# Two Percent a Year: The Slow Curve Behind the Robot Headlines

> Each year robots learn about 2% of the physical work they couldn't do before. Why that slow curve, plus 3% yearly price drops, sets the real robot timeline.

- Canonical URL: https://webofmike.com/robots-two-percent-a-year/
- Author: Mike Moore (https://webofmike.com/about/)
- Published: 2026-10-24
- Last modified: 2026-10-24
- Tags: Robotics, Metrics, Thoughts
- Cite as: Mike Moore, "Two Percent a Year: The Slow Curve Behind the Robot Headlines", Web of Mike (webofmike.com), 2026-10-24. https://webofmike.com/robots-two-percent-a-year/


The scariest robot charts go up and to the right. The most useful one in Anthropic's report goes up slowly.

Here it is in one line: each year, robots become able to do about **2%** of the physical work they previously couldn't.

That's from [What work can robots do?](https://www.anthropic.com/research/what-work-can-robots-do), and it's the number I'd use for any planning conversation about robots and jobs.

## How they got it

This is a clever bit of method, so it's worth a minute.

Anthropic took job task descriptions from 1977. Then they rated those same tasks two ways: what the robots of 1977 could do, and what today's robots can do.

- In 1977, robots could **not** do **62%** of physical tasks.
- Today's robots can do all but **24%** of those same tasks.

So about half a century took the "can't do" share from 62% down to 24%. If you assume a steady rate, that's roughly 2% of the remaining tasks per year.

I checked it. 24 divided by 62 is about 0.39. Take the 49th root and you get about 0.981. One minus that is about 1.9%. I did this math on the back of a paper plate. It holds up.

## Two slow curves multiplied together

Capability is one curve. Cost is the other.

From the [first post in this series](/capable-isnt-cheap-robots/): robot prices have fallen about 3% a year since the 1990s, and at that rate it takes about 40 years for robots to be cost-competitive on 10% of work.

Anthropic puts both curves together:

| Scenario | Half of today's physical work cost-competitive by |
|---|---|
| Base case: 3% yearly cost decline, historical capability growth | 2085 |
| Fast case: costs fall up to 4x faster, capabilities grow 2x faster | 2050 |

And even in the fast case, automating 90% of today's physical work takes 53 years.

That's not a prediction that nothing happens. Real jobs change a lot along the way, and the report is clear that the most exposed jobs show up first. But it's a long way from "robots take most jobs in five years."

## Why slow curves still matter

I've spent a lot of my career looking at metrics on dashboards. The lesson I keep relearning is that the slow metrics are the ones that get you.

A spike pages someone. A 2% drift doesn't. It looks like nothing on any given day. Then one quarter you notice the baseline moved and everything you planned around is off.

Robot capability is a drift metric. Two percent a year is nothing in a news cycle. Across a career, it's a different world. Someone starting work at 22 today will see robots learn a large share of the tasks they can't do now before that person retires. That's the right frame for a training plan or a career plan. "Robots are coming next year" isn't.

## The case for faster

Anthropic is careful not to treat history as a ceiling. It says AI-powered robots "could leapfrog our scale and do work they cannot today, for example by learning to climb ladders or use their arms and grippers more deftly." Humanoids or new manufacturing methods could push costs down faster than 3% a year.

The industry version of that argument is the flywheel. At the SCSP AI+ Robotics summit, Agility Robotics CTO Pras Velagapudi [said](https://www.understandingrobots.org/p/washington-is-worried-about-chinas) "the single most important factor in advancing robots is getting robots out there to have practice doing it." The idea is that more deployments mean a better product and lower costs, which means more deployments.

Kai Williams, who covered the summit, also quoted a roboticist who isn't sold on the data part of that story. Chris Paxton wrote that "most robotics deployments are fundamentally not scalable," and that the data they produce "is almost certainly less valuable than you think it is." Williams' own take was that the broader flywheel idea, where early deployments help get a market going, "seems much more solid."

I land in the same place. Deployment speeds things up. How much is the open question, and nobody has a clean number for it yet.

## What I'd do with this

If you're planning anything around robots, a factory, a warehouse, a training program, a career, use two numbers:

1. **About 2% a year** for new capability.
2. **About 3% a year** for falling cost.

Then watch for evidence that either one is speeding up. A real change in either curve will show up in the deployments before it shows up in the headlines.