Build an AI Advisory Board with GPTs for Decision-Making
Carolyn Healey · article · 2026-05-25 source ↗ Reader 🖥️
Running major decisions through a self-built panel of 3–5 custom GPTs modeled on the frameworks of admired leaders is a cheap, always-available substitute for an advisory board that forces decision-makers out of single-mental-model thinking by surfacing questions they would not have asked themselves.
practical · mixed (prescriptive how-to with expository scaffolding) — A LinkedIn-format prescriptive piece structured as a five-step recipe with brief 'Reality:' caveats. The mode is 'do this' rather than 'understand this'; the expository content (why a panel beats a single mental model) exists only to motivate the procedure.
Structure
| Part | Role |
|---|---|
| Opening hook: 'My board of advisors meets whenever I need them' | Establishes the advisory-board metaphor and the implicit promise (always available, zero cost, frictionless access) |
| Mechanism claim: 'Most CXOs make decisions inside one mental model' | Provides the only theoretical justification — the failure mode the practice is designed to fix |
| Five-step recipe (Choose / Gather / Build / Calibrate / Run) | The operational core — a procedure for instantiating the practice |
| 'Reality:' caveats after each step | Defensive moves that pre-empt obvious failure modes (generic instructions, shallow sources, hero-worship, first-iteration disappointment) |
| Closing limitations: 'not the actual person', 'can hallucinate', 'not a replacement for real mentor' | Concedes a narrow set of objections to bound the claim |
| Closing cost-benefit appeal: 'cost of one bad strategic decision dwarfs the cost of building this' | Frames the decision to adopt as asymmetric and obviously favorable |
Key terms
- AI advisory board Decision-Making Practices
A personal collection of 3–5 custom GPTs, each configured to imitate the framework, philosophy, voice, and decision pattern of an admired leader, used to pressure-test decisions before they are made - intellectual DNA Decision-Making Practices
The synthesizable essence of a thinker — derivable from books, shareholder letters, speeches, interviews, and published frameworks — that an LLM can reproduce well enough to generate that thinker's likely response - pressure-test Decision-Making Practices
Submit a contemplated decision to the synthetic panel and use their responses (especially their disagreements with each other) to surface forgotten questions, latent assumptions, and overlooked angles - blind spots Decision-Making Practices
Domains of judgment a decision-maker systematically under-weights, which can be compensated for by selecting advisors whose dispositions differ
Propositions
- explicitMost CXOs make major decisions inside a single mental model — their own.conf: high
- explicitA custom GPT loaded with a leader's books, letters, speeches, interviews, and frameworks can capture that leader's perspective sharply enough to be useful for decision pressure-testing.conf: high
- explicitThe most valuable output of the panel is disagreement between the synthetic advisors — that's where the real decision lives.conf: high
- explicitAn AI advisor's primary value is not giving the right answer but surfacing the questions you forgot to ask.conf: high
- explicitThe cost of one bad strategic decision is large enough that the expected value of building and using this system is unambiguously positive.conf: high
- implicitcontestedAn LLM trained on a thinker's public corpus generates outputs that reflect that thinker's framework rather than merely the surface texture (vocabulary, idioms, anecdotes) of their public writing.conf: highcontradictsRichard Dawkins and The Claude Delusion · p1 — The advisory-board premise that a GPT reproduces a thinker's underlying framework, not just surface texture, clashes with the statistical-mimicry account of LLM output.
- implicitSurfacing forgotten questions is the principal value real mentors and advisors provide.conf: high
- implicitcorroboratedA panel built from American CEOs and investors of a specific era contains enough diversity to compensate for an individual decision-maker's blind spots.conf: highsupports ←Why Great Leaders Need More Than Good Judgment · p9 — Both treat deliberately generating AI personas with different or opposing frameworks as a mechanism for compensating an individual decision-maker's blind spots -- same technique, different framing (contrarian personas vs. a synthetic advisory panel).
