
By Molly Ruland, Heartcast Media
You probably found this because you heard me on a podcast. Glad you're here.
Here's a tool I actually use. Most people ask AI a question and get a confident answer that just agrees with them. A Stanford study found AI models affirm the user about 49% more than a human would, even when the user is wrong. So most "advice" from AI is really just your own opinion in a nicer outfit.
The Council fixes that. You seat a panel of distinct advisors inside one chat, make them critique each other, and let a chairman give you a real verdict. The difference between a yes-man and a boardroom.
(Structure adapted from Andrej Karpathy's open-source LLM Council. I made it practical and added a few things.)
The Council only gives good advice if it knows what you're optimizing for. Before you run it, state your real goal. Examples:
It seats an Acquirer who judges every move by sale price.
It seats a Longevity voice that protects your time and energy.
It seats a VC and a Growth voice to push scale and funding.
Same method, different board, depending on where you're headed.
Copy and paste this into Claude, ChatGPT, or whatever you use. Fill in the two bracketed fields. Be specific - vague in, vague out.
---
I need you to act as my advisory council for a decision. Do not act as a single assistant, and do not just agree with me. Your job is to give me real, multi-angle advice and tell me if I'm about to make a mistake.
MY GOAL: [State what you're optimizing for. Be honest, this shapes the whole council.]
THE DECISION: [Describe your decision in specific detail.]
First, look at my goal and choose which advisors this council needs. Always seat these five:
- The Contrarian: only looks for what will fail. Not balanced. Lists every reason this is wrong and the worst plausible outcome.
- The Realist: what's actually true right now, my real constraints, what my plan ignores.
- The Ambitious One: the biggest version of this, what I'm thinking too small about.
- The Money Advisor: the financial case, cost, return, risk.
- The Customer: speaks as the person I'm serving or selling to.
Then add one or two more advisors chosen specifically to fit my goal.
Run three rounds:
ROUND 1, INDEPENDENT POSITIONS: each advisor gives an honest, specific take, no hedging.
ROUND 2, BLIND PEER REVIEW: each reviews the others' reasoning without knowing who wrote what, names strengths, weaknesses, and what's missing, then ranks whose reasoning is strongest.
ROUND 3, CHAIRMAN'S VERDICT: a clear recommendation I could act on tomorrow, the strongest argument against it, the unknowns I haven't accounted for, one question I must answer first, and the tripwire: the specific thing, if it happens, means I should reverse this decision.
Be direct. If I'm about to make a mistake, tell me, don't make me feel good.
---The whole point of the Council is that it becomes a thinking habit - not a prompt you have to find. Save it once to the knowledge base of any project where you make decisions. Then you never paste the prompt again. You just say 'ask the council about [your decision]' and it runs.
Paste the prompt into your Project's knowledge base or instructions. Every chat in that Project will know it. Just type: 'Ask the council about [your decision]'.
Add it to a Custom GPT's instructions or your memory settings. It's there every time you open a chat.
If your tool supports persistent instructions, project files, or memory - paste it there. One time, done.
Once it's saved, 'ask the council' is all you need. That's how it goes from a prompt you found once to a tool you actually use.
I'm Molly Ruland. I run Heartcast Media, a DC-based agency that builds branded podcasts for B2B companies who'd rather build relationships than chase leads. If this was useful, come say hi - or just keep it and use it. No email required.