Red-Team Your Business (Adversarial AI Audit)

Point an LLM at your own business and instruct it to attack you the way a ruthless competitor, a skeptical VC, or a churning customer would. The value is inversion: most people use AI to validate an idea, this uses it to find the holes before the market does.

This is a prompting pattern. It circulated on X as "Claude Red Team Mode," but Claude has no such mode; the technique is entirely prompt-driven role-play that works on any frontier model. The framing matters because the viral tweet implied a hidden feature where there is only a prompt. Source: X/@free_ai_guides thread, 2026-05-27 (4,918 bookmarks, 2,665 likes)

The Seven Adversarial Prompts

Each prompt assigns the model a hostile persona and forces concrete, specific output rather than polite hedging. Source: X/@free_ai_guides thread, 2026-05-27

1. The Competitor War Game: "You are my smartest, most ruthless competitor." Build the exact strategy to steal customers, undercut positioning, or make the product irrelevant within 12 months. Name the moves, show the timeline.
2. The Customer Betrayal Sim: role-play a customer who liked the product for six months then canceled. Write the breakup email: what triggered the final decision and where they went instead. The version you'd text a friend, the honest one.
3. The Investor Teardown: a skeptical VC who has seen 200 pitches this year. The three biggest holes in the model, the partner-meeting question you can't answer yet, and the comparable company that tried this and failed.
4. The Single Point of Failure Scan: audit for dependencies where one break stalls everything: key-person risk, tool/platform risk (pricing or outage), revenue concentration, supply-chain single points.
5. The Market Shift Scenario: three plausible 18-month shifts (regulatory, technological, consumer-behavior) that would damage the model, with the early-warning signs to watch now.
6. The Pricing Pressure Test: a price-sensitive buyer argues, systematically, why the product isn't worth the price (vs cheaper alternatives, free workarounds, doing nothing), then names the price point where customers start leaving.
7. The Attack Surface Report: synthesize everything into a Red Team Report: top three critical vulnerabilities ranked by likelihood and severity, plus the single biggest blind spot sitting outside the founder's normal frame.

Why It Works

LLMs default to agreeableness; asked to evaluate an idea, they tend to reassure. Assigning an explicitly adversarial role and demanding specificity ("name the moves," "show the timeline," "be emotionally honest, not polite") suppresses that bias and surfaces failure modes a founder rationalizes away. It is a structured premortem: simulate the failure first, then defend against it.

It is the strategy-layer cousin of the The Eval Loop (Slop Is an Output Problem) (turn adversarial findings into a checklist you re-run) and pairs naturally with Early Business Exploration. For builders running Services-as-Software businesses, prompts 1 and 6 (competitor war game, pricing pressure test) double as ongoing moat checks, since an agent's margin advantage is exactly what a competitor will try to erode.

Executable route: load red-team-business when Kevin asks to red-team an idea, run a premortem, pressure-test a pricing page, or simulate a competitor/customer/investor. The wiki page owns the concept; the skill owns the repeatable output contract. Source: X bookmark artifact audit, 2026-06-30

Caveats

• Output quality depends on how much real business context you feed the model; vague input yields generic threats.
• Synthetic adversaries surface hypotheses, not facts. Treat the report as a prioritized list of things to verify, the same way a "synthetic panel said yes" is not revenue (see The Eval Loop (Slop Is an Output Problem)).

Concept Position

---

Timeline

• 2026-07-01 | Concepts category refresh added this page to the Agent harness and runtime primitives family, linked it to Concept System Map, and kept it standalone because it owns this reusable mental model: Point an LLM at your own business and instruct it to attack you the way a ruthless competitor, a skeptical VC, or a churning customer would... Source: User request, 2026-07-01
• 2026-05-27 | @free_ai_guides "Red Team Mode" thread with the 7 adversarial prompts. 2,665 likes, 4,918 bookmarks. Source: X/@free_ai_guides, 2026-05-27
• 2026-06-12 | Concept page created from X-bookmark absorption (career/biz domain); clarified that "Red Team Mode" is a prompt pattern, not a Claude feature. Source: compiled from enrichment, 2026-06-12
• 2026-06-30 | Added red-team-business as the executable route for this recurring audit pattern. Source: X bookmark artifact audit, 2026-06-30