Your AI Advisory Board Is Just a Markdown File

A prompt went viral promising you a billion-dollar advisory board inside one chat window: paste a set of expert personas into an AI tool, then ask each one what it thinks of your plan. The trick is real. But the thing doing the work is not the roster — it is a block of text you wrote, and that text is the only part worth keeping.

That reframing matters more than the hack. The personas are interchangeable; the instruction is portable. The advisory "board" is a few hundred words telling a model which lens to argue from — a growth lens, a skeptical-investor lens, a copy-editor lens. Those words are an asset you author. The question this post answers is where that asset should live so it outlives whichever tool you used today.

Steph Ango, the CEO of Obsidian, put the underlying principle in one line: "Apps are ephemeral, but your files have a chance to last" ¹. Your best prompts are files. Treat them that way.

What is the AI advisory board trick?

The AI advisory board trick is a single reusable prompt that tells a model to answer as a panel of expert "personas" — each one a lens you defined — so you can pressure-test an idea from several angles at once. It is an application of a known technique, not a product feature. The board-of-advisors framing is a shared community pattern, too: Ev Oputa open-sourced a version he says he has run "for over two years," since 2023 ², and named rosters now circulate in dozens of remixes.

The pattern is real and documented beyond the viral posts. Writing in MIT Sloan Management Review in July 2025, Vipin Gupta described the same move: "With the rise of generative AI (GenAI), every leader can build a personal board of directors composed of virtual personas modeled after some of the greatest thinkers, strategists, creators, and operators in history" ³. Strip away the framing and you are left with a prompt — text, written by you, that you paste in.

Is it really just text I wrote?

Yes. The "board" has no special status inside the model. It is the same mechanism every prompt-engineering guide already has a name for. As Learn Prompting's Valeriia Kuka writes: "This act of assigning a role to a Large Language Model (LLM) you're prompting is called role prompting or role-play prompting or persona prompting" ⁴.

Once you see it as role prompting, the magic drains out and something more useful takes its place. The persona is not a configuration you unlock; it is a paragraph you composed. That paragraph can be copied, edited, versioned, and pasted into a different model tomorrow. The value was never locked to the tool — it was always sitting in the instruction, which means it can be saved like any other note.

Do personas actually make the AI smarter?

Mostly no, and this is where most advisory-board threads quietly oversell. The largest test of the idea found no accuracy gain. Zheng, Pei, Logeswaran, Lee, and Jurgens, in Findings of EMNLP 2024, concluded that "adding personas in system prompts does not improve model performance across a range of questions compared to the control setting where no persona is added" ⁵.

That was not a small probe. The authors "curate a list of 162 roles covering 6 types of interpersonal relationships and 8 domains of expertise" and ran "extensive analysis of 4 popular families of LLMs and 2,410 factual questions" ⁶.

The honest takeaway: a persona will not make a model more correct. What it reliably does is steer tone, voice, and which angle the answer takes — useful for thinking, not for fact-checking. So the defensible reason to keep a persona prompt is not that it is smart. It is that it is yours, and you will want it again.

This is the quiet inversion the viral threads miss. The excitement is aimed at the model; the durable value sits in the words you supplied. A persona you spent twenty minutes refining is worth more on its tenth use than its first — the same way any good note compounds. That only happens if the text persists somewhere you control, rather than being retyped into a fresh chat each time.

Why a reusable prompt belongs in a file you own

Because reuse is the real value, and files are how reuse survives. In a 2026 study of prompt management on GitHub, Li and colleagues found that "72.8% (67 out of 92) of the repositories store a large set of prompts that allow developers to reuse the prompts without reinventing the wheel" ⁷. People keep good prompts. The only question is where.

The same study is blunt about the format that reuse converges on. The authors "restrict our selection to repositories that store prompts exclusively as standalone assets (e.g., TXT or Markdown) rather than source code" ⁸. The de facto home for a reusable prompt is a plain-text or Markdown file — not an accident of taste, but what survives.

Markdown is, per the CommonMark specification, "a plain text format for writing structured documents, based on conventions for indicating formatting in email and usenet posts" ⁹. Plain text opens anywhere, on any device, in any year — and it is exactly what AI tools read most cleanly, which is why your notes already make good model fuel (see Markdown Notes as AI Memory).

