Multi-Lens AI Thought Partner
Challenge an idea through named thinking methods and adversarial review.
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 5
- Confidence
- 94%
Configure an AI thought partner with several explicit reasoning methodologies, such as first-principles analysis and adversarial red teaming. Present an idea together with the decision being considered, then instruct the assistant to reconstruct the problem through one named lens. Repeat with a contrasting lens so hidden assumptions, weak analogies, and alternative mechanisms become visible. Finally, request an unusually candid critique and ask for evidence that would invalidate the proposal. The output is not a decision delegated to AI; it is a structured set of perspectives that helps the human test reasoning, strengthen the idea, or abandon it before costly execution.
Origin
Kieran Flanagan described a custom GPT trained on multiple thinking methodologies and used as a brutally honest thought partner.
Core principles
- 01A model can deliberately switch between thinking methodologies.
- 02Different lenses expose different assumptions.
- 03Adversarial feedback is more useful when explicitly requested.
- 04AI critique supports rather than replaces accountable judgment.
How to run it
- 1
Frame the decision
Describe the idea, intended outcome, constraints, and decision that must be made.
Pro tip Separate facts from assumptions.
Watch out A promotional description will bias the critique.
- 2
Apply first principles
Ask the assistant to remove inherited assumptions and identify the necessary underlying facts and mechanisms.
Pro tip Require each premise to be stated explicitly.
Watch out A model may label ordinary decomposition as first-principles reasoning.
- 3
Change lenses
Reanalyze the proposal using another relevant methodology, stakeholder perspective, or failure model.
Pro tip Choose a lens that creates genuine tension with the first.
Watch out Near-identical lenses add volume rather than insight.
- 4
Red-team the proposal
Ask for the strongest case against the idea, including failure modes and disconfirming evidence.
Pro tip Request severity and likelihood separately.
Watch out Harsh tone is not a substitute for sound analysis.
- 5
Synthesize human judgment
Compare the analyses, investigate critical unknowns, and make the accountable decision.
Pro tip Turn uncertain assumptions into tests.
Watch out Do not accept the model’s confidence as evidence.
In the wild
A founder asks a trained assistant to assess a new enterprise tier through first principles, customer incentives, and a premortem. The assistant identifies an unsupported assumption about willingness to pay and proposes a small pricing interview test before development.
→ The founder replaces a large speculative build with a bounded validation experiment.
Common mistakes
Using brutality as a quality metric
An insulting response may feel rigorous while offering no valid evidence or mechanism.
Delegating the decision
The framework is designed to broaden analysis, not transfer accountability to a model.
Is it for you?
Best for
It is best for reversible strategic decisions, product ideas, positioning, and plans that benefit from structured challenge.
Not ideal for
It is not ideal for high-stakes expert decisions where the model lacks authoritative evidence or domain competence.
From the transcript
“So I've taught it first principles. I've taught it these different thinking methodologies.”
“And it's really good as a thought partner. I love it to red team ideas. I tell it to be really, really brutally honest.”
From the episode
The AI Workflow That Lets 50 People Do the Work of 500 ($2B Founder Reveals)