The Car Wash Problem
Source: Limited Edition Jonathan Substack
Author: Limited Edition Jonathan
Original source: https://limitededitionjonathan.substack.com/p/the-car-wash-problem-why-the-most
Source type: essay
Private backup: the full article text is archived in the private repository at archives/articles/limitededitionjonathan-substack-com-the-car-wash-problem-why-the-most.source.md. It is not published on the public Quartz site.
Summary
The article uses a viral AI reasoning failure—whether to walk or drive to a car wash 100 meters away—to argue that AI and humans often solve the question presented rather than asking whether it is the right question. The author extends the pattern into video editing tools, eye-contact correction, AI agents, Zapier, Canva, AI-generated infographics, and database/workflow design. The central claim is that AI makes bad problem framing more dangerous because it can confidently, quickly, and fluently solve the wrong problem. The proposed remedy is to state the desired outcome, identify unknowns, resist easy shortcuts, and learn primitives rather than only products.
Big ideas
- Students need to bring the purpose; AI should not supply it for them
- AI literacy should help people notice how AI changes what counts as knowing
- Schools should start with learning values before choosing AI tools
- AI tools should be judged by the work they will actually do
- AI is changing what knowledge work asks people to do
Claims
- AI can make a poorly framed problem worse
- “Use case” language can hide what AI adoption changes
- Prompting AI is a literacy practice, not just a technical trick
- AI literacy only works when it is connected to subject-area knowledge
- AI tools should be tested on the real tasks they will be used for
Key evidence and examples
- The car wash prompt fails because models optimize travel time instead of recognizing that the car must be at the car wash.
- The article distinguishes solving the wrong problem, defining the right problem wrong, treating a feature as a problem, and having no real problem beyond wanting to use a tool.
- Examples include silence-removal tools that clip natural speech, eye-contact correction that removes a trust signal, and beautiful AI infographics that fail communication goals.
- The author’s framework: state the outcome, ask what you do not know, be suspicious of easy, and learn the primitive rather than the product.
Pull quotes
Optimizing the wrong question
The AI turned a non-question into a full optimization analysis of a problem that doesn’t exist.
Perfect tool, wrong problem
The tool worked perfectly. It did exactly what it was told. The problem definition was wrong, and the person operating it didn’t have enough domain knowledge to catch it.
Wrong problem at speed
The tools are incredible. AI is the most powerful problem-solving technology humans have ever had access to. But it solves whatever problem you point it at - including the wrong one, with absolute confidence, in record time.
Education relevance
High for AI literacy, project-based learning, instructional technology, workflow design, media production, and teaching students to define outcomes before using tools.
Durability note
The particular model examples will age, but the article’s warning remains durable: AI can optimize confidently for a badly framed problem unless a human brings domain understanding.