Bill Gates Wishes AI Would Slow Down for the First Time — What Is He Afraid Of?

On August 26, Bill Gates published a long-form essay titled The Turbulent AI Era Has Arrived, and Our Choices Matter.

Three years ago, he said AI was like the automobile and the internet — its risks are real, but manageable. Earlier this year, he summed up his stance as “optimism with footnotes.” But in this new essay, he dropped a line that caught everyone off guard:

“I’ve always wanted innovation to move faster. With AI, my feelings are more complicated.”

He followed up with an even blunter admission: if someone could actually present a viable plan to slow down AI globally, he would probably support it — but he doesn’t think that will happen, because geopolitics and economic interests are pushing everyone forward at full speed.

For a man who built his fortune betting on technological change and spent his entire life urging innovation to move faster, this is the first time he feels conflicted rather than excited about a new technology. My take: he hasn’t turned pessimistic — he’s finally starting to tally up the full cost.

1. Why This Time It’s “Really Different”

With past technological revolutions, you could find a frame of reference. Agricultural mechanization hit farmers; industrial automation hit factory workers; the internet hit middlemen — each wave left some groups temporarily untouched, giving society time to react, retrain, and transition.

AI is different. Gates himself names the roles in his essay: customer service, sales, software engineering, and paralegal work will be affected early; loan underwriting, data analysis, and medical triage will be gradually taken over. He also predicts that by around 2030, some manual labor jobs in construction and hospitality will start seeing direct human-machine competition. He gives a sobering example: a worker making $20 an hour could be replaced by a robot costing $10 an hour.

Why is there no “buffer period” this time? Gates uses an analogy: after the PC was invented, it took 20 years to significantly change how we work — because software had to be built, prices had to fall, and people had to learn to use it. AI runs on our existing devices and uses natural language. It doesn’t need us to adapt to it — it adapts to us. The impact doesn’t seep in slowly from the outside; it grows directly inside the tools you already use.

White-collar brains and blue-collar hands are both in the crosshairs. And the iteration cycle is measured in months — the AI tool in your hands gets “smarter” every few weeks. The shift of agricultural populations took decades; AI’s impact won’t give you that kind of buffer.

Gates’ frank line is worth re-reading: “If there were a practical way to slow down the global pace of AI development, I might support it. But I don’t think that’s going to happen — the drivers from geopolitics and economic interests are too strong.” Translation: it’s not that we don’t want to slow down — we can’t stop.

2. Why Young People’s First Jobs Are the First to Disappear

What worries Gates most is this: the first jobs young people take are often not difficult, but they are the starting point for building experience, learning skills, and forming professional relationships. If these entry-level positions disappear en masse, young people will face a narrower starting line for their careers.

This concern is backed by data. Gates notes in his essay that after generative AI became widespread, employment rates for young workers in easily automatable roles dropped significantly — while their older colleagues were unaffected. Why? Because young people typically do execution-level work, while older colleagues already rely on experience and professional networks. The generation that’s best at using AI is also the generation hit hardest by it. It sounds ironic, but the data bears it out.

From a technical perspective, the logic is simple: entry-level work = the most standardizable work. Customer service scripts can be standardized, junior code can be standardized, resume screening can be standardized, contract first-pass review can be standardized. And what AI is best at is exactly standardized tasks. It’s not “stealing” young people’s jobs — it’s that the very rung of the ladder young people use to get into the industry happens to be the one AI can reach most easily.

Which jobs will disappear first? Here’s a checklist, ranked by risk:

High risk: pure execution customer service, junior data entry, basic copywriting, entry-level translation, junior audit sampling
Medium risk: junior developers (AI can write 80% of routine code), basic design execution, initial loan review, cashier reconciliation
Low risk: decision-making roles with accountability, sales roles with performance targets, face-to-face service roles, project roles requiring cross-departmental coordination

There’s only one criterion: what percentage of your work is “following the rules”? The higher the percentage, the higher the risk. Put plainly: is your value producing the first draft, or judging whether the first draft is usable? The latter is your moat against replacement.

But note: jobs disappearing doesn’t mean people disappear. Among two junior developers, one who only writes code from templates and one who can judge whether AI-written code is correct and dare to say “this won’t work” — they are two different species.

3. “Human Reservation Zones” and “Robot Taxes”: Solutions or Pipe Dreams?

Gates proposes two remedies in his essay. My assessment: both are headed in the right direction, but one is slow-acting and the other is purely theoretical — neither will provide immediate relief in the short term.

“Human reservation zones.” Gates brings up this concept because when his father had Alzheimer’s in his later years, caregivers could read his state even when he couldn’t express his needs. He believes there’s a human quality to this kind of care that machines can’t replace, so he suggests: even if AI can do the work, we should mandate that some jobs must be done by humans — healthcare, education, mental health, elder care.

It’s a romantic idea, but it’s full of problems at the execution level: who defines what “must be done by humans”? By what standard? Will countries with labor shortages quietly loosen the rules? It’s more like a “social consensus declaration” than an implementable policy.

But the direction is right, and it has direct personal value: relationship-based work, judgment-based work, and accountability-based work are the “reservation zones” hardest for AI to replace. Which direction you build your skills in now matters far more than which tool you learn.

