Figure 03 at BMW: Win a Hard Frontier by Starting Narrow

For years, "humanoid robots" meant a sci-fi fantasy: a general-purpose robot that walks into your home and does everything, cooks, cleans, folds laundry. That dream is really hard and still years away. But in 2026, something quietly significant happened: humanoid robots stopped being just lab demos and started doing real, paid work. Figure's robots are now working inside a BMW car factory, doing one specific, unglamorous job: sorting and organizing parts. Not a butler. A warehouse worker. And that, starting boringly narrow instead of impossibly broad, is exactly the smart way onto a hard frontier.

The sci-fi trap: the general robot

Start with why the dream version is a trap, at least for now. A general-purpose home robot has to handle an almost infinite variety of tasks in a wildly unpredictable, chaotic environment, every home is different, every task is fuzzy, things move, and "fold the laundry" actually hides a thousand sub-problems. That is staggeringly hard, basically unsolved, and chasing it head-on means you build something that does a little of everything badly and is not reliable enough for anyone to use or pay for. You would burn years and fortunes and have nothing deployable to show for it. The fantasy of the do-everything robot is seductive, but aiming straight at it is how you fail at the frontier, you take on all the hardest problems at once, in the messiest possible setting, and succeed at none.

The smart move: one narrow, controlled job

Figure and BMW went narrow and controlled. Instead of a robot that does everything in your chaotic home, deploy a robot that does one specific job in a structured, predictable place: a factory.

Think about why a factory is the perfect starting point. The environment is controlled and consistent, same layout, same lighting, same conditions every day, unlike the chaos of a home. The task is well-defined and repetitive, "pick these parts, put them in this trolley, in this order", the kind of narrow, repeatable job a robot can actually learn to do reliably. And the value is immediate and measurable, a factory has clear, paid work that needs doing, so if the robot does it well, there is real money and real demand today, not someday. By picking one narrow, valuable, achievable task in a controlled environment, Figure gets its robots doing real work now, succeeding where it counts instead of failing at everything at once. Narrow is not a compromise. It is the whole strategy.

Why narrow wins: the expansion path

Starting narrow is not just easier, it compounds into the big goal. Watch the ladder. The robot masters one task reliably. That does three things.

One: it earns revenue, real paid work, which funds the next stage of development, instead of burning money with nothing deployed. Two: it generates priceless real-world data, every hour working in a real factory teaches the robot and its makers things no lab ever could, about how the machine actually performs in the messy real world. Three: once it has nailed task one, you add task two, then task three, each a bit harder, gradually building toward a more capable, general robot, but now funded by revenue and trained on real data at every step.

So "start narrow" is not abandoning the big dream of a general robot; it is the realistic path to it. You climb to the ambitious goal one achievable, paid, data-rich rung at a time, instead of trying to leap to the top and falling. The narrow job is rung one of the ladder to the dream.

The catch: we're at rung one

The hype around humanoid robots is wild, so be clear-eyed. What is happening is genuinely important, humanoids doing real, reliable, paid work in a real factory is a real milestone, not a demo. But it is still rung one. These robots do a narrow set of tasks in controlled settings; they are not close to the general do-anything robot of the movies, and the road from "sorts parts in a structured factory" to "handles anything in your chaotic home" is long, hard, and genuinely uncertain, humanoid timelines have a long history of slipping. There are also big open questions about cost, reliability at scale, safety around people, and what this means for human jobs. Hold both: a real, meaningful step onto a hard frontier, and the very beginning, with a long climb ahead.

Risks and open questions

• Long, uncertain road. The leap from narrow factory tasks to general capability is unproven and may take far longer than hype suggests.
• Cost and reliability at scale. Working in one cell is different from being economical and dependable across many.
• Safety around people. Humanoids operating near workers raise real safety questions.
• Jobs impact. Automating physical work raises hard questions about human labor.
• Details are fluid. Verify deployment specifics against the latest reporting before relying on them.

Takeaways for product managers

1. Don't leap at the whole frontier. For an incredibly hard, ambitious goal, attempting the entire thing at once is how you fail at everything. Find one narrow, well-defined slice where you can actually succeed.
2. Pick a valuable, controlled slice first. A narrow task, in a predictable environment, with immediate measurable value, lets you ship something real and reliable now instead of someday.
3. Narrow compounds into the goal. Mastering one task funds the next (revenue), teaches you (real-world data), and lets you expand step by step. It is the ladder to the moonshot, not a detour from it.
4. Revenue and data beat demos. A deployed, paid, data-generating product is worth far more than an impressive lab demo of the full vision.
5. Stay honest about where you are. Real progress at rung one is not the finished dream. Communicate the milestone and the long climb both, that credibility is itself an asset.