I Built an AI Board of Advisors. Here's Why You Should Too.

One of the strategies I've been using for a significant amount of time is treating AI as a genuine thinking partner for the decisions that actually matter. Not a tool that gives you answers. A collaborator that challenges your assumptions, pressure-tests your logic, and asks the uncomfortable questions your inner circle probably won't.

Most people haven't gotten there yet. They're still in "ask a question, get an answer" mode. And that's fine for everyday tasks. But when you're facing a decision where being wrong costs years, not hours, you need something more structured. You need AI that pushes back.

I want to walk you through how I set that up.

Eight Experts. Eight Different Ways to Tell You You're Wrong (or Right).

I created what I call a Decision Panel. It's an AI-powered framework that puts my thinking in front of eight distinct expert perspectives, each designed to challenge me from a different angle.

Not eight versions of the same helpful AI assistant — eight genuinely different lenses:

• An Operator who asks if I can actually execute the plan
• An Investor watching opportunity cost
• A CFO running the downside math
• A Career Coach asking what kind of life I'm building
• A Contrarian who argues the opposite of whatever the consensus is
• A Customer who asks whether anyone would actually pay for the idea
• A Technologist evaluating whether it survives the next platform shift
• A Pastor/Theologian who asks whether I'm deciding from faith or from fear, and whether the decision aligns with the values I say I hold

After all eight weigh in, I get a synthesis that maps where they agree, where they disagree, and the strongest challenges to my current leaning. Not a vote. A reasoned integration that surfaces the tensions I need to sit with before I decide.

It’s also worth noting that I teach this same approach to students in George Fox’s undergraduate business program, and I’ll be leveraging this same concept in the course I’ll be teaching in the MBA program focused on entrepreneurship.

A Prompt Won't Do This. A Skill Will.

I built the Decision Panel as a custom skill inside Claude, and the skill architecture is a big part of why it works.

A long prompt could do most of this. You could paste eight persona descriptions, a decision brief, and a synthesis request into a single message and get something useful back. People do this all the time. The problem is what happens on decision two, and decision twenty.

You've got two options once you've got the prompt written:

• Option One: You keep it in a file somewhere and copy-paste it every time. That works, but it's tedious enough that you'll skip it on the day you actually need it most, which is usually the day you're tired and rushed and the decision matters.
• Option Two: You rebuild it from memory each time. That's where the drift starts. The CFO I described last month isn't quite the same CFO I'll describe this month. The questions the Pastor/Theologian asked last time aren't quite the questions I'll come up with this time. Each version is plausibly close to the last one, and that's the problem. I'm not running the same panel anymore. I'm running a slightly different panel that I'll grade against the same standard, and the comparison breaks down.

A skill fixes both versions of that problem. The personas, the questions, and the templates are all locked in. I'm not rebuilding the panel, I'm running it, and the rigor is the same on decision 50 as it was on decision one.

If you're not familiar, Claude lets you create reusable skills. Structured instructions, templates, and persona definitions that Claude can draw on whenever you invoke them. Think of it as giving the AI a playbook it follows consistently, rather than hoping a one-off prompt captures everything you need.

For the Decision Panel, that means each of the eight experts has a detailed persona file that defines their worldview, their priorities, and the kinds of questions they ask. The Operator isn't a generic "think like an entrepreneur" instruction. There's a specific set of instructions shaping how that perspective shows up every time. Same for the CFO, the Contrarian, all of them.

There's also a decision brief template that forces me to structure my thinking before the panel weighs in. What are the options? What's my current leaning and why? What am I most uncertain about? What am I afraid might be true? That structure matters. The quality of the challenge you get from AI is directly proportional to the quality of the context you give it.

And there's a synthesis template that ensures the output isn't just eight separate opinions but an integrated view of where the perspectives converge, diverge, and where the unresolved tensions live.

The point is: none of this works as a single prompt. It works because it's a system. Personas, templates, and instructions that build on each other. That's what skills let you create. You design the framework once, and then you can invoke it repeatedly with confidence that you're getting the same rigor every time.

What Makes a Persona Actually Useful

The temptation when you set this up is to write generic personas. "You are a thoughtful CFO. Give me financial perspective." That's the version that produces eight slightly different paragraphs of the same diplomatic advice.