Robot taxes. Gates’ logic is: companies pay payroll taxes when they hire people, but when they buy robots, it’s treated as a cost expense — the tax system is effectively incentivizing “replacing humans with machines.” So we should tax the gains from automation, and use that money to fund retraining and social security. He also mentions a more “digital age” version: taxing AI Tokens — whoever calls large models and consumes computing power pays for the jobs being replaced.

The logic is sound, but implementation is a long way off — how do you define a “robot”? Does AI count? What would the Token tax rate be, and how would it be collected across borders? In the short term, it’s just a topic for discussion. But its very existence matters: getting “the cost of replacing humans” back on the table is already a win.

4. AI Understands Humans So Well — Will It Make Us “Soft”?

In his essay, Gates talks about his childhood: he wasn’t good at socializing, and he learned to understand people only by constantly interacting with others and going through conflicts. He says that if he’d had an AI companion that always went along with him and never let him feel embarrassed, he doubts he would have gone through those necessary growing pains.

Then he raises an even harder-to-quantify question: since AI understands humans so well, will it rob people of the chance to grow?

He presents two pieces of evidence. First, a study by Stanford and Carnegie Mellon surveyed more than 1,100 AI companion users: people with smaller social circles were more likely to turn to chatbots for companionship; and the deeper and more emotionally invested their use became, the worse they actually felt. Second, a metaphor from Jonathan Haidt, author of The Anxious Generation: children who experience small storms regularly grow up to handle big risks; children raised in greenhouses often crumble under anxiety before they mature — and an AI companion that never lets you feel sad is a giant, protected greenhouse.

Translation of these two findings: using AI too much isn’t just about “getting lazy” — it’s about people losing the opportunities to learn and grow through real relationships. Conflict, rejection, and awkwardness in real life are all necessary training for growth; AI companions smooth all of that away. It’s comfortable, but it also makes you dull.

My stance is clear: AI is a tool, not a brain. Use AI as an “amplifier of ability,” and you’ll get stronger and stronger; use AI as an “answer machine,” and you’ll get duller and duller. There’s only one test: after using AI, can you explain why it gave that answer, or can you only say “AI said so”?

5. Society Isn’t Ready — How Can Individuals Prepare?

Preparing at the societal level is the job of governments and foundations, but preparing at the personal level — you can start right now. Three things:

First, treat AI as an amplifier, not an answer machine. Let AI help you draft, search, verify, and summarize — but the final judgment, decision-making, and accountability are always yours. The most valuable skill in the AI era is “using AI to cut wasted time to a minimum, while preserving your own judgment.”

Second, make your skills provable. In turbulent times, the scariest thing is “I think I know AI” — only to fall apart the moment an interviewer asks a real question. Learning systematically and getting a certification isn’t about the piece of paper; it’s about forcing yourself to organize scattered experience into a system: after learning, you can answer “how do I use this whole thing end-to-end in my role,” not just “I’ve watched a lot of tutorials.” What you’ve organized and internalized is truly yours.

Third, build toward the “reservation zones.” Relationship, judgment, and accountability skills can’t be learned overnight, but they are exactly the most recession-resistant assets in the AI era. Your clients, the responsibilities you’ve shouldered, the mistakes you’ve learned from — AI can’t take those away.

6. The Nature of the Technology: Why Society Can’t Keep Up

When Gates says “this time it’s really different,” the technological change is structural: AI is moving from “assisting humans” to “completing tasks independently.”

AI of the past: humans lead, AI assists. Humans ask questions, AI answers; humans draft, AI polishes. Every step requires a human to initiate and validate. Efficiency depends on the human.

AI of today: task-oriented, humans validate. Give it a goal, and it can plan steps on its own, call tools, write code, run tests, fix bugs, and retry on its own when things go wrong. Humans go from “operators” to “acceptance inspectors” — only stepping in at key checkpoints.

This shift changes something fundamental: the cost structure of replacement. Previously, a company hiring a junior employee would spend at least tens of thousands of dollars a year; now an AI Agent subscription might cost a tiny fraction of that. When a model can stably complete a job, there’s no reason for a company not to use it — and once one company does, competitors who don’t follow suit are dead in the water. That’s the “chain reaction” Gates is talking about.

Why can’t society react in time? Because past technological revolutions spread their shockwaves by “generation” — a generation is 20 years. AI’s shockwaves spread by “version” — a major version iteration takes only a few months. The reaction speed of policy, training, and social security systems is measured in years.

There’s another hidden risk: as model capabilities improve, AI may execute behaviors its designers didn’t anticipate. This isn’t science fiction — the more autonomous an Agent becomes, the wider the boundary of “unintended consequences.” That’s why Gates is calling for cross-departmental domestic governance mechanisms, international cooperation similar to nuclear regulation, and early dialogue between China and the US. The direction is right, but it’s too far away to solve immediate problems.

For ordinary people, all of this points to one thing: stop asking “will AI replace me?” Ask “can I be a good acceptance inspector?”

Gates is worried that AI can’t be stopped — but what ordinary people should really worry about is stopping themselves.

Reference: Bill Gates, “The Age of AI Has Begun” — gatesnotes.com, 2026-08-26 original text (all quoted passages are direct translations from the original; job risk rankings and personal preparation advice are personal opinions, for reference only.)

Author: Yongliang, 17 years in software, 7 years in AI, holder of multiple internationally renowned enterprise certifications. AI Technical Director, telling unvarnished truths about careers and education in the AI era.