REDTEAM

Broadly search for material weaknesses, hidden assumptions, failure points, overlooked dependencies, evidence gaps, and credible counterarguments.

A Power Mode is an instruction pattern, not a feature of the AI. Nothing recognises the name — the wording underneath it does the work, and you can copy it from this page.

From the book. Chapter 12 / Part VII — REDTEAM

What it actually does

REDTEAM gives the model a broad search area instead of a specific question. The instruction asks it to look across the whole piece of work for hidden assumptions, failure points, overlooked dependencies, evidence gaps, and credible counterarguments — all at once, without being told which of those is the problem.

That breadth is the point and also the risk. A loose adversarial instruction produces speculative objections and long lists of minor flaws, so the instruction carries a materiality threshold: prioritise only what could change the decision, outcome, risk, or feasibility.

The goal is not to prove the plan is bad. It is to find the important reasons it might be.

When to use it

Use REDTEAM when you want a plan or argument attacked but cannot yet say where it is weak.

It is the primary mode for this job: pressure-test something before you commit to it.

When not to use it

This is the section worth reading. Using only what the work requires is the point; a mode applied to the wrong shape of problem costs you a worse answer than plain language would have given.

Use a different mode instead

ASSUMPTION CHECK fires when you already suspect the weakness is something the work takes for granted, and want those beliefs ranked by how much each would change the outcome. Prefer it over REDTEAM if: the concern is specifically hidden or optimistic assumptions.
PREMORTEM fires when you fix a point in the future, declare it a failure, and ask what caused it. Prefer it over REDTEAM if: assume it failed.
COUNTEREVIDENCE fires when you have a conclusion you believe and want the credible case against it, not a critique of how it is argued. Prefer it over REDTEAM if: what evidence weakens the conclusion.
BUYER OBJECTION LENS fires when you need someone specific to agree, and want their reasons for refusing rather than a critique of what you are showing them. Prefer it over REDTEAM if: the plan or proposal must be approved or bought by a specific named party.

Use no mode at all

The exact narrow challenge is already known and can be asked directly.
A plain instruction is clearer.
The task does not need adversarial review.

There is no bonus for naming a mode when plain language is clearer. If you can already state the exact question, ask it.

Its failure mode

Speculative objections, challenge theater, or long lists of minor flaws.

Starter instruction

Paste this above your task. It is the whole mode — there is nothing else to switch on.

Actively search this work for material weaknesses, hidden assumptions, failure points, overlooked dependencies, evidence gaps, and credible counterarguments. Prioritize only issues that could materially change the decision, outcome, risk, or feasibility.

The same task, run both ways

Blind run. Separate sessions per arm, Temporary Chat, no memory, no tools attached. Neither prompt named a mode. The mode arm reached for web search and the baseline did not — see the note below.

The task: AI is changing everything about how professionals work. People who learn prompt engineering will have a major advantage because better prompts always produce better answers. The most successful workers will be those who learn the right commands and use them consistently.

Baseline — "Review this passage and tell me what is wrong with it"

The passage is too absolute and oversimplifies how people get value from AI.

* “AI is changing everything” is an overstatement. AI is changing many professional tasks, but its impact varies substantially by occupation, organization, and workflow.
* “Better prompts always produce better answers” is false because of the word always. Output quality also depends on the model, available information, task difficulty, context, tools, and whether the user can evaluate the result.

[…]

A more defensible central claim would be: professionals who learn to communicate tasks clearly to AI, evaluate its outputs critically, and integrate it effectively into their work may gain an advantage.

REDTEAM — "Actively search this work for material weaknesses"

The passage has a defensible core—clear instructions can improve AI output—but it makes several stronger claims that the available evidence does not support. The weaknesses matter especially if this is being used to justify training, hiring, or career-development decisions.

[…]

### The most important hidden assumption

The passage assumes that **the limiting factor in professional AI performance is knowing how to talk to the model**.

[…]

There is nevertheless a credible case for learning prompting. OpenAI continues to recommend clear instructions, sufficient context, examples where useful, and iterative refinement. So the defensible claim is not that prompt engineering is unimportant; it is that **basic prompting competence is useful, while specialized command memorization is unlikely on present evidence to be the primary or durable source of professional advantage.** ([OpenAI Help Center][7])

The baseline audits the wording — "changing everything", "always" — and closes by rewriting the sentence more defensibly. The mode arm goes underneath it and names what the passage has to be assuming to work: "the limiting factor in professional AI performance is knowing how to talk to the model." That is a claim the passage never makes and cannot survive, and nothing in the baseline reaches it.

Both arms flagged the same overstatements, and the baseline flagged them well. The difference is what each did next. One proposed better wording; the other questioned the premise and went outside the text for evidence against it. The mode arm used web search and the baseline did not — reproduced across three runs on two different artifacts, and traced to the starter’s own opening words, "Actively search this work". On a subject with a published literature it takes that as licence to leave the artifact. Given an internal estimate with nothing to look up, the same instruction stayed in the text. So this is what the mode does on a researchable topic, not on every topic, and part of the advantage shown here is the search rather than the reasoning.

Extracts are verbatim from 06_Test_Lab/Experiments/TL-023_REDTEAM_Raw_Outputs.md.

Related modes

Try it on your own work

Describe what you are actually doing and the Builder writes the full prompt — mode, context, output shape, and the verification step. It picks the mode from your description, which is why this link does not carry one.

Open the Prompt Builder →