The Problem
Early-stage founders and PMs make high-stakes decisions largely alone. Not because they're bad at their jobs, but because they don't have a full executive team to pull into a room. There's no CFO to gut-check the pricing model. No Head of Legal to flag the contract risk. No VP of Product to push back on the feature scope.
What they have instead is a Slack thread. A call with one advisor. A gut call.
Advisors help, but they have blind spots. The finance advisor sees the numbers. The product advisor sees the roadmap. Getting both of them in the same conversation, reacting to each other, at the same time? That rarely happens. Boards are slow and structured. Consultants are expensive and episodic. Neither is what you need when you're deciding whether to raise prices next quarter.
So the decision gets made with partial context. You get a Finance take but not a Legal take. A Sales perspective but not a Product one. The blind spots compound, quietly. Wrong hires. Mispriced products. Missed timing on strategic moves.
It's not that founders lack judgment. It's that good judgment needs more angles than most founders have access to.
The Proof of Concept
The Council is a demo of what it would look like if you had that boardroom.
You describe a company and ask a strategic question. Eight AI agents, each with a distinct functional expertise, deliberate in sequence. Engineering weighs in on feasibility. Finance runs the unit economics. Legal flags what could go wrong. Strategy zooms out. The conversation builds on itself, and at the end, the Council surfaces a structured decision with reasons and considerations.
I want to be clear about what this is: it's demo scope, not MVP scope. The agents here run on pre-scripted responses that are tuned to reflect realistic positions for each function. They're not hitting a live AI API with each question. The architecture supports that, but the demo doesn't need it to show the core idea.
What this is, honestly, is a structured exploration of what multi-agent deliberation could look like. Does it feel like a real boardroom? Can you actually learn something from it? Can the format hold up across different kinds of questions? Those were the things I wanted to test.
How It Works
A few design choices make the deliberation feel real rather than just fast.
Relevance-based speaking order. The agents don't all pile in at once, and they don't speak in a fixed rotation. Whoever is most relevant to the question leads. Ask about hiring a sales rep and Sales speaks first. Ask about pricing and Finance leads. This mirrors how a real meeting starts. You don't open by asking Legal to go first when the question is about go-to-market.
Context-chaining. Each agent can see what the agents before them said. Agent 3 isn't just answering the original question from scratch. They're responding to it in light of what Agents 1 and 2 already said. The conversation builds. If Engineering says something is technically risky, Finance might adjust their position. The deliberation compounds instead of just accumulating.
Follow-up awareness. Ask a follow-up and the routing re-evaluates. If you follow up with a question about marketing budget trade-offs, Marketing takes the lead on that round. The context from the previous deliberation carries over, so the agents know what's already been said. You're not starting from zero each time.
Structured decision. After all eight agents have weighed in, the deliberation surfaces a verdict, key reasons the Council aligned on, and considerations worth watching. Not a vote count. A synthesized position. The goal was to give you something actionable, not just a transcript.
Meet the Council
Eight agents. Each with a distinct lens.
Engineering
Technical feasibility, implementation complexity, operational risk
Design
User experience, visual direction, product aesthetics and usability
Product
Product roadmap, user needs, feature prioritization, market fit
Strategy
Big-picture thinking, long-term planning, competitive positioning
Finance
Budget, ROI, financial sustainability, resource allocation
Legal
Legal risk, compliance, contracts, regulatory constraints
Sales
Revenue generation, customer acquisition, pipeline and deal strategy
Marketing
Market opportunity, customer adoption, brand positioning
Every agent has a functional focus that shapes how they interpret a question. Engineering is looking at complexity and risk. Design is thinking about usability and experience. Finance is running the numbers. Legal is watching for what could blow up. The deliberation is designed so each of them pulls in a different direction, and the verdict reflects where those directions converge.
Demo vs. MVP Scope
What the demo does:
- Pre-scripted responses tuned per agent and per question type
- Three sample companies, three preset questions per company
- Follow-up question routing via keyword scoring
- Context-chaining built into the response scripts
This is enough to show the shape of the idea. You can see the deliberation format, feel the agent voices, and understand what a structured decision looks like coming out of the other side.
What an actual product would do:
Live OpenAI inference per agent, each with a distinct system prompt that enforces their functional perspective. Finance would always be running numbers. Legal would always be flagging risks. The persona would be baked into the prompt, not just implied by a label.
Persistent sessions and decision history. Save a deliberation, come back to it, continue from where you left off. A decision log over time. "Here's what The Council said in March about pricing. Here's what changed."
Custom company context. You describe your own company, not a preset. The agents respond to your actual situation.
A custom agent roster. You define the roles that matter for your org. If you have a Chief Risk Officer instead of a Legal VP, you configure that. If you want to add a Customer Success agent, you add one.
Team collaboration. Multiple people watching the deliberation together, reacting to individual messages, building on the conversation as a team.
Decision export and sharing. Send the deliberation to someone who wasn't in the room. Let them catch up without a summary.
The goal was always a tool that makes multi-angle thinking accessible to people who usually have to think in a straight line. Not to replace judgment, but to surface the questions you might not have thought to ask.