Not “if it takes your job.” I mean Everything. Every task. Every role.

We keep getting told: “don’t worry, humans always adapt, new jobs will appear.” 🙄

Except the people saying that can’t tell you what those jobs are. Or when they’ll show up. Or who gets them.

Here’s what’s actually keeping me up at night 👇

🏁 Three companies are setting the pace of the most consequential technology race in human history. Their internal safety timelines, their real capability levels, their disagreements with each other — most of that never reaches daylight. We’re spectators to a race we can’t see the scoreboard of.

⚖️ Regulation is sprinting to catch a plane that already took off. By the time a policy is drafted, debated, and passed, the model it was written for is two generations old.

💰 If AI really can do “everything,” the question stops being about jobs and starts being about who owns the compute. Because if machines do the work, the money doesn’t automatically follow the humans anymore. It follows the owners.

🧠 Even the researchers and safety teams closest to this don’t agree on the basics — how fast, how far, how dangerous. When the experts are split, “trust the process” isn’t a plan.

So no — I don’t think “we’ll adapt, we always do” is good enough this time. That answer worked when machines replaced muscles. This is the first time they’re coming for judgment too.

But I know pretending this is business as usual is the most dangerous position of all.

📚 A few reads that shaped this post:

International AI Safety Report 2026

Yale economist Pascual Restrepo on what happens to labor markets if AI can do “essentially all economically valuable work”

Inside the race for power (literally) — why electricity, not chips, is now the bottleneck on AI growth

Who’s actually winning the AI race — models, money, compute and distribution, mapped out

80,000 Hours’ plain-language breakdown of “loss of control” risk — what it means and why experts disagree so much