THE ROAST closes. It's a deliberately diverse cast of archetypal advisors — the Competitor, the Regulator, the Skeptic, the Customer, the Auditor — that you can run your idea past on demand, with a heat dial from "constructive nudge" to "scorched earth." Think of it as a permanent, on-call Team B for whatever you're building next.
The pre-mortem checklist
- It is 18 months from today. The company has failed.
- Write the headline of the news story.
- List the top five proximate causes of death.
- For each, write what you would have done differently in month one.
- For each, decide: do that now, monitor it, or accept it.
Do this once per quarter. It is the cheapest insurance you will ever buy.