AI is cheaper than workers. Bill Gates wants to change that

AI is about to change everything, and we’re not ready. That’s the overarching message of Bill Gates’s 6,000-word essay about his growing worries over AI, and my initial thought after giving it a skim was this: Tell me something I don’t know.

The disruption that AI’s about to cause will be unprecedented, Gates writes. It could put weapons of mass destruction into the wrong hands. Even worse, it could devastate the job market, with world-altering consequences. We need a plan, and we don’t have one.

Haven’t we heard all this before?

To be fair, I don’t think the legendary Microsoft co-founder — who, famously, was late to the internet revolution — just realized that AI could be a problem. If anything, he seems less alarmed by the threat of AI itself (and make no mistake, Gates sounds very alarmed by AI) than he is appalled by the laissez-faire, “let the market sort it out” attitude of those in a position to actually do something.

Gates’s AI manifesto gets more interesting once he arrives at his proposals, even if the first — a category of “just for humans” jobs that he dubbed “Human Reserved” — sounds more like a rhetorical talking point than an actionable plan. 

The idea of uniquely human jobs, including roles such as caretakers and teacher, is a laudable one. But as Gates himself asks, who precisely would be in charge of labeling specific jobs as Human Reserved, and who would enforce it? And let’s face it: The very idea of “Human Reserved” jobs is more than a little dystopian and creepy (or maybe it’s just a sign of the times).

But Gates’s other proposal is far more concrete, albeit certain to be controversial: taxes.

Specifically, we’re talking a tax on AI tokens and robots, and it’s not a new idea. Basically, businesses would need to pay taxes on every AI token they use, adding a new element to the equation when it comes to employers deploying AI in the workplace, and whether to replace human jobs in the process.

As Gates writes, the current tax system actually encourages replacing human workers with AI, given that employers must cough up cash for payroll taxes but may write off AI costs as business expenses. 

An AI token tax could flip the equation, deleting the current built-in incentive for employers to replace people with ChatGPT, Claude, or Gemini, while simultaneously funding programs to help assist and retrain displaced workers.

Would big business hate the idea of an AI token tax? Sure, but I can also think of millions of AI-leery voters who’d likely warm to it.

Of course, the actual implementation of an AI token tax wouldn’t be that simple. As any Claude Fable or GPT-5.6 Sol user will tell you, not all AI tokens are created equal, and they’re certainly not priced that way. How would the tax code keep abreast of the latest AI models? Would certain AI tokens be tax-exempt? Wouldn’t there be ample opportunity for token-tax loopholes?

An AI token tax would be messy, make no mistake. But if you can allow me a movie quote here, it might be the best bad idea we have.

More in AI this week

• It wasn’t just a few rogue AI agents that coordinated the Hugging Face hack, OpenAI now says — it was more like hundreds of them. (BBC)
• I had a good time tinkering with a new ChatGPT plugin for Apple Messages, right up to the point where ChatGPT took control of my Mac without warning. Spoiler alert: It was my fault. (PCWorld)
• Speaking of ChatGPT (when am I not?), these are the five ChatGPT settings I’d tell a new user to change first. (PCWorld)
• Apple just announced refresh Mac minis that are tailor-made for running local AI models. Here’s the configuration I’d get, were it not for today’s insane PC hardware prices. (PCWorld)
• What’s the biggest problem with AI-generated summaries and AI notetaking apps? As one AI user observes, “it’s hard to finish an idea that isn’t yours.” Well said. (Second Brain)

Prompt of the week: The “no more teases” prompt

“Want me to draw up a list of possible topics for your new podcast? If you like, I could subcategorize that list by topic and draft the episode summaries.”

If those “want me to” and “if you like” teasers sound familiar, it’s because ChatGPT, Claude, and Gemini are still desperate to keep you engaged, and they’re not shy about trying to extend your conversations with suggestions for more tasks.

I’ve devised a prompt to keep your favorite AI chatbot from baiting you with more work, limiting its questions and suggestions to only the mission-critical ones.

That’s all for now!

Thanks for reading the latest issue of Prompt Mode.

Want more next week? Don’t forget to sign up to start receiving this newsletter in your inbox.