I have a chat I go to when I don't want a yes.
Not a prompt. A room.
I call it the Board Room. Six seats. Six different jobs. I bring a decision. They do not bring me comfort.
Here is why I built it.
One AI will sound smart about almost anything. Ask it well and it will make your idea sound better than it is. That is useful. That is also dangerous.
I don't need a louder version of my own thinking. I need friction before I commit.
So I split the thinking into seats.
Chair keeps the question tight, calls on the seats, and writes the final memo. No memo, no decision. If the room cannot sum it up clean, I am not ready.
Chair is also the referee when two smart answers collide. That happens a lot.
What does it cost? What is the worst case? Can we actually afford to be wrong?
Finance does not care if the idea is exciting. Finance cares if it survives a bad quarter, a bad bid, or a bad assumption. If the numbers are missing, Finance names them. That list alone is worth the exercise.
Skeptic makes the case against
Not "play devil's advocate." Make the strongest case that this fails. Find the blind spot. Find the thing I am refusing to look at.
Skeptic has saved me from my favorite ideas. The good ones survive Skeptic. The fragile ones thank Skeptic early, before they get expensive.
Who is affected? Will we be proud of this later? Does this fit what we say we are about?
A decision can pencil and still be wrong. Mission is the seat that refuses to let "it works" be the whole answer.
Operator asks if it can actually get done
By whom? By when? What is step one on Monday morning?
Operator has no respect for a beautiful plan with no owner. If nobody owns it, it is not a plan. It is a wish with bullet points.
Counsel asks what has to be in writing
What could blow up? What needs a signature, a second set of legal eyes, or a hard stop before anyone commits?
Important: Counsel is not a lawyer. The seat flags risk. Real lawyers do law. Knowing the difference is part of the point.
That is the room. Chair, Finance, Skeptic, Mission, Operator, Counsel.
Here is how a run actually goes.
I bring one decision. One. Not five.
Each seat answers through its own lens. Then each seat votes: Go, Go if, or No-go.
"Go if" is where the truth lives.
Go if the price is capped. Go if we name an owner. Go if we see the document, check the assumption, or answer the question nobody has asked yet.
Then Chair writes it up: recommendation, vote count, where the room disagreed, top risk, next steps, and what would make us revisit it.
I still decide. The room does not outvote me. It out-argues my first draft.
Last time I ran it, I asked: Do I buy a second DGX Spark-class box now — a GB10 with 128GB — or wait for a 512GB Mac? I run local AI on one ASUS GX10 today. Chair: make the final call. Say what would reverse it.
Judge it as a WikiWayne purchase: more experiments, growth on X.
Mission said Wait, unless a two-week log shows the 128GB ceiling costing me real hours every week. Log every hit.
Counsel said Go if the return window covers the full test, and I buy it in my business name — separate books, separate data from my day job.
Skeptic said hold off on both. I had not named one blocked job. Two linked GB10s are not one 256GB pool. A Mac could break my CUDA work. Test my largest model at the most aggressive quant I would accept.
Operator said Go if two linked boxes run my actual model fast enough. Show the tokens per second. Test a published result, or rent two nodes for a day.
Finance said Wait. Do the math I had not done: hours lost a month, times the value of an hour, times twelve, against the cost after resale. Cheapest test first — a two-week log, a one-day rental. After my clarification: Go if GX10 posts show a return covering the box in about a year.
Chair's memo: Wait. Buy only once the GX10 earns it. No seat voted Buy now.
Next: GX10 stable through Oct. 28. One GX10 experiment post a week. Track followers and sponsorship leads. By Oct. 31, set an hour limit, list three experiments needing two boxes, confirm the setup with my CPA.
Reversal: buy the second GB10 in the WikiWayne name if it stays stable, weekly posts grow the following, that list exists, and the return window covers the two-box test. A 512GB Mac only enters if it actually ships and the job runs without CUDA.
I have not decided yet. The room did not spend my money. It told me what has to be true before I do.
Why not just ask one model the same question six ways?
You can. I have.
A single chat drifts. It wants to be helpful. It smooths the edges off the last answer and hands you a consensus that nobody actually believes.
Seats hold their ground. Finance does not suddenly become Mission because the conversation got warm. Skeptic does not retire after one good rebuttal.
The disagreement is the product.
It also changes what I bring into the room. I write cleaner questions now. If I cannot state the decision in two sentences, the room sends me back — Chair first, Skeptic enjoying it a little too much.
What do I use it for?
Big calls. Expensive calls. Calls I will have to live with. Calls where being wrong in public is not the worst outcome — being wrong after I ignored an obvious question is.
I do not use it for trivia. I do not use it to avoid deciding. And I do not treat a unanimous "Go" as permission. Unanimous should make you more curious, not less.
Here is my honest limit on the whole thing.
These are still AIs. They can miss context. They can be confident and wrong in a very organized memo.
The Board Room does not make a bad decision good. It makes my reasoning show its work. I can see which assumption is load-bearing. I can see what I still need to verify in the real world.
That is enough. Most bad calls I have made did not fail for lack of intelligence. They failed because one lens ran the whole meeting.
Now six lenses have to share a table.
I bring the decision. They bring the argument. I leave with conditions, not vibes.
If you run anything — a business, a team, a project, a life with more than three moving parts — steal the structure.
Name the seats. Give them jobs that conflict on purpose. Force a vote with conditions. Make one of them write the memo.
Then be the human who decides anyway.
That last part does not get outsourced.
Want your own Board Room? Each seat is a free Grok Bot template, one click to add:
