Fantasy Board: My Second Agent

I wrote my second agent — my Fantasy Board of Directors — inspired by Mike Collins, CEO of Alumni Ventures (I’m on Mike’s board). Mike had been experimenting with building AI personas of famous business leaders as thinking partners, and the concept stuck with me immediately. What if I could build a full board of advisors — people I’d never be able to recruit in real life — and use them to pressure-test my thinking?

The idea got some great coverage in Fortune and Business Insider, so clearly it resonated. But what I didn’t share in those pieces was the detailed how — the construction instructions, the prompting approach, and the management discipline required to make it actually useful. Every CEO can build one of these. Here’s exactly how.

Build

1. Pick your Board members. You need people where there’s enough publicly available information — books, interviews, speeches, shareholder letters, biographies — to do deep research on their profile as a thinker and advisor. My board includes Warren Buffett, Steve Jobs, Oprah, Marc Benioff, and others. The key is diversity of perspective, not a room full of people who think alike. My leadership team had fun with this process, doing a “fantasy draft” akin to how you’d draft a fantasy football team.

2. Build Board Profiles. For each member, create a detailed persona summary that captures their decision-making style, communication patterns, known biases, areas of expertise, and advisory approach. Here’s a sample of the profile I built for Warren Buffett. And here’s a sample of the prompt I used to generate the profiles — you’ll need to fill in the blanks for each person.  I tried a few different engines to generate these. The most effective ended up being Gemini, though Mike used ChatGPT Deep Research. Experiment and see what gives you the richest output.

3. Create an Instruction Set for the agent. This is the backbone of the whole thing. Here’s the one I use for my Fantasy Board – You’ll want to customize this heavily and make sure it’s really consistent — the instruction set tells the AI how the board members interact with each other, how they challenge you, and what kind of output you expect.

4. Create a container. Set up the Fantasy Board in whatever platform works for you — Notebook LM, a Claude Project, a custom GPT, or an agent in your corporate AI environment. I’ll skip the basic platform instructions since most people reading this know how to set up an AI project at this point.

5. Load it with context. This is what separates a toy from a tool. Upload your company values, product briefs, investor materials, financial results, strategic plans — anything that grounds the AI in your actual organizational reality. I’ve given my agent access to my Google Drive, email, Slack, and call recordings, so it has most of my work-related content and communications as part of what it scans. The richer the context, the more relevant the advice.

Use

Clear prompting is key to getting high-quality outputs. Think of it as briefing a high-level advisory team: the clarity and precision of your ask directly shapes the quality of the result. The more specific and structured your inputs are — including decision context, constraints, and success metrics — the more actionable and persona-aligned the responses will be.

I ask the Fantasy Board a lot of basics related to both my real Board and my exec team. What will those groups expect to see at the next meeting? Can you pressure-test and critique the materials I’m preparing? Can you help me draft those materials? Can you look at my strategy from each board member’s perspective and tell me what’s missing?

The outputs can be surprisingly sharp. After uploading a recent board book, I got Marc Benioff telling me to “go bigger, faster” and “own the category, don’t just participate in it,” while Warren Buffett wanted to dissect unit economics and customer retention. That tension between perspectives is exactly what a great board does in real life.

Manage

This is the section most people skip, and it’s the one that matters most. Garbage in, garbage out (more on hat topic next week).

Kamil Banc, who puts out tremendously valuable content around enterprise use of AI, wrote a fantastic piece highlighting the risks. Here’s the key quote:

“The agreement trap in high-stakes decisions. If you use AI to evaluate strategic options, vet candidates, or assess risk, the model’s tendency to validate your framing means you’re getting a biased second opinion dressed up as an objective one. Before trusting AI-assisted analysis on anything consequential, feed it the counter-position first. Make it argue against your preferred outcome. If it flips easily, the first answer was agreement, not analysis.”

That’s dead-on. It is really easy to get caught up in crappy advice or self-referential advice with these tools. A Fantasy Board that just agrees with you is worse than useless — it’s dangerous, because it gives you false confidence.

Beyond managing for bias, you have to actively feed the agent to keep it current:

• Internal updates. Financial results, key decks, team changes. The agent needs to see results, not only your broadcasts out.
• External press about the company. Reviews, coverage, competitive mentions.
• Current context. This is critical, because models are trained at a point in time. Recently, I shared some thinking about our AI work with the Fantasy Board, and it was just missing all kinds of context around the current state of AI and productivity — so it gave me feedback that made no sense. Once I fed it some articles about the current state of the technology, the advice quality improved dramatically.

Your Fantasy Board is only as good as the information it has. Treat it like a real board: keep it informed, challenge its conclusions, and never confuse agreement with wisdom.