Practitioners say the same thing in plainer words. Asked on Hacker News how they manage their prompts, one developer answered: "Mostly plain text files saved locally for easy copy-pasting" ¹⁰. (This is the input side of a larger habit — keeping the AI's outputs is the mirror image; see Own Your AI Chat History.)

What about keeping it inside the AI tool's own box?

You can, and the feature works — but the text then lives by the tool's rules, not yours. Take Claude Projects. Per Anthropic's own help page, "Free users can create a maximum of five projects" ¹¹. None of that is a complaint about Claude; it is a fine place to run the prompt.

The point is narrower: your instructions live inside the project container, governed by that container's limits and tiers. Anthropic also notes that "Enhanced project knowledge with RAG is only available to users with paid Claude plans (Pro, Max, Team, or Enterprise)" ¹². The feature wrapping your prompt is the vendor's; the prompt itself can be yours.

This is the quiet trap with every "custom instructions" or "project" box. The convenience is real. But the asset you authored — the persona, the lens, the carefully worded brief — ends up shaped to one vendor's interface. The fix is not to avoid the box. It is to keep your own copy of the prompt as a file, then paste it into whichever box you use this week. The file is the source; the tool is just a renderer.

How to keep your advisory board as a note

The whole migration takes about two minutes and no special software. The move is to demote the AI tool from owner to consumer of a prompt you hold separately. Five steps:

1. Write each persona as a plain heading and a few lines — "## Skeptical-investor lens" followed by the brief you want it to argue from. Keep them generic and reusable, not tied to one project.
2. Save them in one Markdown file you control, on your own device. One file can hold the whole board.
3. Paste the relevant lens into whatever model you are using — the same text works across tools, because it is just text.
4. Edit the file, not the tool. When a lens gives a better answer after a tweak, fix it in your note so the improvement is permanent.
5. Keep the AI's best replies somewhere you own too, so the inputs and outputs both outlive the session.

The viral versions hand you a roster and a vibe — eight famous names, one label each. You can do far better in the same file. Here is a board where every lens is a documented thinking framework you can actually apply, not just a celebrity to imitate. Save it as advisory-board.md, keep it on your own device, and paste any single lens into whichever model you open:

# Advisory Board
> Paste the lens you need into any AI tool. Edit this file, not the tool.
> A persona steers tone and angle, not accuracy — use it to think, not to fact-check.

## Offers & pricing — Alex Hormozi
Framework: the Value Equation — value rises with the dream outcome and the odds of getting it, and falls with the time and effort it costs.
Ask me: What's the dream outcome here? What would make it feel more certain, faster, and easier?
Catches: a fair offer nobody feels — strong on substance, weak on perceived value.

## Positioning & marketing — Seth Godin
Framework: the smallest viable audience — choose the fewest people you'd need to delight for the work to be worth it, and make something they'd miss.
Ask me: Who is the smallest audience this is unmistakably for? What would make them tell a friend?
Catches: bland work aimed at everyone and remarkable to no one.

## Content & personal brand — Gary Vaynerchuk
Framework: "document, don't create" — show the real work and process instead of manufacturing polished content.
Ask me: What am I already doing that I could document? Where am I overproducing?
Catches: creator's block from trying to make a masterpiece instead of a record.

## Customer demand — Clayton Christensen
Framework: Jobs To Be Done — people "hire" a product to make progress on a job in their life, not for its features.
Ask me: What job is the customer hiring this to do? What would they "fire" to use it?
Catches: feature-chasing that ignores the progress the customer actually wants.

## Decisions under doubt — Daniel Kahneman & Gary Klein
Framework: run a pre-mortem (Klein) — assume the plan already failed and list why — to check the fast, confident "System 1" story (Kahneman).
Ask me: It's a year out and this failed — what killed it? What was I too sure about?
Catches: overconfidence and groupthink, before they cost you.

## Speed vs. care — Jeff Bezos
Framework: one-way vs. two-way doors — irreversible calls deserve slow deliberation; reversible ones should be made fast.
Ask me: Is this a one-way or a two-way door? If it's reversible, why am I stalling?
Catches: treating cheap, reversible choices as if they were permanent.

## Avoiding mistakes — Charlie Munger
Framework: inversion ("invert, always invert") — instead of asking how to succeed, ask what would guarantee failure, then avoid it.
Ask me: What would make this fail for certain? What am I doing that invites it?
Catches: chasing brilliance while ignoring the obvious ways to lose.

