Why Uptime Matters More Than Viral Videos
Buyers do not buy capability. They buy availability — the same useful work, shift after shift, with few interruptions and a clear recovery path.
A viral clip wins the morning. A reliable shift wins the contract.
Buyers do not pay for capability. They pay for availability — the robot being ready and able to do useful work when it is needed, today and tomorrow and the day after, with few enough stops that no-one has to babysit it.
Can it work the whole shift, the next shift, and the shift after that?
What uptime really means
Uptime is plain: the share of scheduled work time the robot is actually able to work. It is not just running. It is ready, allowed, and able. A robot that is jammed, charging, waiting for a person, or stuck on a software error is not contributing.
Manufacturing teams already have a name for this thinking. Overall equipment effectiveness, or OEE, combines availability, performance, and quality into a single picture of whether the machine is earning its place.
If you cannot answer “how many hours did it actually work?”, you do not have an uptime story.
What downtime actually costs
- Lost output
Every minute stopped is product not made, totes not moved, shipments not packed.
- Human babysitting
An operator standing by a stuck robot is a person not doing their normal job.
- Cascade effects
One robot stopping a line can stop a whole sequence of work behind it.
- Trust damage
After enough stops, people stop relying on the robot, even when it is running.
- Support load
Every callout, ticket, and overnight visit costs money — and patience.
What the named humanoid examples show
The most interesting humanoid numbers today are not the highlight reels — they are the runtime totals. Agility Robotics reports Digit has moved 100,000+ totes at GXO's Flowery Branch facility. Figure reports 1,250+ runtime hours and 90,000+ parts loaded at BMW. BMW says Figure 02 ran ten-hour weekday shifts supporting production of more than 30,000 X3 vehicles.
These numbers are still company-reported, and they are not full OEE. They are the right shape of evidence: continuous time, real output, named site. That is what uptime evidence looks like when it is being built honestly.
Why a video is not a shift
- 01A video shows
What the robot can do in a chosen moment, often after many attempts.
- 02A shift shows
What the robot will do in every moment over hours, with no choice of which moments.
- 03A video hides
Failures, resets, charging, support calls, and the gaps between actions.
- 04A shift exposes
Every one of those, because the clock keeps running whether the robot does or not.
Capability is a peak. Uptime is an average.
What people often misunderstand
- Mistake 01
“If the demo is good, the deployment will be good.”
A peak performance and a sustained average are different problems.
- Mistake 02
“A robot that needs help sometimes is broken.”
Every real deployment has interventions. The question is the rate and the trend.
- Mistake 03
“Uptime is the vendor's problem.”
Customers run the floor. They share responsibility for charging, layout, and operator training.
- Mistake 04
“Faster is better.”
A fast robot that stops often produces less than a steady one that does not.
A buyer's checklist for uptime
Across one full week of real operation, how many hours did this robot actually do useful work?
- Scheduled work hours.
- Available hours (ready and able to work).
- Interventions per hour, and trend over weeks.
- Mean time to recover from a stop.
- Throughput per shift — and how it compares to the alternative.
- Uptime is the share of scheduled time the robot is actually able to work.
- Manufacturers measure this as availability — part of OEE.
- Downtime costs more than lost output: it costs trust and operator time.
- Runtime hours and named output are the right shape of humanoid evidence.
- Capability is a peak; uptime is an average — buyers care about the average.
- Uptime
- Time the robot is ready and able to do its work, as a share of scheduled time.
- Downtime
- Time the robot is scheduled to work but cannot — stopped, charging, waiting, or broken.
- OEE
- Overall equipment effectiveness — availability × performance × quality.
- Mean time to recover
- Average time from a robot stop to the robot working again.
- Intervention
- A person stepping in to help, unblock, reset, or restart the robot.
- Throughput
- How much useful work the system produces per unit of time.
Sources and evidence notes
What this essay leans on
| Claim | Evidence | Strength | Note |
|---|---|---|---|
| Manufacturing judges equipment by availability, performance, and quality. | IBM explainer on Overall Equipment Effectiveness. | Medium | Standard productivity framing. |
| Digit has accumulated material runtime at GXO. | Agility Robotics public reporting — 100,000+ totes at GXO's Flowery Branch facility. | Strong | Company-reported. |
| Figure 02 ran sustained shifts at BMW Spartanburg. | BMW Group and Figure reporting — ten-hour weekday shifts, 30,000+ X3 vehicles, 1,250+ runtime hours, 90,000+ parts loaded. | Strong | Named site, named task, partial measurement. |