Fourth rule.
Structure your data.
Once you stop losing information and you constrain how it comes in, something has quietly happened. Your data arrives in a shape you decided. Not prose. Not screenshots. Fields, states, dates, amounts.
That’s not tidiness. That’s the difference between a workshop and a factory.
In a workshop, a skilled person looks at each job and figures it out. It works, and it doesn’t scale. In a factory, parts arrive in a known shape, so the same step can run a thousand times without a decision. The intelligence moved from the person to the design.
Structured data is what lets a process be repeated. It’s also what lets it be verified. You can check a field. You can’t check a paragraph.
And this is where it connects to the tools we all now have. Andrej Karpathy made the point that LLMs do their best work where results are verifiable. Clean inputs, checkable outputs. The vaguer the input, the more the model guesses, and guessing at scale is just automated error.
Structure isn’t tidiness. It’s the design of the factory floor. Get it right and the same step runs a thousand times without a decision. Get it wrong and every task is a workshop again, with a smarter tool.
