Last week, I ran a business idea past my advisors, and they pulled it in three different directions.

One said it wasn't really a business, at least not the way I'd described it. Another said the opposite, that the part I was most nervous about was its strongest part. A third waved them both off and said I was circling a question I could settle for fifty dollars.

None of my advisors is a real person. They're files.

For a few years now I've been keeping what I half-jokingly call a “board of directors”: a folder of talks, essays, interviews, and other content from people I trust, which my AI reads before it weighs in on a decision.

I’ve handed my AI plenty of jobs before. In Issue 1, I gave my AI a name, a memory, and a job. In Issue 8 I ran it like an employee. But this is the opposite kind of help. It doesn't do the work, it tells me whether it's the right move in the first place.

It can do that because the advisors don’t all see it the same way. When you ask a single AI for advice, you get a single answer, quietly bent toward agreeing with you. But a board doesn't just agree with itself. A board gives you several honest takes, and where they don't line up is often where my blind spots turn up.

So here's how to build your own board of directors.

It comes down to three pieces:

A folder your AI has access to. Mine is just a folder on my desktop that my AI can read. In ChatGPT or Claude, a "Project" does the same job: a place the AI always sees, so you're not re-uploading your advisors every time you want their take.

Real files, cleaned up. When I find a talk, an essay, a podcast, even a chapter of a book worth keeping, I don't just drop the raw version in. I grab the text (tip: YouTube provides a transcript you can copy for almost any video or podcast) and ask my AI to clean it up. I ask it to keep all the substance and cut only the noise. Here’s an example:

Here's [a transcript / article / chapter] from [name]. I'm saving it to a folder of thinkers I reference, so rewrite it as a clean version in their own words. Keep every opinion, argument, and claim, plus the reasoning, evidence, and examples behind each one, and keep their caveats and nuance. Cut the filler, the sponsor reads, the small talk, the repetition, and the messy formatting. Don't summarize or shorten the ideas themselves; just strip the noise around them. Return tidy markdown with light headers so I can find what I need later.

I keep both the original as its own file, in addition to the cleaned one. The clean copy is what the board reads day to day; the original is there when I want an exact quote or to dig deeper. Then I file both under whoever said it.

And don't just reach for a famous name. Telling your AI to "answer like Steve Jobs" just invents a Steve Jobs. Give it files extracted from what Steve Jobs actually said or wrote. I don't curate the folder, either. I keep anything smart or insightful, from anyone, and a wider range makes for a more useful board.

The instruction that turns files into a board. I keep a short note at the top of the folder telling the AI how to behave. You can do the same, or drop this into a prompt whenever you need to invoke the opinions of your board:

You are my board of advisors. This folder is the real work of people I trust. When I bring you a decision, work out which of them has something useful to say, and answer as each of them, in their own view. Don't merge them into one voice. Where they'd see it differently, show me the disagreement. Finish by telling me where they split and what it means for my call, not a single tidy recommendation.

Here's a real example I ran through my own board. I'd been kicking around the idea of a Gimbal Marketplace (work in progress), a curated shop for the small "vibe-coded" tools creators make, stocked only with the ones our team actually loves to use.

I dropped the idea in front of the board. Most of the room said nothing, which is how it should work. Out of the 289 advisors I’ve collected, only 10 had something relevant to say about my topic. A marketplace question only pulls the opinions of those who've actually built one.

The first advisor came back blunt: "By your own test, this isn't a marketplace. If you make money sending people off to buy somewhere else, the sale happens off your platform." But the brand people pushed the other way: “A shop people trust is a real business, textbook or not.” And the advisor who broke the tie wouldn't pick a side. He told me to stop turning it over and put the list in front of the podcast viewers we already have, decide in advance what a flop looks like, and only build the real version if the small one works.

The board swapped my question for a better one. Not "should I build this marketplace," but "is this a marketplace at all, or a shelf people trust?" That reframing is what you're actually after, and you only get it because the advisors don't all think alike.

Summarized in 30 seconds

  • Ask one AI and you get one answer that's quietly agreeing with you. A board that weighs in from several angles shows you the blind spot instead.

  • Build it from three pieces: a folder your AI reads, real files cleaned up and filed by person, and an instruction to answer as those people.

  • Two prompts to add to your arsenal: one to clean a source into a digest, one to make the AI answer as a board.

  • It's still a model. It can be confidently wrong, so you make the call.

  • This week: start one folder, clean up one piece from one person you trust, and run a real decision past them.

As for Gimbal Marketplace, I'm not building the full vision yet. I'll be running the small, cheap test the board talked me into first.

me+machine

Keep Reading