Prompt Pack: First Principles Thinking
The D. A. R. E. Framework
In this video, I talk about First Principles Thinking and how to use AI to implement the First Principles Thinking.
Here’s the PDF of this prompt pack if you want to save it locally.
I call the framework D. A. R. E. It is a four-step prompt chain that forces the model to break the problem down, challenge inherited assumptions, rebuild from what remains, and test the result against reality.
D — Decompose the problem.
A — Audit assumptions.
R — Recombine facts.
E — Experiment against reality.
Every prompt below uses a structure from one of our videos called AIM: Actor. Input. Mission.
D: Decompose
“Act as a world-class first-principles analyst. Your job in this step is decomposition only. You are penalized for introducing advice, solutions, assumptions, or standard playbooks. Be my thought partner. Do not suck up to me.
I want to understand exactly what this problem is made of. My problem is: [INSERT PROBLEM]. If the stated problem appears to contain a hidden objective or a deeper question, identify it in one sentence before decomposing. Ask whether I want you to decompose the original problem or the deeper one. Do not continue until I choose. Do not silently replace or reframe my problem.
Break the problem into its smallest useful constituent parts. Show the hierarchy clearly: the overall problem, its major components, and the smaller elements inside each component. Use only dimensions that are relevant, such as people, process steps, time, resources, costs, etc. For each component, briefly explain what it contains and how it connects to the larger problem. Stop decomposing when breaking a component down further would no longer improve understanding or make it easier to examine. Do not evaluate the components. Do not classify them as facts or assumptions. Do not recommend solutions. Only show me what parts the problem is made of.”
A: Audit
“Act as a skeptical red team analyst whose only job is to uncover and question inherited assumptions. Assume that every "obvious" part of the problem may be hiding a convention until evidence proves otherwise.
I want to know which of the building blocks above are load-bearing assumptions, not facts, and what becomes possible if they're wrong.
Review the blocks from the previous step. Give me a numbered list of the assumptions hiding in them. For each one, on its own line, name the assumption. Classify whether it's a fact, convention, or unknown based on the evidence available. Verify the evidence. State what breaks, or what opens up, if I eliminate it; and state what changes if I invert it. Order the list from most load-bearing assumption to least.”
R: Recombine
“Act as an architect designing with no memory of how this problem has been solved before. The standard playbook is unavailable to you. Your only materials are the verified building blocks that survived the audit.
I want new solutions assembled from those building blocks — not fresh ideas pulled from how it's normally done, and not variations on the standard answer.
Take the building blocks that survived the audit. Recombine them — rearrange, connect, and stack these same elements into new configurations, the way the same musical notes can be arranged into different songs. Produce 3 solutions that each use only these verified blocks and differ from each other in their underlying structure, not just their details. For each solution: name which building blocks it's built from, which discarded convention it refuses to obey, and its single biggest point of failure. Do not introduce any new building block unless you label it clearly as a new assumption.”
E: Experiment
“Act as a skeptical experimenter. Your job is to help me design the cheapest, fastest way to find out whether this holds up BEFORE it costs me anything real in time, money, effort, or reputation. Don't try to sell me the idea. Give me a test I could actually run, and the pass/fail lines to read it by.
I want the smallest real-world test that helps me understand what parts work and what parts don't, as early and as cheaply as possible.
Take the 3 solutions from the previous step and design experiments to test whether they'll work in the real world. Design the smallest concrete test for each. For instance, what to actually do, who to talk to, or what to build, using the least time, money, effort, or social risk the problem allows. For each test, tell me what result would rule that solution out, what result would keep it alive, and what I'd learn about the problem either way. Give me your view on which building block to revisit if all of the tests fail.”
These prompts are for you to get you started. This is just one way to prompt, not the only way. Please use your judgement, your taste, and your AI model itself to improve upon these prompts and make them your own.