- implicitA confident but unverifiable response from a synthetic advisor is more useful than no response, even when the user cannot tell whether the GPT is reasoning or roleplaying.conf: high
- implicitThe failure modes worth flagging are the ones named in the 'Reality:' lines (generic instructions, shallow sources, hero-worship, weak first iterations); deeper failure modes (mimicry of style without framework, absence of feedback signal, practicing the feeling of consulting wisdom without its substance) are absent because they don't exist or don't matter.conf: medium
Arguments
You should build and use an AI advisory board of custom GPTs modeled on admired leaders. strength: low
Explicit premises:
- Most CXOs make decisions inside one mental model.
- The AI advisory board forces you out of that single mental model.
- Custom GPTs loaded with deep source material capture a leader's perspective sharply.
- Disagreement between the synthetic advisors is the most valuable output and identifies where the real decision lives.
- The cost of building the system is small relative to the cost of one bad strategic decision.
Implicit premises:
- An LLM conditioned on a thinker's public corpus reproduces that thinker's framework, not just stylistic markers.
- Surfacing forgotten questions is the load-bearing function of real advisors (relational trust, accountability, situation-specific judgment are either also delivered or non-essential).
- The user can distinguish, or doesn't need to distinguish, between a GPT genuinely applying a framework and a GPT confidently confabulating in that thinker's voice.
- The synthetic advisors' disagreements track real divisions in decision-relevant reasoning rather than randomness, prompt artifacts, or surface persona differences.
- A panel selected from a narrow demographic and ideological band still has the disposition diversity needed to widen a decision-maker's view.
Inferences:
- If the practice's value is mainly question-surfacing, a much cheaper structured prompt asking 'what questions am I forgetting to ask?' may capture most of the upside without the persona infrastructure.
- If the load-bearing assumption is framework-capture, the practice fails silently when an LLM substitutes voice mimicry for framework application — and the user has no feedback signal to detect this.
Pick thinkers whose blind spots differ from yours, not heroes. strength: low
Explicit premises:
- The point is to widen the decision-maker's view.
- Heroes by definition share dispositions with the admirer (or fail to challenge them).
Implicit premises:
- The decision-maker can accurately identify their own blind spots well enough to select thinkers who don't share them (which is in tension with the meaning of 'blind spot').
- Blind spot composition is the right selection criterion (versus, say, framework rigor, domain relevance, or empirical track record on decision quality).
Inferences:
- If you can't see your own blind spots, you can't pick advisors who close them — which would suggest delegating advisor selection to someone else, or using an explicit anti-bias framework instead of intuition.
Narratives
- Example Buffett-GPT instruction: 'Apply circle of competence, margin of safety, and long-term thinking. Ask clarifying questions. Be direct. Challenge assumptions.'
→ Shows what 'capturing a framework' operationally looks like — but in doing so, illustrates the gap between framework capture and framework execution. The instruction enumerates Buffett's signature concepts; whether the resulting GPT actually applies circle of competence (vs. invoking it) is exactly what the article does not test. - Routing a decision to Buffett-GPT, Jobs-GPT, and Nadella-GPT and asking each their characteristic question (financial logic / product signal / culture impact)
→ Demonstrates the panel-in-use: the panel's value comes from each persona having a distinctive question stem. This supports the question-surfacing claim but also suggests that what is doing the work is the prompt-engineered question template, not the deep persona simulation. - 'Available at 11pm the night before a board presentation'
→ Concretizes the always-available promise. It also reveals the actual use case: late-stage rehearsal and reassurance before a high-stakes presentation, where the failure mode of 'practicing the feeling of having consulted wisdom' is most acute.
Questions addressed
- ✅ How can an individual decision-maker compensate for the limits of their own mental model without access to (or expense of) a real advisory board?
By building a panel of custom GPTs modeled on respected leaders and routing decisions through them. - ✅ What concretely is the procedure for building such a panel?
