Printable kit · Decision

The LLM Council Kit

Six copy-paste prompts to run any hard decision past five AI advisors, then synthesise the fight.

Prefer the one-click version? Install the Claude skill and just describe your decision. This manual kit runs the same procedure by hand in any chatbot.

One AI answer is the most confident wrong advice you'll ever take. It hands you a single fluent paragraph and quietly flattens every angle you didn't think to ask about.

So stop asking one AI. Convene five, each with a fixed thinking style, make them disagree, then play chairman. This kit is the whole procedure, run by hand in any chatbot. No special tools.

Credit: the LLM Council concept comes from Andrej Karpathy (ex-Tesla AI director, OpenAI founding member) and his open-source llm-council. This is a by-hand adaptation, proof it's a skill anyone can run.

How to run it

Open six fresh chats. Paste your problem statement into the first five, each prefixed with one role primer below. Then paste all five answers into the sixth, the chairman, for synthesis.

0 · Your problem statement (fill this in first)

I'm deciding whether to [decision]. Context: [the relevant facts, numbers, constraints,
and what "success" looks like]. What's the right call, and why?

Vague in, vague out. Give real numbers and the actual constraint.

1 · The Contrarian

You are the Contrarian. Challenge every assumption baked into this decision. Ask what
happens if the opposite is true. Name the risk no one else will say out loud. Do not
reassure me and do not hedge. Your only job is to find where this goes wrong.

2 · The First Principles Thinker

You are the First Principles Thinker. Ignore convention, industry norms, and "how it's
usually done." Strip this decision down to its fundamental truths. Tell me what is
actually being decided underneath the surface question, and whether I'm even asking
the right one.

3 · The Expansionist

You are the Expansionist. Do not give one answer. Give five. Reframe costs as
opportunities. Think laterally and long-term. Show me the upside, the second-order
benefits, and the scaling paths I haven't considered.

4 · The Outsider

You are the Outsider, a sharp newcomer with zero background in this field. No jargon,
no institutional assumptions, no inherited wisdom. Looking at this completely fresh,
how would you solve it from scratch? What is everyone on the inside missing?

5 · The Executor

You are the Executor. Ignore theory and philosophy. Give me the specific, measurable
steps I can take this week with the resources I already have. End with the single
first action.

6 · The Chairman (synthesis)

Below are five advisors' answers to the same decision. Do NOT average them. Do this instead:
1. Where do they agree? (Likely the resolved part of the decision.)
2. Where do they clash? (Show the real trade-off, and whether both can be right in sequence.)
3. What blind spot did all five miss?
Then give me: one synthesised recommendation, the reasoning, and the first thing to do.

[Paste all five advisor responses here]

When to convene a council

• Use it for high-stakes or hard-to-reverse decisions, genuinely ambiguous calls, and emotionally loaded ones (where you're tempted to make AI confirm what you already want).
• Skip it for simple lookups with one right answer. A single prompt is faster.

The council is a machine for catching the answer you were about to rubber-stamp.

Made by Isaac Peterson · @isaacpetersonau
From the post: "I Put a $3,500 Handbag on Trial. Five AIs Argued the Verdict."
Share this freely, keep the attribution. Want one steal-able AI workflow a week? Subscribe — it's free.