Ask a chatbot whether our work society is efficient and it will give you a confident answer. Ask it again, slightly differently, and you may get the opposite one. That is the first thing to understand: there is no “it” with a fixed view of work. A model reflects the argument you bring it.
But the questions are worth asking, and the evidence behind each one is real. Here is what it actually shows: our work society is not mainly an efficiency machine. We keep it running for reasons that have little to do with output. And the animal kingdom, supposedly nature’s factory floor, is full of workers doing nothing at all.
Does AI think our work society is efficient?
A model will tell you AI boosts productivity. It has been trained on a mountain of vendor claims saying exactly that. The measured record is messier.
In July 2025, the research nonprofit METR ran a randomized trial with 16 experienced open-source developers working on 246 real tasks in their own repositories. When the developers were allowed to use AI, they finished 19 percent slower. That number matters less than what happened next: they had forecast a 24 percent speedup, and even after living through the slowdown they still believed AI had made them 20 percent faster.
You can see the same gap at scale. A European Central Bank blog post in August 2026 reported that AI use at work has doubled over two years, with workers reporting large time savings, but that perceived productivity gains vary widely from person to person and task to task. Saving hours is not the same as producing more value, because the hours still have to go somewhere useful.
That is the crux, and a 2026 paper from researchers at NYU and the University of Illinois names it. Their Counterfactual Resilience Framework asks a question companies rarely ask: what happens if the AI goes away? A budget cut, an outage, a vendor dispute. The answer exposes whether a workflow is genuinely efficient or merely AI-shaped. They warn that AI can distort performance expectations when its usefulness differs sharply between roles, and that leaning on it can erode the skills people need to work without it.
There is a telling detail in how firms talk about this. New data released on September 29, 2026 found that executives chasing AI “efficiency” through headcount cuts were nearly half as likely to invest in the training needed to actually realize the gains they wanted. They are optimizing the spreadsheet, not the work.
This is an old pattern with a name. In 1987 the economist Robert Solow quipped that you can see the computer age everywhere but in the productivity statistics, and it took about a decade of reorganizing work, not just buying machines, before IT showed up in the numbers. AI looks like it is on the same curve. The tools land first; the gains arrive only when the work is redesigned around them.
So is our work society efficient? Not in the way the word implies. Efficiency would mean output per hour rising and hours falling to match. What we mostly get is faster drafts and unchanged schedules. That is a productivity story with the productivity left out.
Why do we have work at all?
If work existed only to produce things, we would have largely solved it. Industrial productivity multiplied many times over the last century, and working hours did not fall in proportion. Something else is going on, and it is not mysterious. Work, as we practice it, does at least four jobs at once:
- Distribution: it is the main mechanism by which money reaches people.
- Identity and status: “what do you do” is the first question strangers ask for a reason.
- Coordination: shared schedules and roles let large groups act together.
- Structure: it gives the week a shape, and many people would feel its absence.
Only the first of those is about output, and even that one is a design choice, not a law. Anthropologists have long pointed out that “work” as a separate thing you have, sell, and retire from is a modern invention, not a human universal. Marshall Sahlins called hunter-gatherers “the original affluent society” precisely because they met their needs with far fewer hours than a modern wage worker. James Suzman made the same case in his 2020 book Work: for most of human history, labor was woven into life rather than carved out of it as a commodity.
There is a blunt version of this in David Graeber’s 2018 book Bullshit Jobs, built on surveys in which large numbers of workers said their own job did not need to exist. Whatever you make of that claim, the pattern is telling: if work were purely about output, that many people could not describe their own output as pointless.
And the arrangement itself is younger than it feels. The five-day, 40-hour week is not ancient. Henry Ford popularized it in 1926, and the United States wrote it into federal law only in 1938. What looks like human nature is often just a recent settlement that nobody has revisited.
What does the animal kingdom say?
The usual story is that nature is a relentless workplace and everyone pulls their weight. Nature disagrees.
In 2015, biologists Daniel Charbonneau and Anna Dornhaus tracked 250 individually paint-marked Temnothorax ants across five colonies for three weeks. About 25 percent of the workers were inactive the entire time, and the same ants were idle whenever the camera looked. The researchers concluded that some workers “effectively specialize in inactivity,” the way others specialize in foraging.
Ants are not the exception. Earlier work found that a minority of bees does the bulk of the labor in a hive. And humans themselves, before agriculture and wages, worked far fewer hours than we do now. Hunter-gatherers had slack; we invented the 40-hour week.
What the animal kingdom actually shows is that division of labor, reciprocity, and idle reserves are ancient and normal. What is uniquely ours is the layer we built on top: wages, markets, careers, and the assumption that a person’s worth is their output. The animals work. They just do not have jobs.
There is a lesson in that. Colonies keep reserves because conditions change, and a colony that ran every ant at capacity would break the first time a forager died or a trail flooded. Markets and wages optimize for throughput in steady conditions. Nature hedges. We mostly do not.
The Short Version
- AI has no fixed view of work. Ask twice and you get two answers, shaped by how you ask.
- Our work society is not mainly an efficiency machine: METR found developers using AI were 19 percent slower while believing they were 20 percent faster.
- We keep work for functions beyond output: income distribution, identity, coordination, and structure.
- Nature is not a factory: about a quarter of the ants in one study did nothing at all, and most bees leave the work to a few.
- The 40-hour week is a human invention, not a law of nature or of markets.
- Only one of work’s four jobs is about output, and automating the output does not settle the rest.
FAQ
Is AI really making work more productive?
Sometimes, and unevenly. Time savings are real and widely reported, but value gains depend on what organizations do with the freed hours. Rigorous studies still find cases where AI slows experienced people down.
Is the animal kingdom really full of idlers?
Yes, more than the story allows. About a quarter of one ant species studied were consistently inactive, and a minority of bees does most of the work. Slack appears to be part of how colonies stay resilient.
So what is work for?
Mostly distribution, meaning, and coordination. Output is the part we talk about, but it is not the whole machine.
Why does this matter for AI?
Because automating output is the easy half. If work mainly moves money and status around, then taking over the tasks does not solve the payroll. The hard question is the distribution, not the labor.
The companion question: what does AI think about us?
If a model has no fixed view of work, it has no fixed view of humans either. We laid out the full evidence, from sycophancy to the lab fight over whether there is a mind there at all.
About The Author
