The thousand-hand facility

The picture in everyone's head

Ask someone to imagine the robot that will run tomorrow's factory and they'll describe a humanoid: two arms, two legs, walking the floor, reading gauges, turning valves, doing what a technician does today. It's a compelling picture, and for some jobs it's the right one. A robot built in our shape can work in spaces built for our shape.

But look at what keeping a facility healthy actually involves. A water works has hundreds of pumps, valves and tanks spread over acres. A dam has piezometers buried along its length and seepage weirs at its toe. A bridge has bearings and girders over a river. A substation has transformers at the far end of a gravel track. The work isn't hard because each task needs a skilled pair of hands. It's hard because it is everywhere at once, all the time, and most of it is waiting.

A single robot, however capable, can only be in one place. The work of a facility is in hundreds.

Rounds are a symptom

The defining ritual of physical operations is the round: a person walking from asset to asset, looking, listening, writing in a logbook. Rounds exist because the facility can't tell anyone how it's doing. Everything we know about a pump, we know because someone walked past it.

A humanoid robot could walk the rounds instead. That's an improvement in labour, but not in kind: the pump is still only observed when something walks past it, and the knowledge still lives in whoever — or whatever — did the walking. The 03:12 trip still waits for the 06:00 round.

The real fix is to stop needing the round. Put a small node on the pump that measures it all the time and can switch it over to its standby. Put another on the valve, the tank, the damper, the breaker. Each one is cheap, simple and always there. Together they make the facility able to see itself.

Distributed robotics

Once nodes can act as well as sense — open a valve, start a pump, switch on cooling, close a door — something interesting happens. The facility becomes a robot. Not a humanoid one, but a distributed one: a body spread across the whole site, with a sense organ and a hand at every point that matters.

A distributed robot has properties a single robot can't match:

• It is everywhere at once. Nothing waits for it to arrive.
• It degrades gracefully. If one node fails, the rest carry on, and the platform knows which one went quiet.
• It scales by addition. Need more coverage? Add more nodes, where you need them.
• It goes where people shouldn't. Nodes can be placed by drone on spans, walls and stacks, and stay there.

A body needs a mind

A thousand nodes acting on their own would be chaos. What makes them one robot is shared intelligence: a single memory of everything that has happened on the site, a model of how its parts relate, forecasts of what is likely to happen next, and a clear rule about who decides.

That is what we've built Vigil to be. It learns what normal looks like for every reading without anyone setting a threshold. It remembers every event, finding and human note, with who recorded it and when. It forecasts every channel and says how likely each limit is to be crossed. It knows which pump feeds which tank. And when the right answer is to move something, it proposes the action, shows exactly what will change, and waits for a person to say yes.

The nodes are the hands. The platform is the mind. The person is the judgement.

Where humanoids fit

None of this is an argument against mobile or humanoid robots. Some jobs need hands in a person's space: replacing a seal, clearing a blockage, carrying a part. A distributed layer makes those robots far more useful. Instead of patrolling in the hope of finding work, they're dispatched to the asset that needs them, with the context of what happened, what was tried and what to check — and what they do is recorded in the same memory as everything else.

The distributed layer finds the work. The mobile robot does the part only it can do. The person decides.

What changes for the people

The point of all this isn't fewer people. It's people doing the work only people can do, more safely. Fewer climbs onto structures to read a gauge. Fewer entries into confined spaces to check a valve. Fewer nights woken to check a reading that turns out to be fine. More time spent on the judgement calls, with the evidence already gathered.

And something quieter but just as valuable: the site stops forgetting. When the engineer who knew that pump 3 sticks below freezing moves on, the note they left is still attached to pump 3, read by the next shift, by the weekly report and by the assistant that helps whoever comes next.

The path

We built the mind first: a platform that senses, remembers, foresees and acts, proven across a demonstration estate of over a hundred simulated sites and full-scale drone surveys of real kinds of structures. In September 2026 the first Vigil Node boards — the hands — arrived from manufacture. Next come pilots with design partners: sensing first, then the first actions people approve.

We think the facility of the next decade won't be run by one remarkable machine. It will be looked after by thousands of small ones, and the people who direct them.

Talk to us about a pilot →