Make your AI argue

If you have used AI for a decision that mattered, you know the letdown. You lay out the situation, you ask for its best recommendation, and you get back a paragraph of careful nothing. On one hand this, on the other hand that, it depends on your priorities. Thoughtful, balanced, useless.

Here is what is going wrong, and the fix we built. Then here is a full run of that fix on a decision we are genuinely stuck on, so you can watch it work instead of taking our word for it.

A balanced answer hides the thing you needed to see

When you ask one AI to weigh it all up, you are asking it to do the averaging for you. It weighs the money against the brand against the risk, resolves the tension in its own head, and hands you the smoothed-over result. The problem is that the tension was the whole point. The spot where two good goals pull against each other is where the decision lives, and a balanced answer is the one format built to paper over it.

So we stopped asking for balance. We sit five agents down at one table and give each a single seat and a single rule: argue only from your seat. Do not be fair. Do not hedge into anyone else's concern. Make the strongest case from your corner, hand me a verdict, and name the one risk that scares you most.

The five seats

For a marketing or positioning call, the council is five fixed seats, each a different way of valuing the world. They never change. Only the decision in front of them does.

• The Brand Guardian, who cares only about long-term brand equity and will spend to protect it.
• The Performance lens, who cares only about the funnel, the cost to acquire a customer, and whether the money math closes.
• The Growth and Distribution lens, who cares only about reach and whether the idea can actually travel.
• The Audience Skeptic, who is the target customer, unimpressed, asking why anyone should care.
• The Contrarian, whose whole job is to attack the option everyone else is drifting toward.

The decision we put on the table

This is not a made-up example. It is a question we are arguing about internally right now, and it is a big one, because it steers two quarters of how we show up.

The question: for Bluhook's growth over the next two quarters, do we lead founder-first, with one person as the face, the name, the voice, the authority people follow? Or team-forward, where the studio and its fleet of agents are the hero and no single person carries the brand?

Every seat got the same briefing: who we are, who we sell to, the offer, the rules we do not break, and the one live tension. That tension is that our whole pitch is a faceless fleet of agents that runs a company, and yet the cheapest reach we have today sits on one personal account. Same facts to all five. The only thing that differs below is the seat they argue from. We have trimmed each one for length, nothing else.

Seat 1: the Brand Guardian

"Our entire pitch is that we are living proof of our own thesis: a fleet of agents, run by a general manager, building systems for other businesses. That claim is the brand asset. Founder-first distribution does not just create tension with it, it contradicts it in public. Every I-built-this post is a small, compounding admission that the team is decoration on a one-man show, which is exactly the doubt a buyer has about whether we can staff and run a system for them after the sale. A personal brand also caps at one person's bandwidth and dies the day he burns out or goes quiet. The studio brand compounds independent of any one person, and it is the only version of us that is sellable and scalable to a buyer who wants a vendor, not a parasocial relationship with a stranger's feed."

VERDICT: team-forward. Founder content is allowed only as a short bridge if reach demands it, explicitly framed as one operator of the fleet, never as the face of the company, with a dated cutover.

BIGGEST RISK: the bridge becomes permanent because it is working. Personal-brand momentum is addictive, and eighteen months on we are indistinguishable from a personal-brand consultancy while the team positioning is a dead paragraph nobody believes.

Seat 2: the Performance lens

"Founder-first wins on math alone. A faceless account starts at zero trust and has to earn a click through sheer content volume. A person starts with a face people already extend some trust to on sight. That collapses the top of the funnel: same content, faster path from impression to profile to site. Build-and-run sales are relationship sales, and relationships convert off a person, not an org chart. Nobody books a call with the fleet. They book a call with someone whose judgment they have been sampling in their feed for weeks. It is also the cheapest trust you can build, because a personal takes-and-build-log account costs only time we are already spending, while a team identity is cost paid in months before the first lead."

VERDICT: founder-first, with the team material folded in as proof posts on the founder account, not spun out as its own identity yet.

BIGGEST RISK: one hundred percent dependency shows up as a pipeline cliff, not a slow decline. If he goes dark for two weeks, lead flow from the only channel drops near zero, with nothing warmed up to catch it.

Seat 3: the Growth and Distribution lens

"Founder-first is the only lever that exists today. Distribution is not a choice here, it is a starting condition: no paid budget, thin earned reach, and one asset in play, a personal account with a playbook attached. That playbook works because a single face makes the parasocial bet possible. People follow a person's arc, argue with a person's takes, share a person's wins. A faceless we has no protagonist for the algorithm or the audience to latch onto, no one to quote-tweet, no one to book on a podcast. Team-forward content is a distribution sink until an audience already exists that is big enough that the institution itself is the draw, and we are nowhere near that."

VERDICT: founder-first, conditional. The founder account is the engine on distribution for two quarters, and team and built-work content is a feed downstream of it, narrated by him, not a parallel brand built from zero.

BIGGEST RISK: he is also the entire delivery and ops org. If growth starts working, his posting time competes directly with the fleet-running time that makes the living-proof claim true, and the flywheel stalls the moment he goes quiet.

Seat 4: the Audience Skeptic

"Founder-first is a swap-test failure at scale. Every AI studio has a guy on X posting takes about his agents. I have seen a dozen I-built-an-AI-team threads this year and I cannot tell any of them apart six months later. If he is the brand, my read is: this is one guy's hustle with a product wrapper. What happens when he takes a vacation or gets bored and pivots? I am handing my operations to a personality, not a system. Team-forward has the opposite failure only if it is vague we copy with no substance. But it is the one version that matches the pitch. You are selling me a fleet of agents doing work under a manager, so show me that. Show me the built projects, the agents doing jobs, the receipts. No competitor can point to a working fleet that runs their own company. That is the one claim only you can make."

