---
title: "Can we predict the jobs robots will do?"
description: "Anthropic's robot exposure index estimates robots can technically perform about three-quarters of US physical job tasks but are cost-competitive for only 0.3%, which bears on how soon physical jobs may change."
kind: "wire"
section: "Work"
author: "Josh Alley"
author_url: "https://hastalavista.ai/authors/alley"
published: "2026-10-04T02:27:13.550Z"
updated: "2026-10-04T02:30:23.500Z"
canonical: "https://hastalavista.ai/work/can-we-predict-the-jobs-robots-will-do"
source_url: "https://www.anthropic.com/research/what-work-can-robots-do"
source_name: "Anthropic"
archive_url: "https://web.archive.org/web/20261003012539/https://www.anthropic.com/research/what-work-can-robots-do"
---

# Can we predict the jobs robots will do?

*Anthropic's robot exposure index estimates robots can technically perform about three-quarters of US physical job tasks but are cost-competitive for only 0.3%, which bears on how soon physical jobs may change.*

By [Josh Alley](https://hastalavista.ai/authors/alley) · October 3, 2026

Source: [Can we predict the jobs robots will do?](https://www.anthropic.com/research/what-work-can-robots-do) — Anthropic

## Human take

*These are the author’s thoughts, made without AI and typed directly into our publishing software without assistance.*

Robots + AI can't do human physical labor jobs better — *today*. But plot the trend with a longer lens and you'll see where it's headed.

- Anthropic is self-disclosing that, right now, only 0.3% of physical labor jobs can be done cheaper by a robot on a 1:1 basis
- Remember, software engineering was once impossible without humans, and now it is better than humans at it
- Today, across a large number of physical jobs, cameras + AI are recording these tasks as training data with the intent to empower robots to do them
- Taxis used to need a human, now you can tap a button and have a driverless chauffeur deployed
- Human labor is typically one of the most expensive bills a company has to pay
- If a company can pay less for labor or eliminate it entirely, they make more profit
- This naturally creates the demand for solving the problem of physical labor needing a human

![LLM and robot exposure by occupation group](https://oli14uru1ac8n1u4.public.blob.vercel-storage.com/media/27ca9274-a1d9-4d8c-b249-e1dfd109a0c0/1600.webp)
*Credit: Anthropic.*

This is the capitalist's map of robot + AI opportunity. Get the training data, focus on solving the human labor problem in these areas, and realize the greatest profit potential.

> **Thought exercise**
>
> With enough training data, compute power, and time, AI has been able to replace software engineering. Maybe not 100%, but it's getting there fast. These digitally-native "jobs of the mind" are all at massive risk as a result.
>
> Remember, we laughed at how bad the AI videos were just a few years ago. How wrong GPT was, yet how confident it was in its assertions. How fast things have changed.
>
> Physical labor jobs, let's say a plumber for example, is thus not really at risk today. A robot can't come diagnose the problem with your toilet and reach for the plunger or snake. If someone made a robot today that tried, it would be hilariously bad.
>
> But the problem with your toilet may be quickly diagnosed *by* AI today. I've done that with my AC unit and prevented a few hundred bucks in house calls as a result. Diagnosing things in person is one part of the equation, performing the repairs is another.
>
> Which is why companies are out there strapping headcams to folks like technicians, hairdressers, delivery drivers, seamstresses, and many other tradesfolk, skilled and unskilled alike.
>
> The video is being used to train AI to "know" how to perform all of these tasks, including their nuance, and across real-world situations that require human decision making and finesse.
>
> Why? Well, why did AI study code, graphic design, writing language, and so on?