In June 2026, Fei-Fei Li — the “godmother of AI” — sat down with MasterClass founder and CEO David Rogier for a podcast conversation. The title was a precise provocation: In 10 Years, Only Two Types of Workers Will Remain.
This is not another piece of AI-fortune-telling. Read the whole dialogue and you will find Li offering almost no sensational predictions. Instead she did something rarer: she pulled up by the roots the narrative that tech media and techno-optimists keep reselling, laid it on the table, and said — this story is wrong.
A Wrong Premise on Repeat
Start with the line currently fashionable in AI circles: “the cost of intelligence is approaching zero.” It sounds right. ChatGPT writes copy far cheaper than hiring a copywriter. Claude generates code faster than any junior engineer. Reasoning, data analysis, research — the tasks that once required professional training genuinely are getting cheap.
Then Li says the thing that cuts through it: “Anyone out there claiming the cost of intelligence is approaching zero is being irresponsible.”
Why? Because human intelligence is not one-dimensional. Linguistic intelligence is one layer. Above it sit perceptual intelligence, spatial intelligence, bodily intelligence, emotional intelligence. We do not yet understand where creativity comes from — on what basis is its cost zero? A teacher diagnosing why a student isn’t learning relies on more than text analysis — on reading expressions, tone, the moment of hesitation. A team lead deciding whether to say the hard sentence in front of a key client has no algorithm standing by to make that call.
This is not denial of AI’s power. Li states plainly that large language models “have shown powerful capability in business intelligence, software engineering, and deductive reasoning.” The problem is that equating linguistic intelligence with intelligence as such — a concept swap — is becoming the starting point of management decisions.
A wrong assumption, from which a series of correct-but-dangerous actions are derived. That is the scariest thing of all.
Three Automation Mindsets That Send AI Adoption Sideways
Li’s core position repeats throughout the conversation: AI is a tool, an extraordinarily powerful tool. But is the tool deployed to replace people or to lift them? The answer does not depend on the technology. It depends on people. Her discussion surfaces three of the most common automation mindsets.
Mindset one: AI as headcount replacement. Her product-manager example is precise. Ten years ago, product managers “were more like conductors. They didn’t write code; they usually weren’t software engineers.” Prototypes went to designers, development waited on engineers, and a full lifecycle — prototype, ship, feedback, integrate — took months. Now many PMs code with AI themselves, vibe-prototyping, and the cycle has compressed from months to days. AI replaced no one. It pushed everyone up one level — PMs from conductor to hands-on builder, designers and engineers from executors to the people who attack only the hardest problems. So what is the automation-minded manager’s first reaction to this same story? “Then we can hire two fewer engineers.” Same facts, same tool, opposite conclusions. The difference is not the technology. It is the understanding of the technology.
Mindset two: buying tools equals doing it right. Purchase the licenses, run the training, teach employees to write prompts — training complete, box checked. That is the standard flow of most corporate “AI transformation.” Li said something about education in the interview, and it maps one-to-one onto management if you swap “education” for “management” and “students” for “teams”: “The goal of education is to cultivate people — to let every person become a meaningful contributor to their community and society, and to live a meaningful life. AI should not strip away any of these basic goals; it should help achieve them better and more effectively.” Meanwhile Rogier offered a live demonstration. He is not from a technical background, yet he now builds every application he uses — his CEO toolstack, his productivity apps, his to-do lists — himself, with Claude Code or Cursor. The cost of making an app has fallen from months to a weekend. He is not showing off. He is demonstrating a deeper logic: the excellent people of the AI era are not “better at executing tasks” — they are “better at designing work systems.” Buy every tool on the market, and without design thinking the tools just become digitized versions of the old process.
Mindset three: announcing “full AI deployment” as a technology directive. Send a memo saying “the company will fully deploy AI,” and employees hear “layoffs are coming.” Sit down and say “let’s look at what you could do with AI that you couldn’t do before,” and employees hear “we get to be stronger.” Two framings, same tool, same budget, same people — different premise, completely different outcome. Employees hesitate not because they can’t learn the tools but because they don’t know whether management intends AI to replace them or to lift them. This is the most easily missed signal in adoption: the first threshold of AI deployment is never technology. It is trust. Li’s starter advice is disarmingly simple: find someone under twenty-five — a kid, a niece, a nephew — and with pure curiosity watch how they use AI. Once you actually understand what it is, the world is less frightening.
