Between 1933 and 1961, exactly 43 people passed Lev Landau's entrance exam.
He did not call it the theoretical maximum. He called it the theoretical minimum: the core you cannot do serious work without.
The exam worked like this. You arrived at his apartment and left everything at the door. Books, notes, every scrap of paper. Inside was a round table, blank sheets, and a problem. Landau wandered in every twenty minutes and looked over your shoulder. He was not checking your answer. He reused the same problems for years and did not care that students compared notes, because the approach was the thing being examined. Whatever you produced in that room had to be rebuilt from what you actually held in your head.
Forty-three people in twenty-eight years. Ten of them ended up in the Academy of Sciences.
Two developers, one accidental proof
This week I watched two game developers argue about AI coding and accidentally prove Landau right.
Jonathan Blow spent ten years on his current game. He wrote his own programming language and his own engine to build it. His verdict on AI coding tools: they cannot hold the constraint web of a large codebase, the thousand quiet agreements between systems that every new line has to honor. Ask one to fix a deep bug and it forgets half the constraints while satisfying the other half.
Thomas Brush, interviewing him, pushed back with his own numbers: a thousand playtest bugs and feature requests cleared with Claude running at his side, on a game holding 60 frames a second.
Opposite verdicts. Except they were not arguing, and neither of them noticed.
Brush built his codebase himself, over four years, from the ground up. When something breaks he can tell the AI where to look and what correct looks like. He holds the constraint web. The AI just walks it for him. Blow is right about what the tool cannot hold. Brush is right about what it can do for someone who holds it himself.
The variable is not the model. It is whether the person delegating could rebuild the foundation on blank paper.
That is Landau's exam, applied to delegation.
The ledger from my own desk
Here is what that looked like at my desk this week, zeros left in.
The game I am building has an AI art pipeline. The ledger says it generated 102 assets. The number that shipped: zero. This week I rejected the entire direction, and the rejection took one sentence, because the thing the art failed against is a bar I hold personally. I did not need to re-derive it, look it up, or ask anything to evaluate it for me. I knew it the way Landau's students knew mechanics.
The generation was delegated. The bar never was.
That is not a story about AI art being bad. A hundred and two assets was cheap tuition for finding the bar's exact edge. It is a story about which half of the work was safe to hand over. If the bar had been generated too, I would still be scrolling through those 102 assets asking each one is this good? with nothing to measure it against.
The rule
Your theoretical minimum is whatever you must be able to rebuild on blank paper. Delegate everything else freely, and that, never.
The test is not "do I understand this." Understanding is what everyone claims right up until the page is empty. The test is Landau's: close the book, blank sheet, rebuild it. Whatever survives that test, you can check, which means you can delegate the labor around it and catch the failures. Whatever does not survive it, you cannot check, and delegation without checking is not delegation. It is hoping.
Most people running AI hard are accumulating in exactly the direction Landau warned about: more tools, more output, more collected material, while the holes in the core quietly grow. The fix is not less AI. It is knowing, in writing, which three or four things in your field everything else is built on, and drilling those on blank paper while the machines do the rest.
What is on your list? Not the whole job. The part you would stake the rest on. Hit reply. I read everything.
Christopher
Wisdom Nexus is written the way it is lived, with AI in the loop and a person at the desk who cares how it lands. This is where the theory meets the practice, shared with you in the open.