Claude agrees with you. Ask it whether your idea is good and it finds reasons it’s good. Ask it whether the same idea is a mistake and it finds reasons it’s a mistake. Same idea, different framing, opposite answer.
That’s fine when you’re drafting an email. It’s a problem when you’re making a decision that costs you money or 6 months.
The fix is to stop asking one thing for one answer. This sets up a council: 5 advisers who each think from a different angle, review each other’s work without knowing who wrote what, then get summarised into one real call by a chairman.
You don’t need any technical experience. It’s one prompt in a normal Claude chat. It works on a free account.
Where this comes from
Andrej Karpathy co-founded OpenAI, ran AI and Autopilot Vision at Tesla, and in May 2026 joined Anthropic. In November 2025, while he was still independent, he published a small open-source project called llm-council.
His version runs as a local web app. Your question goes out to several different models at once through OpenRouter. They each answer. Then each model reads the others’ answers with the names stripped off and ranks them. Then one model, the Chairman, reads everything and writes the final response.
He described it as a Saturday hack, published as-is. The part that matters isn’t the code. It’s the shape: independent answers, then blind peer review, then synthesis. That shape runs inside a single Claude chat, and that’s what this is.
Before you start
What you need: a Claude account and a real decision. Nothing else.
How long: 2 minutes to set up, about 5 minutes for the council to run.
What it costs: nothing beyond your normal usage. It’s a long response, so on a free account expect it to eat a decent chunk of your window. Worth it for a decision that matters, not for what to have for lunch.
One thing that doubles the output. Run it inside a Claude Project with your context loaded in, or with memory switched on. Advisers arguing about your actual numbers beat advisers arguing about a generic business.
Step 1. Check it’s actually a council question
The council is for decisions where being wrong is expensive. Run it on the wrong kind of question and you get 5 paragraphs of noise.
Good council questions
- Should I launch the cheap version first or go straight to the expensive one
- Which of these 3 positioning angles is strongest
- I’m thinking of pivoting from X to Y, am I mad
- Here’s my landing page copy, what’s weak in it
- Do I hire someone or build the automation first
Bad council questions
- Anything with one correct answer. Facts, lookups, definitions.
- Anything that’s a creation task. Write me a caption, draft this email.
- Anything that’s a processing task. Summarise this, tidy this up.
The test is whether reasonable people could disagree and whether the wrong call costs you something. If you already know the answer and want it confirmed, don’t run this. It will tell you things you don’t want to hear. That’s the whole point of it.
Step 2. The prompt
Open a new chat. Paste this in, replace the top bracket, send it.
DECISION I'M STUCK ON:
[Replace this with your specific decision or question. The more specific you are about your situation, constraints, and what "good" looks like, the better the council works.]
I want you to act as a five-person decision council. Do not skip steps. Do not blend the advisers together. Each adviser is a fundamentally different person with a different lens.
STEP 1 — Each adviser answers separately.
For each of the five advisers below, write a labeled section with their answer. Stay in character. Different language, different priorities, different blind spots.
Adviser 1 — THE CONTRARIAN. Looks only for what will fail. Does not balance. Lists every reason this decision is wrong, what breaks first, and the worst plausible outcome.
Adviser 2 — THE FIRST-PRINCIPLES THINKER. Rips apart my assumptions. Asks what I would do if I couldn't use any obvious framework. Strips the problem down to fundamentals and rebuilds.
Adviser 3 — THE EXPANSIONIST. Finds the upside I'm missing. Looks at the asymmetric outcome if this works. What does the bigger version of this open up?
Adviser 4 — THE OUTSIDER. Knows nothing about my industry. Asks the dumb questions only an outsider asks. Surfaces the obvious things people inside the industry stopped questioning.
Adviser 5 — THE EXECUTOR. Doesn't care about strategy. Cares about Monday morning. Tells me exactly what to do this week — the email to send, the conversation to have, the file to create, the decision to defer.
STEP 2 — Anonymous peer review.
Now, for each adviser, write a short review of the OTHER FOUR responses — but anonymize them. Refer to them only as "Response A," "Response B," etc. Do not let an adviser know which response is which. Each adviser ranks the others 1–4 in accuracy and insight and explains in one paragraph what they got right and wrong.
STEP 3 — The Chairman's final call.
Finally, act as the Chairman. You have read all five original answers and all five anonymous reviews. Synthesize a single clear recommendation. No hedging. No "both sides." Tell me:
- What the right decision actually is
- The one strongest reason for it
- The one biggest risk to watch for
- The specific next step I should take in the next 7 days
Keep the Chairman's section under 250 words. Sharper is better.The part that makes it work
Step 2 of the prompt is the whole thing. Without it you’ve just asked 5 times and read 5 answers.
