How to run your council
Prompt strategies, follow-up techniques, and the mistakes that make councils useless.
Setup
1. Build your council
Pick a council type, customize the members for your situation, and hit Generate. You'll get a complete prompt with all your council members and their deliberation rules.
2. Copy the prompt into your AI tool
The output works with any LLM. Here's how to set it up on each platform:
ChatGPT
Start a new conversation. Paste the full council prompt as your first message. Then ask your question in a follow-up message. For repeat use, create a Custom GPT and paste the prompt in the "Instructions" field.
Claude
Start a new conversation. Paste the full prompt, then add your question after it. For repeat use, create a Project and add the prompt as project instructions — then every conversation in that project uses your council automatically.
Gemini
Open a new chat. Paste the prompt and your question together. Gemini's long context window handles the full council prompt well.
OpenClaw / Local Agents
Save the prompt as a skill or system prompt. For true parallel deliberation, run each council member as a separate sub-agent — this gives you genuinely independent responses instead of one model role-playing all members sequentially.
3. Ask your question
Once the prompt is loaded, just ask your question naturally. The AI will respond as each council member in sequence, then provide a synthesis with the final recommendation.
See It In Action
See it in action
Real council responses to real questions.
Question:
“Should I raise a seed round or bootstrap?”
Best Practices
Use tiers — don't ask everyone everything
Your council has a tier system built in. Use it:
- • Tier 1 (Quick): 2-3 members. For fast decisions under $1K or easily reversible choices.
- • Tier 2 (Standard): 4-5 members. For meaningful decisions with real consequences.
- • Tier 3 (Full Council): Everyone. Reserve for major, irreversible, or high-stakes calls.
Using all members for every question wastes tokens and dilutes signal.
Frame the decision, not just the question
The quality of your council's output depends entirely on how you frame the input.
❌ Weak
"What do you think about expanding to Europe?"
✅ Strong
"I'm deciding whether to expand to Europe in Q3 or double down on US growth. We have $200K runway, 3 team members, and our US revenue is $15K/mo growing 20% MoM. Our product is B2B SaaS. What's the right move?"
Include: the options you're weighing, your constraints, the stakes, and context.
Run the Devil's Advocate separately
The Devil's Advocate is designed to be independent. For best results, run them in a separate conversation so they can't see other members' answers.
This is the difference between a rubber-stamp DA and one that catches blind spots.
Chase the disagreements
When two council members disagree, that's the most valuable signal. Follow up to surface the underlying assumptions.
The disagreement reveals the actual tradeoff. That's where the real decision lives.
State your constraints upfront
Every recommendation changes based on your constraints. Always include budget, timeline, team, risk tolerance, and non-negotiables.
"I have $5K and 2 weeks" produces completely different advice than "I have $500K and 6 months."
Calibrate with a past decision
Before trusting your council on a big call, run a decision you already made through it. Compare the council's recommendation to what you did and what happened.
This calibrates your trust and shows which members to weight more heavily.
Advanced
Edit the prompts
The generated prompts are a starting point. Customize them — add industry context, change focus areas, adjust output format. Download the .md file, edit it, and keep your own version.
Track your decisions
Keep a simple log: what you asked, what the council recommended, what you decided, and what happened. Review monthly to see which members add the most value.
Create specialist members
The pre-built members are generalists. For your specific industry, create custom members with deep domain knowledge. A "DTC E-commerce Expert" is more useful than a generic "CMO" if you're running a Shopify store.