I was staring at a PDF on a screen that was too bright for the time of day, trying to make sense of the world through the bottom of a coffee mug that hadn’t been washed since the last administration. The document in question was the latest “Anthropic Economic Index,” a sprawling collection of charts and data points released just before they unleash their next digital god, Opus 4.5, upon the unsuspecting masses.
They call the new metrics “economic primitives.” It sounds like something you’d find in a textbook written by a caveman who just discovered compound interest, or maybe a band name for a group of guys who play synths in a basement in Brooklyn. But in reality, it’s just a fancy way of measuring how much of our souls we’re currently outsourcing to the machine.
I lit a cigarette, the smoke curling up against the screen like a ghost trying to read the fine print, and dug in. The report captures a snapshot of humanity interacting with Claude in late 2025. And let me tell you, the picture it paints isn’t exactly the Renaissance. It’s more like a cubist painting of a man realizing he’s just trained his replacement.
The suits at Anthropic have decided to measure five dimensions of our digital co-dependency: complexity, skills, autonomy, success, and use case. They want to know if we’re using the AI to cure cancer or just to write passive-aggressive emails to our landlords.
Here’s the kicker: we’re doing both, but mostly we’re just trying to get through the workday without screaming.
First off, let’s look at the map. The report talks about “geographic variation,” which is polite talk for “rich people use AI differently than poor people.”
In the wealthy nations—the places where the GDP is high and the anxiety is higher—people are using Claude for “personal use.” That means we’re lonely. We’re asking the bot for life advice, therapeutic validation, and probably recipes for sourdough starter because we still haven’t gotten over the lockdown era. We treat the AI like a very smart, very patient bartender who doesn’t charge by the drink.
In the developing world, though, the usage is different. It’s all coursework. It’s education. The kids in countries with lower GDPs aren’t asking Claude about the meaning of life; they’re using it to learn code, to pass exams, to claw their way up the ladder that the rest of us are busy trying to saw the rungs off of. They’re hungry. We’re just bored.
The report mentions that usage in the US is “converging.” The gap between the tech-bro states and the flyover states is closing. Everyone is getting equally hooked. They threw around terms like “Gini coefficient,” which measures inequality. Apparently, when it comes to being dependent on a chatbot, America is becoming a socialist paradise. We are all equally useless now.
There’s a section in there about “Augmentation” versus “Automation.” These are the buzzwords that keep middle managers awake at night, sweating through their Egyptian cotton sheets.
“Automation” is when you toss the keys to the robot and sleep in the backseat. “Augmentation” is when you keep your hands on the wheel, but the robot screams at you every time you drift out of the lane.
According to the data, “Augmentation” is back in style. Just over half the conversations are humans iterating with the bot. We aren’t just saying, “Claude, write me a novel.” We’re saying, “Claude, this paragraph sucks, fix it,” and then Claude fixes it, and then we say, “Now it sounds too robotic,” and Claude apologizes profusely, and we feel a tiny, sick little thrill of power.
But look at the API traffic—that’s the stuff businesses use, the code running in the background while you sleep. That is almost entirely automation. The corporations aren’t interested in a chat. They want the work done, they want it done cheap, and they don’t want to pay for your health insurance. The API data shows a massive spike in “Office and Administrative Support.” That’s code for “firing the secretary.”
They frame it as efficiency. I poured a heavy splash of bourbon into the coffee—it was that kind of morning—and thought about the efficiency of a guillotine. It’s very effective at weight loss, technically speaking.
This is where the report turns from dry statistics into a horror story written by an economist. They have this concept they’re tracking called “deskilling.”
For years, the optimists told us that AI would take away the drudgery. They said the robots would wash the dishes and do the data entry so we humans could focus on the “high-level” creative work. We were supposed to be the architects, the visionaries, the dreamers.
Well, the data is in, and the joke is on us.
Anthropic’s “primitives” show that the tasks Claude is best at are the high-education ones. It turns out, writing a complex analysis of market trends requires a lot of book learning, and the model is great at it. But walking across a room to plug in a cable? The model has no legs.
So, what happens to the jobs? The report uses the example of a Travel Agent.
In the old days, being a travel agent meant you were a bit of a wizard. You planned itineraries, you knew which hotels in Paris had the best view of the Eiffel Tower, you crafted an experience. That’s the high-skill, high-education part of the job.
Claude can do that part in three seconds.