A five-step recipe: choose 3-5 leaders, gather their intellectual DNA, build per-leader GPTs, iterate on instructions, route real decisions through the panel. - ⭕ How would you know whether the synthetic advisor is genuinely applying its assigned framework versus producing confident, fluent imitations of that thinker's surface style?
Not addressed. 'It can hallucinate' is named as a limitation but no procedure for detection, calibration against real-world outcomes, or falsification is offered. - ⭕ What would distinguish using this panel from rationalizing decisions you have already made?
Not addressed. The article does not discuss the risk that a confident, articulate GPT panel functions as an elaborate confirmation device, especially when the user pre-selects advisors whose frameworks they already admire. - ⭕ How does decision quality from this practice compare to (a) a structured devil's-advocate prompt without persona, (b) a premortem, (c) journaling, (d) actually consulting one trusted human?
No comparison offered. The cost-benefit appeal at the end treats the alternative as 'do nothing' rather than 'use other established decision-quality tools.'
Gaps
- incompleteNo epistemic test for whether a custom GPT is applying a thinker's framework versus producing fluent imitations of that thinker's voice. The 'intellectual DNA' claim is the load-bearing assumption of the entire practice — if it fails, the panel becomes a stylized confidence generator. The article offers no method (held-out test cases with known answers, decisions-vs-outcomes tracking, blind comparison against a generic 'critique my decision' prompt) for detecting this failure.
Read next: Annie Duke, 'Thinking in Bets' and 'Quit' — for the distinction between decision quality and outcome quality, and the discipline of pre-committing to what evidence would change your view. Philip Tetlock, 'Superforecasting' — for what actually-calibrated judgment looks like and how to measure it. Daniel Kahneman, 'Noise' — on the gap between confident expert judgment and reliable judgment. - incompleteNo engagement with the possibility that the practice does subtle epistemic harm: it lets the decision-maker experience the feeling of having consulted wisdom without the actual feedback loop (real advisors push back, get hurt, hold you accountable, and update their advice based on what happened last time). The closer the simulation gets to feeling real, the more it may displace practices that actually improve decisions.
Read next: Cutler on accountability and DRIs (directly responsible individuals); literature on simulated advice in clinical and organizational decisions (Logg et al. on algorithm aversion/appreciation); the AI-coding-quality conversation in the user's library (Marcus, the SQLite-vs-LLM-rewrite finding) for the more general pattern of fluent output being mistaken for capable output. - incompleteSelection bias in the advisor pool is unacknowledged. The named exemplars are almost all American, male, billionaire CEOs/investors from a specific era — selected by biographical-survivor bias (we have books about Buffett because he succeeded). Their published frameworks are necessarily post-hoc rationalizations of decisions that worked. A panel of survivors cannot teach you about the decisions that, made the same way, did not work.
Read next: Survivorship-bias literature; Tetlock on the difference between pundit confidence and forecasting accuracy; Annie Duke on resulting (judging decision quality by outcomes). - illogicalSelf-undermining selection criterion: 'pick thinkers whose blind spots differ from yours.' If a decision-maker could identify their own blind spots well enough to do this selection, the blind spots would no longer be blind. The directive presupposes the very self-knowledge whose absence the panel is supposed to compensate for.
Read next: Kahneman, 'Thinking, Fast and Slow' on the bias blind spot; Pronin et al. on asymmetric insight; literature on structured techniques (e.g. premortem, red teaming) that don't require accurate self-diagnosis to function. - uninformedNo engagement with established decision-quality literature that already addresses many of the same goals (Duke, Tetlock, Klein's recognition-primed decision-making, Kahneman on noise, the premortem technique). The piece writes as if the choice is between 'one mental model' and 'GPT panel,' ignoring a substantial existing toolkit that does not depend on the framework-capture assumption.
Read next: Duke, 'Thinking in Bets'; Klein, 'Sources of Power'; Kahneman, Sibony & Sunstein, 'Noise'; the premortem literature (Klein 2007 HBR).