Skip to content

Influences

The framework is not invented from scratch. It borrows from people who thought hard about decisions, failure, and judgment long before AI, and it puts their ideas to work on the specific problem of framing an AI project. These are the thinkers the course leans on, and the one idea it takes from each.

PersonWhat the course borrows
Daniel KahnemanSystem 1 vs. System 2. Fast pattern-matching jumps to a solution; slow, deliberate thinking questions the frame. Framing is System 2 work.
Charlie MungerInversion. Ask "how will this fail?" and design backward from the failure modes instead of forward from the happy path.
Richard FeynmanThe Feynman Test. If you can't explain the project plainly, with no AI jargon, you haven't framed it yet.
Abraham MaslowThe law of the instrument — to a hammer, everything looks like a nail. The course calls the AI version the Maslow bias: reaching for a model before asking if the problem needs one.
Jeff BezosWhen the data and the anecdote disagree, trust the anecdote and go look. A reminder to chase the signal your metrics are hiding.
Andrew GelmanThree ways to attack any problem, in order: first-principles reasoning, then an empirical baseline, then statistical modeling. The simple approaches catch framing errors first.
Teresa TorresThe Opportunity Solution Tree. Map the outcome to opportunities to solutions before committing to a build.
Andrej KarpathyWhen generation gets cheap, the bottleneck moves to taste — telling which output is actually worth keeping.
Mark StouseYour job is to help executives make better decisions, not to ship models. The model is a means, not the goal.
Gregory LewandowskiThe five things executives actually care about: revenue, cost, risk, customer satisfaction, employee satisfaction. Tie the project to one of them.
Jeff CammThe lever test. A recommendation about something the stakeholder can't change is useless, no matter how accurate.
Steve JobsStart with the customer experience and work backwards to the technology, not the other way around. Problem and outcome first, model second.
Elon MuskThe five-step design process — make the requirements less dumb, delete the part, simplify, accelerate cycle time, automate last. It maps almost one-to-one onto framing an AI project.

The course also draws on a set of software-engineering laws — Goodhart's Law, the Principle of Least Astonishment, YAGNI, KISS, and others — as cross-domain anchors. They show up throughout the framework.

See how these ideas turn into a process →

Companion site for the AI Problem Framing course on Maven.