Digital brains / 1 Jun 2026

Building a digital brain

Everything you know, and everything you have ever been curious about, is worth more if it lives in one place a machine can read. Companies need this now. People will need it soon. I keep hitting the same wall inside real businesses, and it is always the same shape. There is not enough context.

The bottleneck is not the model. It is context.

The models are extraordinary and getting better every month. That is not where the work gets stuck.

Ask a frontier model about your business and it answers like a clever stranger. It is not short of intelligence. It is short of information about you. Andrej Karpathy has a clean way to picture this. He describes the model as a new kind of computer, where the model itself is the processor and the context window is the working memory. A processor with nothing loaded into memory is just an expensive idle chip.

So the real skill is not writing a clever prompt. It is putting the right information in front of the model at the right moment. Karpathy calls this context engineering: the art of filling the context window with just the right information for the task in front of it. The prompt is a sentence. The context is everything the model needs to know to answer that sentence well.

We discover and buy by conversation now

Something changed in how people find businesses. They used to browse. Now they ask.

People discover a company inside a chat, get their questions answered inside a chat, and decide inside a chat. This is the conversational economy, and it is already how a large share of buying starts. The question every business now faces is simple and uncomfortable. When someone asks, do you have an answer ready for every reasonable question they could possibly have?

For most companies the honest answer is no. They have a help page. They have maybe twenty FAQs, written quickly years ago and never touched since. That was enough when a patient human read them and filled the gaps with their own judgement. It is nowhere near enough now, because the thing reading them is a machine answering on your behalf, and it can only repeat what it has been given.

• Twenty thin FAQs cannot cover sizing, fit, edge cases and the question nobody thought to write down.
• A machine does not improvise your refund policy. It looks it up, or it guesses.
• The gaps in your knowledge are exactly where a confident wrong answer gets made.

Answers that work while you sleep

We have always solved this with people. Hire customer service reps, train them, let them carry the answers in their heads. That still matters. But more and more of the first response is now handled by AI, and AI cannot carry anything in its head. It can only read what you wrote down.

This is where it stops being a cost and starts being leverage. Naval Ravikant has a line I think about often, that code and media are permissionless leverage, the kind that works for you while you sleep. A well-built digital brain is exactly that. Write an answer once, properly, and a machine can give it ten thousand times, across three time zones, at three in the morning, without getting tired or going off-script.

This is true for a person, not just a company

The same logic runs at the scale of one human. Everything you have learned, every problem you have solved, every subject you fell down a rabbit hole on, all of it is worth more collected in one place than scattered across your memory and forty browser tabs.

Naval calls this specific knowledge: the kind you cannot really be trained for, found by following genuine curiosity rather than whatever is fashionable. It is your edge precisely because nobody else has the same pile. But a pile in your head does not compound, and it cannot be queried. Written down, it becomes something you and a machine can both build on.

This is why I keep my own. I call it Fonda OS, a private store of context, decisions and notes that feeds every AI session I run. It is the reason I do not re-explain myself from scratch each time. Build the same for yourself and it pays back across your résumé, your work, your products and the software you make. The brain is the asset. Everything else is output from it.

Write it so a machine can read it

There is one more move, and most people miss it. Your knowledge has to be in a form a machine can actually use.

Honestly, the way AI reads and uses a knowledge base is still being worked out in real time. Retrieval, context windows, agents, all of it is moving. But the safe bet does not depend on which technique wins. The safe bet is to write more of it down, in plain language, in the open, in formats that are easy to parse. Karpathy makes this point about documentation. A lot of it was written for a human clicking through a screen, and it needs rewriting for a machine reading it directly.

• For a company, that means real help centres, real articles, real depth on products, sizing, policies and the systems behind them.
• For a person, it means writing your thinking down somewhere durable instead of letting it evaporate.
• For both, it means plain text and markdown over buried PDFs, clear pages over "click here", and treating the knowledge as the product that machines, not only people, have to read.

The rule

Stop tuning the model. It is already smarter than your information about yourself.

Write down what you know. Then keep writing it down. The business and the person with the richest, most legible brain will win the conversational economy. Not because they had the cleverest AI, but because when the machine went looking for an answer, theirs was the only one that was actually there.