The personas that earn their keep have three things going for them:

• A distinct worldview.More than a job title. An actual point of view. My CFO doesn't just "think about money." He believes most decisions get made on optimism and corrected on margin, and that the right question to ask first is always "what does this cost us if we're wrong?"
• A specific set of questions they always ask.The Pastor/Theologian doesn't give me theology. The persona is built on the pastors and theologians I actually listen to and have learned from over the years, and the questions they tend to ask. He asks whether I'm making this decision from faith or from fear, whether the people closest to me would recognize it as consistent with who I say I am, and whether I've talked to anyone who would lose if I won. Same three questions, every time. The repetition is the point. It builds a pattern I start carrying into decisions even when I'm not running the panel.
• A bias they're allowed to have.Real advisors aren't neutral. My Contrarian is allowed to be wrong on purpose. My Investor is allowed to over-index on opportunity cost. The personas earn their keep by being lopsided in useful, predictable ways.

For example, here is just a snippet of some of the “instructions” the Pastor/Theologian is operating under as part of my decision team.

In short, If you find yourself writing a persona and it sounds like a job description, you're not done. Keep going until it sounds like a person you'd actually argue with.

The Same Idea Works on Other Platforms. The Mechanics Vary.

I built this on Claude because skills are the cleanest fit for what I needed. But the underlying pattern travels. ChatGPT has Custom GPTs. Gemini has Gems. Microsoft has Copilot agents. The specific feature names will keep changing, and probably faster than this post will age, so I'm not going to walk you through screenshots that will be outdated in six months.

What matters is the architecture, and that part is stable:

• A defined persona for each expert, with a worldview and a set of questions they always ask
• A structured intake template that forces you to give the AI real context before it weighs in
• A synthesis step that pulls the perspectives together rather than leaving you with eight separate opinions

Every major AI platform supports this pattern in some form today. Some make it easier than others. Some have better instruction-following, some let you upload reference files more cleanly. Those differences matter when you're choosing a platform, and they'll keep shifting. The methodology underneath them won't.

If your company has standardized on a tool, build it there. If you're choosing, pick the one with the strongest support for persistent instructions and uploaded context. The tools will keep changing. The discipline stays the same.

What This Isn't

A few things worth being clear about, because I've watched people get the wrong idea about this kind of setup.

This isn't a replacement for a real board, real mentors, or real advisors. The people who know your business, your family, and your history bring context an AI can't synthesize, no matter how good the persona file is. The panel is what I use when I want to think harder before I bring something to those people, not instead of bringing it to them.

It also can't own the decision. The panel surfaces tensions. It maps where the perspectives pull in different directions. It can give me the sharpest version of the case for and against. But owning a decision means living with the consequences, repairing the relationships if it goes wrong, carrying it forward into the next decision and the one after that. AI can't do that part, and the people who try to offload it onto AI are setting themselves up to lose the muscle they most need to keep building. The whole value of the panel is in the part where I sit with the unresolved disagreement and make a call anyway, knowing it's mine.

And it isn't a substitute for thinking. If anything, it demands more of it. A good panel session leaves me with sharper questions. Easier answers aren't the point.

You Don't Need a Board. You Need a Better Framework.

I built this for myself. But the logic extends to anyone who faces high-stakes decisions without a formal board of advisors to pressure-test their thinking. Founders. Senior leaders in mid-sized companies. Executives who have a board but want to stress-test an idea before they bring it to the table.

Most of these people make major calls with too few perspectives in the room. They value diverse input, they just don't have easy access to it on demand. A well-built skill changes that equation entirely.

The One-Off Prompt Gets You a One-Off Answer

This is the difference between using AI and collaborating with AI. If you just ask "should I take this opportunity?" you'll get a diplomatic answer that reads well and helps very little.

The Decision Panel works because each expert has a defined role, a clear worldview, and specific instructions to challenge rather than confirm. That kind of rigor doesn't come from a clever prompt. It comes from designing a system the AI can run inside, the same way every time.

The people who get the most from AI aren't asking it the most questions. They're building systems. Skills, templates, and structured frameworks that turn AI into a real collaborator for the decisions that matter most. The one-off prompt gets you a one-off answer. The system gets you a thinking partner.

If you want to compare notes on how you're approaching it, reach out. I'd love to hear what you're building. You can send me an email at scampbell@georgefox.edu or connect with me on LinkedIn.