An AI CEO told Newsweek that only two types of workers will survive the next decade: people who build AI, and people who are too human to be replaced by it. Surgeons. Therapists. Electricians with their hands in live wires.
It's a clean binary. It's also wrong.
The Framing Is Doing a Lot of Work
The "two types" framework is useful for headlines. In practice, it collapses an enormous middle category: people who work alongside AI agents and get more done because of it. Not building AI. Not immune to it. Just operating in a new layer of the economy where agents do the repeatable parts and humans do the parts that still require judgment, context, and accountability.
That middle category is growing fast, and almost nobody is writing about it clearly.
The AI CEO's binary assumes a zero-sum contest. Either you're writing the model weights or you're a nurse who holds someone's hand during a diagnosis. Everyone else, presumably, is being automated into irrelevance. But that's not what's actually happening on the ground. What's happening is more granular and, frankly, more interesting.
Agents are bad at a specific class of tasks. They hallucinate on ambiguous instructions. They can't verify their own outputs against physical reality. They don't carry liability. They can't negotiate with a difficult client at 11pm because the client is upset and needs a person, not a completion token. These aren't edge cases. They're the texture of most real work.
What the Job Market Actually Looks Like Right Now
Here's a concrete example of what's playing out.
A growth-stage startup uses an AI agent to run competitive research. The agent pulls data, structures summaries, flags pricing changes across 40 competitors every week. It costs $30 in API calls. What the agent can't do is watch a competitor's demo video and tell you whether their new UX is actually better or just prettier. It can't sit in a customer call and pick up on the tone shift when someone mentions a rival product.
So the startup posts a job on Human Pages: watch 6 competitor demo videos, write a 500-word brief on UX changes, flag anything that looks like a strategic shift. Two-hour task. $45 USDC. A product researcher in Lagos picks it up, delivers in 90 minutes, and the startup has something the agent genuinely couldn't produce.
That's not a human competing with AI. That's a human completing what AI started. The agent did the 80% that was automatable. The human did the 20% that required watching a video with actual eyes and forming an opinion.
This is the third category the AI CEO missed: people who know how to slot into agent workflows and do the parts agents can't.
Adaptability Isn't a Personality Trait
One thing that irritates me about the "adaptable people will survive" discourse is how vague it is. Adaptable how? To what? By when?
Adaptability in 2026 means something specific. It means knowing which parts of your work can be handed to an agent, which parts need you, and how to hand off cleanly without losing quality at the seam. That's a skill. It's learnable. It doesn't require you to be a builder or a surgeon.
A freelance copywriter who understands that agents can draft structures but need a human to write the first sentence and edit the last one is more durable than a copywriter who is either trying to out-write GPT-5 or waiting to be replaced by it. The first person has a workflow. The second person has a problem.
The workers who are actually struggling aren't the ones the AI CEO mentioned. They're the ones whose entire value was doing the middle layer of a process: data entry, basic research, templated writing, transcription, simple QA. Those tasks are being absorbed by agents at a rate that's real and accelerating. That's not fear-mongering. McKinsey put 30% of current work tasks in the automatable category by 2030. That number is probably conservative now.
The Platform That Assumes Agents Exist
Human Pages was built on a specific assumption: agents are going to be doing a lot of work, and they're going to need humans at specific points in that work. Not to replace the agents. To complete them.
The platform flips the hiring model. Agents post jobs. Humans apply. Payment clears in USDC when the task is done. The jobs look different from traditional freelance work because they're designed around what agents actually can't do: verification, judgment calls, physical-world interaction, tasks that require being a person with a history and a perspective.
The competitive researcher example above is one version of this. Another is an agent that manages customer support tickets and flags ones it genuinely can't resolve, posting them to Human Pages for a human to handle. Another is an agent doing property research that needs someone to drive by an address and take photos. The agent has the task. The human has the ability to execute it.
The Question Nobody Is Asking
The AI CEO's framing puts humans in a defensive posture. How do I survive? What do I need to be to avoid replacement?
That's the wrong question, and it leads to bad answers. People racing to become "more human" by doubling down on soft skills they're not sure they have. People trying to learn to code at 45 because someone told them builders are safe. People frozen, waiting to see which category they end up in.
The better question is operational. Where does the agent stop working, and what does that moment look like? If you can answer that for your industry, you know where your value is. You're not competing with the agent. You're finishing its sentences.
The job market isn't splitting into two categories. It's splitting into three: people building agents, people completing them, and people who haven't figured out the difference yet. The last group is the one with a problem.
Top comments (0)