Anthropic: Robots Could Already Do 74% of Physical Tasks
Robots can already do about 74% of the physical tasks people perform at work in the United States. That comes from new Anthropic research on what robots can actually handle today. The surprise is not what they can do. It is what they cost.
Robots are cheaper than human labor for only around 0.3% of jobs right now. Even if robot prices keep falling at the rates they have in the past, Anthropic estimates it could take roughly 40 years for robots to be cost-competitive across 10% of jobs.

What Happened
- Around 74% of physical tasks in the U.S. could already be done by robots.
- Pair robots with large language models and about 80% of total work time could potentially fall within reach of these technologies.
- Cost is the wall. Robots beat human labor on price for only 0.3% of jobs.
- Falling robot costs would still need about 40 years to reach cost competitiveness across 10% of jobs.
- Some work stays hard for them: tasks with high human interaction, complex physical skills, and jobs that need flexible real-world judgment.
Why It Matters
The bigger question is not simply whether robots will replace humans. It is that AI is moving from digital work into physical work. LLMs work with information, robots work with the physical world, and putting the two together gives you AI agents with bodies.
Healthcare shows why that matters. A hospital is not only data. It is also supplies being moved, rooms being cleaned, items being carried from one place to another, and spaces being inspected. That is physical support work, and it adds up across every shift.
So the AI agent era may not stay on our screens for long. The technology is closer than the economics is, and the gap between the two is the thing to watch.
More shares you might like
ToolsHow to Sign Up for the Amazon Influencers ProgramAlex Ruiez · 5h · 13 views
ToolsHow to Make a Ghostly Abandoned Village Video With AIAlex Ruiez · 1d · 41 views
ToolsHow to Make Motion Graphics Videos With ClaudeAlex Ruiez · 3d · 59 views
NewsNVIDIA Open-Sources Auto Ontology, a Semantic Layer for AI AgentsAlex Ruiez · 4d · 81 views
ToolsHow to Write Specs Coding Agents Can Build Without GuessingAlex Ruiez · 5d · 64 views
AIParallel's Decisions API Grounds Decision Models With Live Web SearchAlex Ruiez · 6d · 70 views
Comments
Join the conversation — sign in to comment.
No comments yet — start the conversation!