The Barbell Law: In 10 Years, the Middle Disappears
Now the part that stings. Rogier proposed a hypothesis about workplace structure: the barbell effect. A passable piece of copywriting — anyone with a large language model now does it decently. But if you are the best copywriter in the world, or in the top one percent, nobody can easily beat you. The barbell’s two ends are the top-one-percent experts on one side and high-agency generalists who can juggle many things on the other. The “decent” middle is being compressed from both directions.
Li agrees with the structure and adds a critical layer: “Whether you’re on the expert side or the generalist side, you need agency.” The experts at the left end are the ones who apply augmentation to its limit: AI filters out 90 percent of the repetitive work so their energy concentrates on the 10 percent that genuinely requires human judgment. Their value is not being compressed — it is being released. The generalists at the right end are the ones who initiate augmentation: they pick up the tools themselves, build their own, define their own workflows. They are not waiting for an augmented future — they are the starting point of it.
The middle’s problem is not a skills problem. It is a posture problem. AI has raised the baseline of “decent execution” to an extraordinary height; whoever remains at the “can execute” level gets caught, no matter the field. But the moment a person switches from “waiting to be told how to use AI” to “let me get my hands on it and see what it can do,” the middle regains a path to either end of the barbell. Li put it memorably: the word “entrepreneur” is, to a large extent, a synonym for “agency.”
Why Augmentation Isn’t Wishful Optimism
Here is the most interesting part. Someone will ask: what if technology keeps advancing until human judgment, creativity, and emotional intelligence are all automated too? Li answers with spatial intelligence and evolutionary biology. Her company, World Labs, researches spatial intelligence — understanding the three-dimensional world, reasoning about it, generating it, interacting within it. She uses the basketball shot as the example. Even the act of shooting a basket is a densely intelligent moment. Linguistic reasoning participates — the player registers whether the ball went in and what it means for the game. Seeing the whole court, reading every player’s position, aiming for the rim — that is deep spatiality. Adjusting the body, knowing how to produce the motion — that is deep physicality. Three intelligences — linguistic, spatial, bodily — fire simultaneously in the instant of a shot, cooperating. Not a pipeline: first language, then space, then body. Most of what we do in daily life is a blend of linguistic, spatial, and physical intelligence.
And the evolutionary judgment: it took more than five hundred million years of evolution for spatial intelligence to mature; linguistic intelligence took far less time. Spatial intelligence is a deep, ancient, foundational capability that animals and humans share. All of this points to one conclusion: what AI can genuinely accelerate today is the linguistic layer of tasks — writing reports, searching literature, analyzing data, producing code, generating images. What it buys is time and energy for everything outside that layer: judgment, creativity, empathy, deciding under ambiguity, staying calm under pressure, holding onto the one signal that matters when every signal conflicts. Augmentation is not a wish or a values choice. It is a scientific judgment about this stage of the technology: humans still hold a great deal AI cannot catch.
So What Do We Actually Do?
Li closed the conversation with a line that deserves to be the touchstone for every AI management decision:
We taught our children to use fire, then knives, then the internet. Now, as a species and as a society, we have to learn how to use AI.
The keyword is not “learn.” It is “we.” Not employees figuring it out alone. Not the IT department deploying it top-down. Managers and teams together, treating AI as something a whole organization needs to work out in the open. A closing frame for four seats at the table: if you are a manager — stop asking how many people AI can replace; start asking how many levels up AI can push each person. If you are a middle-layer executor — stop waiting to be assigned an AI workflow; get your hands on it and see what it can do, even if that just means building yourself one productivity tool. If you are a top-one-percent expert — let AI filter the 90 percent and pour the freed energy into the work that most needs judgment. If you are a generalist — keep the wide curiosity, but train agency until it is muscle memory: don’t just use other people’s tools; define your own work system. In ten years the workplace may hold only two kinds of workers. But the choice is today, and it sits in each person’s own hands.
Source: Fei-Fei Li in conversation with MasterClass founder and CEO David Rogier, on the Silicon Valley Girl podcast (June 25, 2026).
