Red team

The practice of deliberately assigning someone (or something) the job of attacking your plan to find its weak points before reality does. The term comes from military and security work. With AI, you can run a red team on a decision in minutes, and it has no ego or career risk to protect.

What it is

A red team's whole purpose is to argue against you in good faith: poke the assumptions, run the premortem ("it is a year later and this failed, why?"), make the strongest opposing case. The point is not to be negative, it is to surface the failure modes a team of people who all want the deal to happen will quietly skip over.

Why CEOs care

Because the people around you have incentives. They want the bonus, the win, the not-rocking-the-boat. An AI assigned the red-team role has none of that, so it will ask the uncomfortable question before you wire the money. The catch: models tend to agree with you by default (a tendency called sycophancy), so you have to explicitly assign the adversarial role, or you just get a cheerleader. Used right, it is a checklist of risks, not a verdict.

Where you'll see it

In the pressure-test-a-decision workflow, where you hand AI a one-page decision memo and make it argue the other side.

Example

Before a big acquisition, a CEO has Claude run a premortem and argue the strongest case against the deal. It surfaces three assumptions the deal rests on, two of which nobody had stress-tested.

Related

• claude-code