The Memory Bread
When I was a kid, I watched an anime called Doraemon. There is a gadget in it that I have never forgotten: the memory bread. You press a slice of bread onto a page of a book, the words soak into it, and when you eat the bread, you know everything on that page. No studying, no flashcards, no years hunched over a desk. Knowledge, transferred at the speed of a bite.
Every child who saw that wanted it, and every adult knows why it stings. Because the real version of knowledge transfer is the slowest process in human civilization. We spend fifteen or twenty years in school loading the basics into one mind. Then we spend another decade on the job, making the mistakes, absorbing the exceptions, slowly turning information into something rarer: judgment. The knowledge of what to do, in what order, with what exceptions. What I call expertise.
And here is the strange, tragic part. After all those years, the expertise stays trapped. It lives in one head, applied one meeting at a time, one client at a time, one company at a time. When an expert retires, or burns out, or gets laid off, most of what they know simply leaves the world. We have built rockets and networks and machines that write poetry, and our best technology for moving expertise from one mind to another is still, essentially, school.
This is the problem I think about every day. It is why we named our company Expertise AI, and why our mission is a single sentence: make knowledge and expertise transfer between humans and AI agents the fastest and most reliable in the world. I want to build the memory bread. Not for trivia on a page, but for the hard-won judgment of people who have spent ten years getting good at something.
I am writing this because 2026 is the first year that has ever been possible, and because the people I most want to reach, the operators who hold that judgment, are standing at a fork in the road that most of them have not fully seen yet.
The year execution stopped being scarce
For all of economic history, if you knew how to do something valuable, you earned by doing it. The knowing and the doing were welded together, and the doing was the constraint. A brilliant consultant could serve maybe a dozen clients a year. A great operator could build systems for one company at a time. Expertise was bottlenecked by execution, and execution was bottlenecked by the human calendar.
AI agents broke that weld, and they broke it very recently. Not the chatbots of 2023, which could talk about work, but the agents of now, which do work: multi-step, tool-using, unsupervised stretches of real execution inside real systems. McKinsey's research puts numbers on the shift: 88 percent of organizations now use AI in at least one business function, and agentic AI, the kind that carries out work rather than answering questions, is already being scaled by roughly a quarter of enterprises, with another 39 percent actively experimenting. The firms selling this transition are drowning in demand for it; Accenture alone booked over four billion dollars in generative AI revenue in a single fiscal year.
I want to be precise about what this means, because it is easy to hear "AI does work now" and miss the economic earthquake underneath. When execution becomes abundant, everything priced as execution collapses in value, and everything that directs execution becomes the scarce input. The hour loses. The judgment wins.
You can watch it happening in the data. Since ChatGPT arrived, freelance postings for writing have fallen by a third, translation by a fifth, customer service by a sixth, and by 2025 eleven of twelve major categories on Upwork were shrinking. The great consulting firms, whose product was always brilliant people billed by the hour, are cutting the junior pyramid that model was built on: McKinsey has shed thousands of roles with more announced, and firms across the industry report that AI now does in minutes what analysts used to deliver in weeks.
But look closer at the same data and you find the part almost nobody talks about. Spending per client on Upwork went up 8.3 percent even as the commodity work vanished. Specialists in narrow, hard-won niches are earning more than ever. The senior operators inside the consulting firms remain in fierce demand while the leverage beneath them disappears. Execution is being repriced toward zero, and genuine expertise is being repriced upward, at the same time, in the same markets.
The lesson is not that AI is coming for what you know. It is the opposite. What you know just became the most valuable thing you own. The question of the decade is who captures that value.
The fork: extraction or ownership
There are two answers forming right now, and this spring gave us a perfect picture of the first one.
In April, Meta began installing software on its US employees' work computers that records their keystrokes, mouse movements, clicks, and screenshots across more than two hundred applications, in order to train AI agents on how skilled humans actually work. There was no opt-out. More than sixteen hundred employees petitioned against it. The rollout landed alongside thousands of layoffs, and the workers drew the obvious conclusion in their own words: they feared they were training their replacements. Then the collected data, which included private conversations and personal records, was found sitting exposed across the company's internal systems, and the program was paused in embarrassment.
Strip away the scandal and look at the underlying arrangement, because the arrangement is the point. The expertise of thousands of skilled people was treated as a deposit to be mined. They supplied the judgment; the company kept the asset. That is the extraction model, and Meta is only its most honest practitioner. Softer versions of it are everywhere: every tool that quietly learns from its users' work, every playbook that stays behind when the person who wrote it is walked out the door.
The second answer is ownership, and it is the one we are building.
In the ownership model, the expert packages their know-how deliberately, on their terms, as an asset with their name on it. At Expertise, we call these packages skills: a working process, the logic and judgment of a real operator, turned into something an AI agent or a team can install and run. The playbook itself stays private to its creator, not exposed, not copyable. What others get is the ability to run it. What the creator gets is a business: every run is attributed and paid, their name travels with every install, and when they improve the skill, everyone running it moves to the better version. One team adopting your skill is a customer. A hundred teams is an income that compounds while you sleep.
Think about what Shopify did for people who make things. Before it, you rented shelf space in someone else's store on someone else's terms. After it, you owned the storefront. We are building that same shift for people who know things. Your expertise should never be trapped in your calendar, and it should never be extracted from you. It should be an asset you own, and a business you build on it.
Between intelligences
I said our mission is knowledge transfer between humans and AI agents, and I chose those words carefully, because the long arc here is bigger than any single product.
Agents are not finished getting capable, and no matter how capable they become, there will always be gaps between what one intelligence knows and another does not. A human who has run enterprise deals for ten years knows things no model was trained on. An agent tuned on one company's motion knows things another agent does not. Human to agent, agent to human, agent to agent: the future is full of these gaps, and every gap is a place where expertise wants to flow. Whoever makes that flow fast, reliable, and fair to the people who supply the expertise will be building one of the most important pieces of infrastructure of the next twenty years.
That is the world I am trying to build toward. A world where the twenty years it takes to load a mind is not the price everyone pays forever. Where a great operator's judgment can outlive their tenure, their employer, and their calendar. Where the memory bread is real, and the person whose knowledge is on the page is the one who gets paid when someone takes a bite.
If you are one of these people
Maybe you are. Maybe you are the person whose scoring model the whole team quietly depends on, or whose outbound system gets rebuilt from memory at every company you leave, or whose way of running pipeline is the thing your old colleagues still describe to their new coworkers. You have spent years turning experience into judgment, and until now, the only way to earn from it was to show up.
The window to change that is open, and windows like this do not stay open. Categories get settled early. The operators who packaged their video knowledge early became the YouTubers everyone else studies; the writers who claimed their beats early own those beats today. The same settling is about to happen for operational expertise, and the names that get attached to the standards will be attached for a long time.
We are forming the first cohort now. We call it the Verified Operator Program: a small, vetted group of operators who publish the workflow strong teams already trust them for. We handle the productization, the infrastructure, the billing, and the buyers; you bring the process and the judgment behind it, and you keep control of both. The first cohort is focused on go-to-market expertise, because that is where we can make operators successful fastest, but the vision has never been one category. The vision is any expertise, moving between any intelligences, at the speed of a bite.
If that is you, the door is at expertise.ai/operator-program. Walk us through what you know. We will show you what it can become.