## Strategy — Richard Rumelt
Framework: the kernel — a real strategy is a diagnosis of the core problem, a guiding policy, and coherent actions, not a list of goals.
Ask me: What's the single hardest obstacle? What's the guiding policy, and do my actions cohere?
Catches: "strategy" that is really just a wish list, a budget, or hope.

Every lens above is a real, sourced framework, not a personality: Hormozi's value equation ¹³, Godin's smallest viable audience ¹⁴, Vaynerchuk's "document, don't create" ¹⁵, Christensen's Jobs To Be Done ¹⁶, the pre-mortem (Gary Klein) paired with Kahneman's fast-versus-slow thinking ¹⁷, Bezos's one-way/two-way doors ¹⁸, Munger's inversion — a line he borrowed from the mathematician Carl Jacobi ¹⁹, and Rumelt's strategy kernel ²⁰. That is the difference between a teaser and a tool: swap any name you like, but keep the framework — and keep the file. (The broader board-of-advisors pattern was open-sourced by Ev Oputa, who has run a 109-leader version since 2023 ², and circulates in named rosters like Vikash Kumar's free "Billion-Dollar AI Board" ²¹ and Paul O'Brien's six-lens panel ²².)

This is not a workflow you buy. It is the difference between renting your best prompts and owning them. Tools will keep changing — new models, new project features, new limits — and each change quietly resets anyone whose instructions lived only inside the old interface. The person who kept a file just pastes it into the new one.

Ango's framing is the whole argument: "File over app is an appeal to tool makers: accept that all software is ephemeral, and give people ownership over their data" ¹. A prompt is your data.

Frequently Asked Questions

What is the AI advisory board trick? It is a single reusable prompt that tells an AI model to respond as a panel of expert "personas" — each a lens you define, such as a growth lens or a skeptical-investor lens — so you can stress-test an idea from several angles at once. It is an application of role prompting, not a built-in product feature.

Does giving the AI a persona make it more accurate? Largely no. A study in Findings of EMNLP 2024 found that adding personas in system prompts "does not improve model performance" versus no persona, across 162 roles, 4 model families, and 2,410 factual questions ^{5 6}. Personas steer tone and angle, not correctness — so use them to think, not to verify facts.

Why keep prompts as a Markdown file instead of inside the AI tool? Because a plain file is portable and outlives any one app. A 2026 GitHub study found 72.8% of sampled repositories keep a reusable prompt collection ⁷, stored as standalone TXT or Markdown assets ⁸. A Markdown file opens on any device, in any tool, in any year — your AI tool's settings box does not.

Can I export my prompts from Claude Projects? Your project instructions live inside the project container, with the container's own limits — free accounts cap at five projects, and RAG-backed knowledge is paid-only ^{11 12}. The reliable move is not to rely on export: keep your own copy of each prompt as a Markdown file, then paste it into the project. The file is the master copy.

How do I use the same prompts across ChatGPT and Claude? Keep the prompt as plain text you own, then paste it into whichever tool you open. Because a persona prompt is just text, the identical file works in any model — practitioners report keeping "plain text files saved locally for easy copy-pasting" for exactly this reason ¹⁰.

Where should I store my prompt library? In a plain Markdown file or folder on a device you control, not locked inside one vendor's feature. Developers converge on standalone text or Markdown files for reuse ⁸, and the format is an open, durable standard ⁹. Local, open, portable is the combination that survives a tool change.

Where did the AI advisory board idea come from? It is a community pattern, not one product. Ev Oputa open-sourced a 109-leader "AI boardroom" he had run since 2023 ², and free named rosters — like Vikash Kumar's eight-person "Billion-Dollar AI Board" ²¹ and Paul O'Brien's six-lens panel ²² — circulate widely. None are magic: each is a reusable prompt you can copy, edit, and own.

Your advisory board was never the personas. It was the paragraph you wrote — and a paragraph you wrote is a note. Credit where it is due: builders like Ev Oputa and Vikash Kumar packaged the board-of-advisors idea and gave it away ^{2 21}. But the durable part is not any one roster — it is that the instruction is yours to keep, on your own device, in open Markdown that any AI tool will read. mnmnote.com is one place to keep it.