What we said
Progress in physical AI shows up as each deployment being less fragile than the last, not as a better demo. Progress is measured by fewer interventions, faster deployments, steadier uptime and less firefighting once the system is live. If that is right, buyers who have run these systems for a year should be able to tell us which of those curves are bending.
How we tested it
To test our hypothesis, we looked at three measures:
- How interventions and support have changed for buyers over the past 12 months.
- Whether each new deployment is getting easier to implement than the last.
- What operational impact buyers have actually seen where they have deployed.
What we found
Buyers report uneven progress across their operations:
- Fewer interventions: 76% say a person has to step in less often than a year ago, and 79% say the system now handles more of the task on its own.
- Better recovery behavior: 64% say the system is better at getting itself unstuck.
- Faster deployment times: 66% say a new site is faster to get running than it used to be.
- More stable uptime: 74% say equipment availability improved.*
- Less bespoke troubleshooting: 64% say each new site needs less one-off engineering; 58% say their people spend less time fixing robots that are already live.
*note: base for this question is n=61
79%
of robotics buyers say the share of the task the system handles on its own has improved over the past 12 months
All qualified buyers, n=85 · “Improving” combines “Improving clearly” and “Improving slightly”
| Category | Improving (net) |
|---|---|
| The share of the task / workflow the system handles on its own | 79% |
| How often a human has to step in | 76% |
| How long it takes to stand up a new deployment | 66% |
| The amount of custom engineering each new deployment needs | 64% |
| How well the system recovers on its own after it gets stuck | 64% |
| The support / firefighting burden as more units are deployed | 58% |
All qualified buyers, n=85 · “Improving” combines “Improving clearly” and “Improving slightly”
See the full benchmark81%
of robotics buyers say each new deployment is easier to implement than the ones before it
All qualified buyers, n=85 · single-select, so the shares add to 100%
| Category | % of buyers |
|---|---|
| Clearly easier to implement consistently than earlier deployments | 19% |
| Somewhat easier over time | 62% |
| About the same each time | 11% |
| Still requires substantial customization each time | 5% |
| Too early to tell | 4% |
All qualified buyers, n=85 · single-select, so the shares add to 100%
See the full benchmark85%
of robotics buyers already in production report throughput improving over the past 12 months, the strongest of the five operational measures. Safety incidents move least, at 64%.
The 61 buyers already in production — scaled or in limited production. Buyers still piloting were not asked
| Category | Improved (%) |
|---|---|
| Throughput / output | 85% |
| Labor cost | 74% |
| Uptime / equipment availability | 74% |
| Error / defect rate (a decrease is positive) | 72% |
| Workplace safety incidents (a decrease is positive) | 64% |
The 61 buyers already in production — scaled or in limited production. Buyers still piloting were not asked
See the full benchmark