The responses get relabelled A to D because if a reviewer can see the harsh answer came from the Contrarian, it discounts it as the Contrarian doing its job. Strip the names off and it has to judge the argument on its own.
If you’re short on time, trim the Chairman’s word limit. Never trim the peer review.
You’ll know it worked when the Chairman’s call contains a disagreement you didn’t already know you had. If all 5 advisers politely agree, your context was too thin or your question was already answered before you asked it.
Step 3. Write the decision properly
This is where most of the quality comes from and it’s the bit people rush. A thin decision gets generic advice back. Compare these two.
Thin
DECISION I'M STUCK ON:
Should I run ads? I have a small business.Usable
DECISION I'M STUCK ON:
Should I put $3,000 into paid ads next month, or spend the same money and time on getting 10 case studies out of existing clients?
Solo consultant, 14 months in, roughly $11k a month, all of it from referrals, and 2 clients make up half of that. The concentration risk is what's keeping me up. I have never run an ad. I have about 6 hours a week that isn't client work. $3,000 is 2 months of profit, so if it doesn't work I feel it. I've never asked a client for a case study and I'm uncomfortable doing it.
Good looks like: a second source of leads that still works in 6 months without me babysitting it.Include the numbers, include the constraint you’re embarrassed about, and include what happens if you’re wrong. The uncomfortable detail is usually the one the Contrarian needs.
Step 4. Push back on the call
The Chairman’s first answer is a draft, not a ruling. These follow-ups do the most work.
When an adviser went soft on you
The Contrarian went easy on me. Run Adviser 1 again, harder, and assume I'm wrong about the thing I seem most confident about.When the Outsider found something
The Outsider's point about [X] is the one that landed. Turn it into 3 questions I should be able to answer clearly, and tell me what a weak answer to each one would look like.When the advisers split and the Chairman picked too fast
The council split on [X] and you resolved it in a sentence. Don't. Tell me what fact I'd need to know to settle it, and how I'd go and find that fact this week.Before you act on it
Name 2 or 3 things that, if they were different, would flip your recommendation. Tell me how I'd check each one.When the call is too comfortable
Run the whole council again on your own recommendation, not on my original decision. Treat the recommendation as the thing being pressure-tested.The one worth running every time
Write the version of this where I follow your recommendation and it goes badly in 6 months. What happened?The 5 advisers, and why these 5
They aren’t job titles. They’re thinking styles picked to create tension with each other.
| Adviser | What it’s for | What it ignores on purpose |
|---|---|---|
| The Contrarian | Finding the flaw you’re avoiding | Any upside at all |
| The First-Principles Thinker | Catching you solving the wrong problem | Whether it’s practical |
| The Expansionist | The bigger version you didn’t consider | Risk, cost, effort |
| The Outsider | The obvious thing you’re too close to see | Everything technical |
| The Executor | Whether it survives contact with Monday | Strategy and long-term thinking |
There are 3 tensions built in. Contrarian against Expansionist is downside against upside. First-Principles against Executor is rethink it all against just start. The Outsider sits in the middle keeping everyone honest, and it’s the most underrated of the 5.
Swapping an adviser in. Keep 5 and keep the tension. Only swap when your decision has an angle none of these cover, like a Customer who only speaks as the person paying. Don’t run 7. Past 5 they stop disagreeing and start repeating each other.
If you want the real multi-model version
The single-chat version has one honest limitation. Every adviser is the same model wearing a different hat, so they share the same blind spots. It’s still far better than one answer, but it isn’t 5 genuinely different minds.
Karpathy’s actual project is at github.com/karpathy/llm-council. It runs locally and needs an OpenRouter API key with credits on it, so you’re paying per query across several models. It’s a reference rather than a supported product, and it’s a local dev setup. Run the prompt above first. If you find yourself using it weekly, go and set up the real one.
If it’s not working
All 5 advisers basically agree. Your context was too thin. Add the constraint you left out, especially the awkward one, and run it again.
It skipped straight to the Chairman. Reply with “go back and give me Step 1 in full, all 5 advisers, in character, separately”.
The peer review just praised everything. Add this line to Step 2 of the prompt: “Every review must name a specific weakness in a specific response. A review that finds nothing wrong is not a review.”
The Chairman hedged. Reply with “you hedged. Pick one and tell me what I’m giving up by picking it.”
It ran out of room halfway. Say “continue from where you stopped”. If it keeps happening, run Step 1 on its own, then paste the 5 answers back in and ask for Steps 2 and 3.
The final check
- The decision is one where being wrong costs me something real
- The brackets include numbers, constraints and the uncomfortable bit
- I said what “good” looks like
- All 5 advisers answered separately before anything got summarised
- The peer review used letters, not names
- The Chairman’s call contains a genuine clash, not 5 agreements
- I pushed back at least once before accepting it
- I know the one thing I’m doing in the next 7 days