What Claude can’t do is the grunt work. It can’t physically issue the ticket on a legacy system from 1995. It can’t act as the payment processor. So, the human travel agent is stripped of the planning, the dreaming, and the consulting. They are left with the “routine ticket purchasing and payment collection.”
They don’t get fired. They just get lobotomized. The job becomes a shell. You aren’t a travel consultant anymore; you’re a fleshy API endpoint for a credit card terminal.
The same goes for technical writers. The AI does the analysis and the writing. The human is left to “draw sketches” or “observe production.” We are being pushed to the margins of our own professions, left to handle the physical scraps that the digital brain can’t digest.
Conversely, they claim “Property Managers” might get an upgrade. The AI does the bookkeeping, leaving the human to handle “contract negotiations and stakeholder management.” I laughed so hard smoke came out of my nose. Have you ever met a property manager? You think they want to spend more time “managing stakeholders”? They want to collect the rent and hide.
There is a dark humor in the “Task Success” metrics. The report admits that Claude is pretty good at most things, but it gets dumber the longer you ask it to work.
If you give it a task that takes five minutes, it’s a genius. If you give it a task that requires five hours of sustained concentration, it starts to hallucinate. It loses the plot. It’s exactly like every drinking buddy I’ve ever had. Fun for the first round, coherent for the second, but by the time you’re asking for the solution to the Middle East peace process at 2:00 AM, it’s just blabbering nonsense.
The chart for “Task Horizons” shows a steep drop-off. The API success rate plummets as the task gets longer. But—and here’s the twist—on the chat interface, the success rate holds up better. Why? Because the humans are there to catch it when it falls. We are the safety net. We are the spell-checkers for the machine god.
They call this “collaboration.” I call it cleaning up vomit.
The report notes that “Augmentation patterns” suggest users are learning to iterate. We are learning to speak machine. We are bending our language, our prompts, and our expectations to fit the limitations of the software. The “Human Education” and “AI Education” metrics are almost perfectly correlated. This means if you write a smart prompt, you get a smart answer. If you write a dumb prompt, the AI dumbs itself down to meet you.
It’s a mirror. A billion-dollar, silicon-based mirror that reflects our own mediocrity right back at us.
And what is all this for? What is the grand prize for turning our travel agents into cashiers and our writers into editors?
The report estimates a 1.0% annual increase in labor productivity.
One percent.
I stared at that number for a long time. I refilled the mug, skipping the coffee this time. One percent productivity growth. That’s the golden idol we’re sacrificing everything for. We are restructuring the entire labor market, deskilling the workforce, and inducing a global existential crisis so that the GDP line can wiggle upward by a single percentage point.
It’s absurd. It’s beautiful in its stupidity.
We’re getting faster at things that maybe shouldn’t be done in the first place. We’re generating more emails, more reports, more code, more content. The “Office and Administrative” tasks are booming in the API data because we’ve built a world where paperwork reproduces asexually. The AI generates the invoice, another AI reads the invoice, and a human sits in the middle, pressing “OK” until their finger goes numb.
The report wraps up with some optimistic drivel about “understanding AI’s impact” and “enabling researchers.” They talk about how this data—these “primitives”—will help us navigate the transition.
But reading between the lines of the bar charts and the scatter plots, the message is clear. The train has left the station, and the conductor is an algorithm that hasn’t slept in three years.
We are entering an age where the definition of “skill” is flip-flopping. The things we spent decades learning—how to write, how to code, how to plan—are becoming cheap commodities. The things we tried to escape—dealing with angry people, fixing physical objects, taking the blame when the system crashes—those are the only things left for us.
The “primitive” isn’t the economic metric. The primitive is us. We are the legacy hardware. We are the wetware running on deprecated drivers, trying to interface with a system that is rapidly outpacing our ability to understand it.
I closed the PDF. The sun was fully up now, mocking my headache. I thought about writing a complex, nuanced analysis of the geopolitical implications of the Anthropic Economic Index.
But then I thought, why bother? Claude can do that.
Instead, I decided to do the one thing the machine still can’t do. I walked down to the corner store, bought a fresh pack of smokes, and wasted thirty minutes watching a pigeon fight a rat over a crust of bread. It was visceral. It was real. It was inefficient as hell.
And it was the most human thing I’d seen all day.
Source: Anthropic Economic Index report: Economic primitives \ Anthropic