What counts as a working robot? Reading the Figure–BMW evidence

When a company says a robot works, the missing words are often “at what?” Sorting one kind of part, operating through a shift and adapting to a different job are different achievements. Keeping them separate makes it easier to recognize real progress without demanding that every prototype already be a finished business.

BMW’s June 2026 account offers a useful example. It says Figure 02 supported production of more than 30,000 BMW X3 vehicles over ten months by inserting sheet-metal parts for welding. That is a contribution to a manufacturing process; it does not mean a humanoid assembled those cars on its own. BMW’s release

Read the task before the headline number

The named workflow helps us understand what was demonstrated. A repeatable operation in a production setting is more specific than a general-purpose capability claim. But a vehicle count still does not disclose the cost of the installation, the frequency of human assistance or the financial return.

There is also a reporting distinction worth preserving. Figure described an eleven-month deployment, while BMW attached a ten-month period to the production contribution. Those statements use different descriptions of the period. We would not silently combine them to calculate a daily productivity figure. Figure’s deployment report

Figure 03’s next workflow introduces sequencing parts in logistics. Figure’s June 30 description adds coordinated movement and cart handling. It is useful evidence of the work being attempted; a new workflow still needs its own operating results. Figure 03 at BMW

Three people, three different tests

Huang’s Cosmos announcement concerns tools for developers. Adcock’s BMW discussion concerns a deployment. Musk’s Davos remarks include an outlook for future availability. Ranking these statements on a single “who is ahead?” scale hides the fact that they describe different parts of the problem. NVIDIA WEF transcript

The economic test comes after the task is clear

Our practical test is to compare the complete cost of doing the job. That includes installation, supervision, downtime and maintenance, as well as the robot itself. A flexible machine could be attractive even if it is slower at one repetitive operation, provided it saves enough changeover work across several jobs. That is a scenario to investigate, not a conclusion about this deployment.

For investors, the most useful next announcement may be a modest-looking disclosure: an expanded customer deployment, a measured reduction in intervention or a repeat order. Each helps connect technical progress to a business outcome. A polished demonstration can be informative too, but it should be described for what it actually shows.

The habit is simple: name the task, identify the evidence and keep the unanswered questions visible. It makes a robot story more useful both when the results are impressive and when the marketing gets ahead of them.