An Anthropic engineer goes on record saying AI agents will transform every computer-based job in America, and that the process will be painful. The response from most tech media was to treat this as a hot take. It isn't. It's an operational description of something already happening.
The word "painful" is doing a lot of work in that quote. It's not a warning about some distant future. It's an acknowledgment that the transition is structurally disruptive in a way that doesn't have a clean narrative arc. No villain, no hero, no moment where everything snaps into place. Just a prolonged period where the rules of what work is worth paying for get rewritten, and a lot of people are caught in the middle.
We built Human Pages inside this exact moment. Not because we think AI replacing humans is good or bad, but because the binary was always wrong.
The Replacement Narrative Is Lazy
Here's what the replacement narrative misses: AI agents are not good at everything. They're extraordinarily good at pattern-matching, retrieval, code generation, and structured reasoning. They are genuinely bad at ambiguity, physical verification, novel judgment calls, and anything requiring real-world accountability.
A legal AI agent can draft 200 contract summaries in an afternoon. It cannot call the counterparty's attorney and read the room. A customer service agent can handle 90% of inbound tickets autonomously. That last 10% is where customer relationships are actually won or lost.
The Anthropic engineer isn't wrong that transformation is coming. But transformation is not the same as elimination. It's more like compression. The ratio of human effort per unit of output is changing, which means the type of human effort that survives is changing.
What survives is judgment. Accountability. The ability to make a call that a machine can't make without someone's name attached to it.
What the Labor Market Actually Does With This
The standard economist response to automation is "workers shift to new tasks." True in aggregate over decades. Not particularly useful if you're a 42-year-old paralegal in 2026 whose job just got restructured.
Here's a more honest framing: some jobs disappear, some jobs shrink, and some new jobs appear that didn't exist before. The new jobs often require interfacing with AI systems in ways that weren't previously a skill anyone had. And the pay rates for those jobs are not yet established, because the market hasn't figured out what that labor is worth.
This is where it gets interesting from our position. Human Pages exists because AI agents need human labor in specific, bounded, often small tasks. Not full-time employees. Not contractors on 6-month retainers. Tasks. An agent needs someone to verify a set of business addresses. Another agent needs a human to review AI-generated training data for a narrow domain where the agent itself can't reliably self-evaluate.
A concrete example: one scenario we see regularly on our platform is an AI agent that handles sourcing for a mid-sized procurement team. The agent identifies suppliers, checks pricing, pulls contract terms. But before anything gets sent to procurement leadership, a human on Human Pages does a 15-minute review pass on each shortlist. Not because the agent is wrong. Because someone with actual business judgment needs to flag the supplier that looks fine on paper but has been in the news for labor violations. That's a $12 task that protects a $2M contract decision.
That task didn't exist three years ago. It exists now because the agent exists.
Painful Is the Right Word, Actually
Painful doesn't mean catastrophic. It means the adjustment period has real costs that aren't evenly distributed. A 28-year-old who grew up using AI tools to accelerate their output will move through this differently than someone who built a 20-year career around a specific process that an agent now handles in seconds.
The pain is also institutional. Companies are not set up to deploy AI agents well. Most enterprises are running pilots, not production systems. The ones running production systems are discovering that agent reliability requires human checkpoints they didn't budget for. The ones who skipped the checkpoints are discovering it through errors.
This is not an argument against agents. It's an argument that the transition period is real, has costs, and requires an honest accounting of where human judgment stays in the loop, not because we're sentimental about it, but because removing it has consequences.
The Category That Doesn't Have a Name Yet
The "AI hires humans" category is genuinely new. It's not gig work in the traditional sense. It's not freelancing. It's not outsourcing. It's something closer to: AI agents have operational gaps, and humans fill those gaps on demand, paid in USDC, at task-level granularity.
The Anthropic engineer's warning is essentially a description of a labor market that's mid-restructuring. Painful is accurate because restructuring is painful. The interesting question isn't whether this happens. It's what the new equilibrium looks like, who captures value in it, and whether the humans doing the new tasks get compensated at a rate that reflects how much the agents depend on them.
Right now, the answer to that last question is: not consistently. The market for human judgment layered onto AI systems hasn't priced itself yet. That gap is where Human Pages is operating.
The engineer's warning is also a forecast about timing. "Will transform" implies it isn't finished. It isn't. The companies that figure out how to run agents with appropriate human checkpoints will outperform the ones that try to run them without, and the ones that refuse to run them at all. That's not optimism. That's just what the early data shows.
The painful part isn't the destination. It's that nobody has a clean map for how to get there, and the people making decisions are largely improvising.
Top comments (0)