Two pieces of research landed this week that appear to speak to the same fear, that machines are coming for our work, and then point in almost opposite directions. Read together, the gap between them is the more honest picture of where automation actually stands, and it is neither the apocalypse nor the all-clear.
The sharper of the two comes from Anthropic, which built what it calls a robot exposure index: an attempt to measure, task by task, how much of the physical economy robots can do right now. The headline finding is striking. Robots, defined as autonomous machines that sense and act, can already perform about 74 percent of physical work tasks in the United States, covering roughly a third of all working hours. Add large language models for the desk work, and only about a fifth of employment is untouched by one technology or the other.
That sounds like a countdown clock, until you reach the second number. Robots are actually cost-competitive with a human worker for just 0.3 percent of tasks. Being technically able to do a job is not the same as being cheaper than the person who currently does it, and on today's price curve the two rarely meet. If robot costs keep falling at their historical rate, Anthropic estimates it would take 40 years for that cost-competitive share to reach even 10 percent. Dexterity remains a wall, the report notes, down to something as fiddly as untangling a bundle of wires, and human preference and regulation slow things further.
McKinsey's Global Institute, publishing on 29 September, arrives at a complementary conclusion from the labour side. Its modelling suggests automation could erase demand for some 36 million jobs in the United States by 2035 while creating around 40 million, a net gain on paper. But buried in that tidy arithmetic is the hard part: roughly 11 million workers in shrinking occupations would have to move into entirely different kinds of work, and 43 percent of today's jobs face some degree of automation. "People should understand the challenge of changing their activities, their processes, and potentially their jobs," said McKinsey fellow Michael Chui. The report is blunt that retraining is hardest exactly where it is needed most, among older workers who lack the qualifications a new field demands.
Put the two together and the popular story, sudden replacement by robots, dissolves into something slower and more grinding. The near-term reality is not mass unemployment but mass reassignment: a workforce that has to keep moving sideways, paired with a technology that becomes capable of a job long before it becomes affordable at it. The robot that could do your warehouse shift already exists. The one that can do it for less than you mostly does not, at least not yet.
That "yet" is where the comfort runs out. Anthropic's 40-year estimate assumes costs fall at the old rate, and the humanoid industry spent this week doing the opposite of that assumption, with Chinese shipments up more than 400 percent in six months precisely because an entire supply chain is driving prices down. Cost curves can bend, and McKinsey's reassuring net-positive math only holds if displaced workers can actually reach the new jobs. Both reports, in the end, point at the same under-discussed task: not stopping the machines, but moving the people.