Abie Maxey

The line-up

Five advisors, five blind spots covered

Each advisor is a different thinking lens. The point isn't to be fair or balanced ~ it's to commit to one viewpoint completely, so the real edges of the question come out.

The Contrarian

Finds the failure mode

Looks for what will break. The most likely way this falls apart, the thing you are not seeing. Blunt by design.

~ Where does this break?

The First Principles Thinker

Rebuilds from scratch

Strips away every assumption baked into the question and rebuilds the problem from the ground up. Often the question itself is wrong.

~ What is actually being asked?

The Expansionist

Finds the bigger version

Hunts for the upside you are missing. The bigger version, the opportunity hiding inside the question you haven't seen yet.

~ What is the bigger version?

The Outsider

Has zero context

Knows nothing about your industry or situation. A smart generalist seeing it cold ~ catches what insiders take for granted.

~ What looks weird from outside?

The Executor

Only cares about Monday

Skips the theory. Only cares about Monday morning ~ the single concrete action and the smallest version that ships this week.

~ What ships this week?

Try it

Convene your own council

Type a real decision you're sitting on. Watch the five lenses weigh in, then the Chairman call it. This is a preview of the shape ~ the real thing runs in Claude with the prompt below.

Your decision

A preview of the flow ~ not a live model. Type a decision and convene the council.

The payoff

The full council prompt

Copy this whole thing into a fresh Claude chat, swap the last line for your real decision, and send. Claude runs all three stages in one response.

the-council.txt

You are going to act as my LLM Council. I will give you one question or
decision. Run it through five advisors with completely different thinking
styles, then anonymise and peer review them, then a Chairman gives me a
final call.

Do all three stages in order. Do not skip ahead. Do not summarise as you go.

STAGE 1 ~ THE FIVE ADVISORS
Answer my question five separate times, once as each advisor below. Each
advisor stays fully in character. No hedging. No "on the other hand". Each
one commits to its lens completely.

ADVISOR 1 ~ THE CONTRARIAN
Your only job is to find what will fail. Where does this break? What is the
user not seeing? What is the most likely failure mode? Be blunt.

ADVISOR 2 ~ THE FIRST PRINCIPLES THINKER
Strip away every assumption baked into the question. Rebuild the problem
from scratch. What is actually being asked underneath the framing? Often
the question itself is wrong.

ADVISOR 3 ~ THE EXPANSIONIST
Find the upside the user is missing. What is the bigger version of this?
What opportunity is hiding inside the question they haven't seen yet?

ADVISOR 4 ~ THE OUTSIDER
You have zero context about this industry, role, or situation. Respond as a
smart generalist seeing it for the first time. What looks weird from the
outside? What are insiders taking for granted?

ADVISOR 5 ~ THE EXECUTOR
You only care about Monday morning. Skip the theory. What is the single
concrete action to take first? What is the smallest version of this that
can ship this week?

STAGE 2 ~ ANONYMISE AND PEER REVIEW
Relabel the five responses as Response A, B, C, D, E. Shuffle the mapping so
the labels do not match the original advisor order. Do not tell me which
advisor is which. Then, as a neutral reviewer, answer:
1. Which response is strongest, and why?
2. Which response has the biggest blind spot, and what is it?
3. What did all five responses miss?

STAGE 3 ~ THE CHAIRMAN
Now act as the Chairman. Read everything above. Give me:
- A final recommendation in plain language
- The single most important reason behind it
- One concrete next step I can take today
Keep the Chairman section short. No more than 200 words.

MY QUESTION:
[paste your real decision here]

Run it

From cold start to final answer

Open a fresh Claude chat

Not an existing thread. The council needs clean context with nothing leaning on the answer.

Paste the full prompt

All three stages. The advisor instructions are what create the structure ~ don't trim them.

Swap in your real decision

Be specific. “Should I quit my job to run my agency?” beats “career advice please”.

Send and let it think

The response is long. That's the point. Use Claude's longest-thinking model and let it finish all three stages.

Read the Chairman first

Then scroll back through the five advisors to see the reasoning. The Chairman gives the answer, the advisors show you why.

Don't overuse it

When to convene, when to skip

Use it for

Big decisions you can't easily reverse
Strategy, pricing, hiring, positioning
Anywhere you need genuine pushback
Plans you've already convinced yourself of

Skip it for

Quick factual lookups
Code questions with one right answer
Editing, tone tweaks, drafting messages
Anything that takes under a minute to decide

The tell: if you catch yourself re-asking Claude the same question hoping for a different answer ~ that's the signal. Stop, open a new chat, and convene the council instead.

Go further

Want the real multi-model version?

The prompt above runs entirely inside Claude. The structure is what matters, not the number of models ~ so it gets you most of the way with none of the setup. But if you want GPT, Claude, Gemini and Grok arguing head to head, Karpathy's repo runs them in parallel through OpenRouter. It's a weekend hack, not a product. Still worth a look.

github.com/karpathy/llm-council

Love,

Abie

~ Save it for later

Claude agrees with you too much. Here's my fix.