Summary and commentary. Source: Dror Poleg, Your Job Can Survive AI. Your Salary Might Not., October 27, 2026.
Tool and medium
Most AI debate comes down to whether the machine will take your job. Dror Poleg calls the usual analogy, AI as a tool like the power loom, the typewriter or the computer, useful but incomplete. AI is also a medium, like the phonograph, radio or the camera.
The difference: a tool extends what a person can do, while a medium extends who they can reach. A medium separates a performance from the performer and distributes it. It rarely wipes out a profession. Instead it changes the profession’s income distribution, job stability and path to success. The essay’s key line: AI does not have to replace you to reprice you.
The musicians’ case
Recording, radio and streaming did not wipe out musicians: their share of the American labor force has barely changed since 1970. But the profession became a different one. Technology gave musicians audiences far beyond their hometowns and, at the same time, put each of them up against the best performers in the world. Today the top stars are billionaires, while most musicians, according to an NEA report, earn less than the national median salary.
YouTube, TikTok and Spotify added millions of people uploading their own performances. They don’t show up in employment statistics, but they compete for the same attention and the same money. As a result, even top performers depend on algorithms and shifting attention rather than traditional measures of quality.
Superstar markets
The same thing happened in creative professions long before AI. Once communication and transport let one person reach a huge audience and the barrier to entry drops, a “superstar” market appears; economists have been studying it for at least 45 years. Poleg shows it in physical professions too: fitness instructors and OnlyFans performers also have a handful of stars and a huge field of competitors. An online workout is not a perfect substitute for a live one, but it already reshapes the economics of the whole profession.
AI does the same thing, harder and wider. Top professionals amplify their impact, and their skills get packaged and made available to anyone. Poleg’s forecast: fewer, larger winners; a longer tail of semi-professionals; a hollowing middle. Access to expertise widens, but the share of people who can make a dependable living from it does not necessarily follow.
Silicon Valley as a preview
In 2025 Meta, Google and Microsoft reportedly offered elite AI researchers packages worth hundreds of millions of dollars, like NBA stars. At the same time, the same companies laid off thousands of experienced employees and grew without the hiring such growth used to require. Polarization is already visible at the very top of the labor market.
“Distinctly human” work is a losing bet
The usual advice is to move into work that is more creative, harder to measure, “distinctly human.” But such work, Poleg points out, tends to be less stable and less equal. And AI is learning it fast too: people happily consume AI-generated content, eight Pulitzer awardees disclosed using AI, writers are accused of letting a model write for them, and AI is solving problems mathematicians have struggled with. Racing ahead of the machine toward tasks that seem “distinctly human” is a losing strategy.
For Poleg, the only thing truly human is being an actual person. People will pay a premium for human attention, opinion and creation, not because it is better but because it is human. Even before AI, many people were paid for work software could do: meetings that could have been emails, questions that could have been a form, personal service. AI will free people, or force them, depending on your view, to do more of that work. But demand for human attention does not guarantee a middle-class income.
What to do about it
We can’t stop AI any more than the musicians of the 1920s stopped the gramophone. But musicians did win something: in 1948 they negotiated an agreement under which a share of record sales went into a trust fund, which has paid for live performances across America ever since. Poleg’s conclusion: we need institutions that turn greater productive capacity into broadly shared prosperity and let people build stable lives while the market value of their work keeps changing.
My take
In software I see the same picture Poleg describes with musicians. Agents don’t remove the programming profession; they change its economics. A strong engineer with agents covers the scope a team used to be hired for, and Agentic Engineering turns their experience into processes and prompts others can use. The risk is not that programmers disappear but that the middle of the market, people hired for clear, well-specified tasks, comes under pressure.
That leaves a practical question for a manager: if agents raise the output of the best, how will you grow the next best? Junior tasks were not just cheap labor; they were training. If agents take them, the path into the profession has to be built on purpose. It’s the same question about institutions Poleg asks about society, at the scale of one team.
It also connects to his previous essay, The Interface Is the Org Chart: there, agents put pressure on departments that only route information; here, on people whose work came down to a replicable skill.
The full argument is in the original.