Bill Gates has published a nearly 6,000-word essay on Gates Notes, calling the AI transition "one of the most turbulent times in human history." In it, he proposes a tax on AI tokens and robots to offset the payroll-tax advantage that automation already holds over hiring, and puts a ceiling of roughly 40% of jobs under the most aggressive version of that carve-out.

Buy a worker, pay payroll tax. Buy a robot, write it off. Bill Gates thinks the math employers face is a problem.

In an essay titled "The turbulent AI era is here. The choices we make now are critical," the Microsoft co-founder argues there is an asymmetry in the current job market: employers pay payroll taxes when they hire a person, but a robot is generally written off immediately as a business expense. "The tax system nudges you toward replacing people with machines," he wrote.

The stakes are fiscal as much as social: payroll taxes are the primary funding stream for Social Security and Medicare in the United States, so a large-scale shift from human labor to untaxed automation would erode the revenue base those programs depend on.

His proposed fix is a tax on AI tokens—the units language models use to process text and other data—plus a tax on robots themselves.

It is not a new pitch. Gates first floated a robot tax in 2017 and was dismissed by economists who called it a tax on productivity dressed up as a tax on progress. The same year, South Korea took a narrower step—limiting tax incentives for automation investment rather than imposing a direct robot tax—making it one of the few governments to move on the idea at all.

He is not backing down this time. Gates believes a usage tax changes employer incentives directly, unlike a corporate tax that only cuts into profits, which can run thin in a hypercompetitive market. Labor is already taxed through payroll levies; robots depreciate and owe nothing to Social Security.

The jobs no machine gets to touch

The essay's second idea is a category of work Gates calls "Human Reserved"—jobs society deliberately keeps for people even when AI could technically do them. His analogy is a nature reserve: land where you could legally build, but don't, because what stands there is worth more left alone.

Childcare and jury service are his clearest examples. Gates has written previously that his father received round-the-clock care before dying of Alzheimer's in 2020, and that "something in the care they gave my dad was irreplaceably human." Education and healthcare sit in a fuzzier middle tier, where he expects humans to remain in the room while leaning on AI to cover more ground.

The concept has a real-world analogue: some jurisdictions already restrict automation by law rather than economics—several U.S. states, for example, require a human operator behind self-driving vehicles. Gates's proposal extends that logic from safety rules to labor policy.

Gates capped the idea at roughly 40% of jobs under its most aggressive form, saying he could not push the number higher without the argument falling apart. He also split the protection into two flavors: permanent, for roles like caregiving, and temporary, aimed at workers—his example was a 55-year-old who has spent his whole career in construction—who cannot realistically retrain into an unrelated field on short notice.

Why he thinks this transition breaks the pattern

Gates spends much of the essay arguing that the usual comparison to past technology shifts does not hold. The move from farm work to office work took generations and created new categories of labor that required human judgment. AI, he argues, skips that step: it can substitute directly for human cognition in law, customer service, medicine, software, and manufacturing—and it can do so within a single decade instead of several.

Anthropic CEO Dario Amodei has made a version of the same argument, putting a number on it: up to half of entry-level white-collar jobs will be gone within five years.

The data has not caught up to the warning yet. A Yale Budget Lab and Brookings study published in October found no sign of mass displacement in federal employment data through July. California's newly launched AI-unemployment tracker tells a similar story so far—no statewide layoff wave—though it did flag rising unemployment claims among college-educated workers in AI-exposed occupations.

Gates is not waiting on the data. "I'd love to be convinced that I'm wrong about the job market, but I'm not," he said, adding that he would stake his reputation on AI turning out "very bad for the job market."

Gates frames the essay as a fork in the road: AI ends up "either the greatest equalizer ever invented, or the worst source of injustice," and which one we get will be decided in the next few years, not the next few decades. He is calling for new national coordinating bodies and an international institution to manage the transition—work he says must start now, while the robot tax he first pitched nine years ago is still waiting for a government willing to try it. That call lands amid an active global policy debate: the EU's AI Act, the U.S. executive orders on AI, and international summits on AI safety have so far focused on regulating the technology itself, leaving the question of who funds social programs if automation displaces the workers who fund them largely unaddressed.