Writing — Robotics and Physical Automation

When Software Learns to Lift

November 16, 20265 min readBy Milan Cheeks

Robotics is where AI leaves the screen and starts answering to gravity, safety, and repair, and that change will sort builders from storytellers.

When Software Learns to Lift

For most of my career, software has lived in a forgiving place.

It lived on screens. If it broke, you restarted it. If it made a mistake, you rolled it back. If a release went wrong, you shipped a fix and moved on. The world it touched was patient. Data waited. Nothing fell over.

Robotics ends that patience.

When software learns to lift, carry, cut, pour, and move through rooms where people live and work, it stops answering only to logic. It starts answering to gravity. It answers to friction, to weight, to worn parts, to a wet floor, to a child stepping into a path nobody predicted. That is why I think robotics and physical automation are the most honest test AI will face.

On a screen, intelligence can look finished long before it is useful. On the floor, useful is the only thing that counts.

The Screen Forgives. The Floor Does Not.

A model can give a wrong answer and the cost is a bad paragraph. A machine that acts in the physical world gives a wrong movement and the cost lands on a person, a product, or a place. There is no undo button for a dropped load or a door closed at the wrong moment.

That difference changes what good work means. In software, we prize speed. In robotics, speed without control is just a faster way to cause damage. The best system is not the one that moves first. It is the one that knows where it is, knows what it is holding, knows what it does not know, and stops when stopping is the right call.

Stopping well is a feature. A machine that can pause and hand control back to a person is not weaker than one that pushes ahead. It is more trustworthy, and trust is the real product in any room where people and machines share space.

Intelligence Is Not the Bottleneck

Most conversations about robots start with the brain. Mine start with the hands.

Human hands are remarkable. They sense pressure, adjust grip, and repair small errors without conscious thought. Rebuilding even a fraction of that in a machine is slow, expensive work. Add sensors that drift, motors that wear, and batteries that fade, and the problem stops being a demo problem. It becomes an operations problem.

A robot that cannot be repaired by the people near it is not a product yet. It is a performance.

Repair is where I judge seriousness. Who fixes this machine when it stops at two in the morning? What parts can be swapped with simple tools? If the answer is a shrug, the technology is not ready, no matter how good the launch video looks.

Trust Arrives Slower in Person

People forgive apps quickly because apps stay at a distance. A robot shares air with you. It moves toward you. It enters the logic of the body, and the body has older rules than any product team.

That is why physical automation will advance task by task, not all at once. The first wins will be narrow and easy to measure. Moving boxes along a known route. Checking shelves. Handling work that is dull, heavy, or unsafe to repeat all day. Those jobs sound modest. They are not. A narrow job done reliably, with clear limits, teaches people that a machine can be counted on. Counted on is a higher bar than impressed.

I build in public, and that habit has taught me to respect this order. Do not ask for trust in advance of evidence. In robotics, evidence is not a benchmark score. Evidence is a shift completed, a floor kept safe, and a person nearby who would choose to work beside the machine again tomorrow.

What This Changes for Builders

Building for the physical world forces a different discipline. You design for maintenance from day one. You design the failure state before the success state. You give a person a clear way to take over, and you make that way obvious under stress, not hidden in a menu.

It also changes the economics. Software can be copied at almost no cost. A machine cannot. Every unit carries parts, assembly, shipping, and support. I have felt a small version of that lesson building hardware of my own. The Manifest Console costs $1,499, its first production run is in production, and it ships by early 2027. Even a device that sits still and runs games will teach you respect for supply and testing.

The builders who win this era will not just be good at models. They will be good at service, spare parts, training, and plain spoken communication with the people who use their machines. The moat is not the demo. The moat is the tenth year of showing up.

Why This Frontier Matters to Me

I founded SELF Labs, and I build Manifest Anything, an AI game creation platform powered by the Manifest Engine and its own AI models. Most of my work turns intent into things people can play, learn from, and build on. Robotics is that same idea pushed into the hardest medium there is. Intent becomes action. Action becomes consequence.

My long project is moving humanity toward a Type I civilization, and that road does not run only through better chat. It runs through energy, food, housing, and care. If AI is going to matter at that scale, it has to learn to work where the work actually is, with care, with limits, and with people firmly in charge of what it is for.

The machines are coming into the room. That sentence can sound like a warning or a promise. Which one it becomes depends on the builders, and on whether we learn the floor lessons now, while we still get to choose them.

If this work speaks to your audience, you can book me for keynotes via https://www.milancheeks.com/#contact

← All writing Book Milan →
The Newsletter

The frontier, in your inbox.

One email when a new essay drops — the Kardashev roadmap, AI-native building, and coexistence, written as it happens. No noise, unsubscribe anytime.

You're on the list.