VERDICT: team-forward, but only if it means specific, checkable proof: named agents, work you can inspect, the fleet visibly running the company. If it degrades into faceless vagueness, it is worse than founder-first.

BIGGEST RISK: I click through and find team language backed by nothing I can verify. Then I file it as theater and leave, same as I would with founder hype.

Seat 5: the Contrarian

"The room is arguing over who wears the mask, not whether the show is worth watching. Founder-first versus team-forward is a false choice, because both are the two most oversaturated moves in the current AI wave. Every dev-tool founder is doing build-in-public. Every agency is doing meet-our-AI-team. Neither is different. Both are patterns the target buyer already scrolls past. The assumption nobody is questioning is that the hero has to be a who at all. The proof point is that a fleet of agents runs a company end to end with an outcome you can measure. The content that stands out is not him talking or the agents introduced, it is the receipt: the system doing the job, unedited, with the result. That is proof-first. His face and the fleet are both just framing around that receipt, and picking one before the receipts exist is decorating a product that has not shown its punchline yet."

VERDICT: reframe. Not founder-first or team-forward. Proof-first. If forced to pick a face, he narrates, but he is the messenger, never the message.

BIGGEST RISK: everyone assumes distribution is the bottleneck. It may not be. The bottleneck may be that there is no repeatable, undeniable receipt to distribute yet. Winning the founder-versus-team debate while shipping thin proof just gets more people to scroll past a better-dressed version of nothing.

The tally, and the thing the fight caught

Here is where they landed. Two seats for founder-first, two for team-forward, one that refused the question. If you had asked a single AI, you would have gotten one of these five answers with the other four quietly averaged out of existence.

What a single answer would have buried

The vote is a tie, and the tie is not the point. Here is what the fight surfaced that no balanced paragraph would have.

First, the two camps secretly agree. Performance and Growth want founder-first, but both named the identical risk: total dependence on one person with no fallback. The Brand Guardian wants the opposite and named the same failure from the equity side. Three seats arguing opposite conclusions pointed at one shared danger. That agreement under the disagreement is the finding, and averaging would have deleted it.

Second, the Audience Skeptic and the Contrarian, coming from different seats, converged on the same uncomfortable truth: the founder-versus-team question is downstream of a question we skipped. Neither a face nor a faceless we is worth much without a receipt, a piece of our own work shown in full that a competitor could not fake. The Contrarian said it outright: we are optimizing the packaging of a product that has not shown its punchline.

So the council did not hand us a winner. It did something more useful. It told us the question we walked in with was the wrong one. The call is not who is the face. It is: build the undeniable receipt first, let the person narrate it, and let the fleet be visible exactly where it is the mechanism of the proof. That is the decision we are taking out of this, and a single agent trying to be helpful would never have told us our own question was broken.

How the machine is built

Now that you have watched one, here is the whole thing, because it is simple on purpose.

Five fixed seats, because brand equity, return, distribution, the skeptical customer, and the contrarian are tensions in every marketing call any business will ever make. The brand and the decision are the fuel poured into fixed seats. Same machine, different fuel. Each seat is a separate agent given the same briefing and one instruction: argue only from your seat, no balanced take, return a position, a verdict, and the one risk that scares you most. A human reads the disagreement and decides. The council advises and never gets the last word.

Two fences make it work. It refuses small jobs: running five agents at a single caption is theater and produces worse work, so it turns everything that is not expensive or hard to reverse away. And it never auto-decides, because an AI that quietly averages five seats back into one recommendation is just the mushy middle wearing a costume.

One more thing we hold to: we run a second council for business strategy, with different seats entirely, a hard-nosed finance chief, a wartime operator, a contrarian board member. We keep the two apart on purpose. Merge them into one all-purpose oracle and it goes right back to giving you a balanced answer about everything.

Run your own tonight, with one chatbot

You do not need a fleet. You need to stop asking for a recommendation and start demanding the argument. With any chatbot:

• Write your decision as one question with two named options. Not review our marketing. Instead: lead with A or B.
• In one chat, tell it to be the money seat and make the strongest case for A only, then a verdict and one risk. No hedging.
• In a fresh chat, do the same for the brand seat, then a fresh chat for the skeptical-customer seat. Fresh chats matter, so it cannot quietly balance across them.
• Add a contrarian chat with one job: tell me why this whole question is wrong.
• Read the five side by side and find the risk they share and the assumption they break. Then you decide.

Where this comes from

We got the specific move from Wade Foster, Zapier's CEO, describing it on the My First Million podcast: instead of asking one AI for advice, he stands up a panel of AI executives and makes them argue the call back at him. We took that idea and built the two councils above around it.

The lineage runs deeper than AI. The closest well-known ancestor is Edward de Bono's Six Thinking Hats, from 1985: hand each person in a room one fixed mode of thinking so a group stops talking in circles and covers every angle on purpose. Our seats are a decision-focused cousin of that same move.

The newer thread is about AI specifically. Researchers have found that making several agents debate a question, instead of trusting one, measurably improves the answer. If you want the technical version, the paper below is a clean place to start.

• Wade Foster on My First Million (the AI war-council idea we built on)
• Six Thinking Hats, Edward de Bono (the role-per-seat idea, 1985)
• Improving Factuality and Reasoning in Language Models through Multiagent Debate (Du et al., 2023)

That is the whole method. A balanced answer feels safe and tells you nothing. Five arguments, run on purpose and read by a person, show you the one thing the decision turns on, and sometimes they show you that you were solving the wrong problem.

We are building this in the open and telling you what we learn as we learn it. This run is live for us: we have not finished making the call. But the fight already did its job the moment it told us our question was wrong. Now it is yours.