Every wave of transformation gets called science fiction. Then, a few years later, it is just how work gets done.

This week, Rivian CEO RJ Scaringe said that by around 2030 we will see things that have historically been science fiction: robots doing broad-spectrum tasks in manufacturing plants, and cars that drive you to the airport and back (Inc.). His robotics startup, Mind Robotics, is building foundation models for factory work like sorting parts, fastening components and connecting wiring.

I believe him. Not because the demos are impressive, but because I have watched this exact moment happen three times before. The technology was never the slow part.

The first time: a lathe in the late 1990s

In the late 1990s, I demonstrated CAD/CAM software connected directly to plotters and CNC machines. The design went in on one end. On the other end, a lathe cut the part.

The human in that loop did one thing: approve the output. No one stood at the machine turning handwheels. The operator became the approver.

The reaction in the room was always the same. People looked at it as science fiction. A machine tool running on approval alone did not match their picture of a shop floor. Today, that is simply how modern machining works.

The second time: seeing the incident before it happens

About a decade ago, I demonstrated anomaly detection at the business process level, not just at the infrastructure level. Layered metric streams from monitoring were correlated in real time. When something drifted, the system identified the root cause as it happened, whether it sat in the environment or in the process itself.

The human was no longer the person reconstructing an outage after the fact. The human acted on a root cause the system had already found, often before the business felt any impact. It did not stay a proof of concept; it went into production.

Again, the room called it science fiction. Operations had always been reactive. Avoiding the incident instead of resolving it did not fit anyone’s picture of how IT ran.

The third time: operations without a service desk or NOC

Then I took a bigger idea to enterprises: eliminate first, automate what cannot be eliminated, and give humans assist for the rest. I demonstrated what that looks like in practice: service desk-less and NOC-less operations.

The reaction was identical to the lathe and to anomaly detection. Leaders who had spent careers scaling ticket queues, shift rosters and escalation matrices saw it as science fiction. Yet the platform behind it ran multi-tenant in production, with customers onboarded, eliminating 60–80% of work and taking out more than 35% of cost.

The technology had arrived. The mental model had not.

The pattern: the operator becomes the approver

Put the waves side by side, and the same shift repeats each time.

  • CAD/CAM, late 1990s. The machinist stopped running the lathe. The design drove the machine directly, and the human approved the output.
  • Business process anomaly detection, a decade ago. Engineers stopped reconstructing outages after the fact. The system found the root cause in real time, and humans acted on it before impact.
  • Eliminate-driven operations, now. Agents stop working ticket queues and NOC screens. Work is eliminated or automated at the source, and humans assist on true exceptions.
  • Physical AI, around 2030. Workers stop doing repetitive plant tasks. Robots handle broad-spectrum work, and humans supervise and approve.

In every wave, the shock is not the machine. It is the moment people see where they now sit in the loop. “Science fiction” is often the polite word for “I don’t see my role in this yet.”

That is why I treat human assist not as a fallback, but as a designed position. The approver role has to be engineered with as much care as the automation.

Why operations gets there first

The Inc. piece names the hard part of physical AI. Robots need real-world data on how objects move and how environments change, and that data is hard and expensive to collect.

Enterprise operations does not have that problem. The data already exists, and in volume: years of tickets, logs, monitoring streams, change records and incident histories. Most enterprises are sitting on a decade of operational memory they have never turned into intelligence.

That is what historical intelligence does. Before designing anything, it reads that history to answer one question for each type of work: eliminate it, hand it to an agent, or give a human assist. The training set for the science fiction moment is already in the building.

So IT and OT operations can reach their 2030 moment well before 2030. The shop floor will follow, and it will need the same operating model.

The real lag is the operating model

CAD/CAM did not spread at the speed of the software. It spread at the speed of shop floors redesigning roles, skills and quality checks around the machine. The same is true now.

You can demonstrate NOC-less operations in a week. Moving an organization to run that way takes deliberate change management:

  • New roles for the people who used to work the queue.
  • New metrics that reward eliminated work instead of closed tickets.
  • New governance for what agents may do on their own.

This is why I hold that AI has to be change-management-led. Actionable intelligence should drive the operating model change, then the transformation, then the transition. Skip that step, and the science fiction stays in the demo room.

Physical AI will hit the same wall. The robots Scaringe describes will be ready before most plants have redesigned the work around them.

The question to ask before 2030

The science fiction moment always arrives early. It arrived on a lathe in the 1990s. It arrived in business process anomaly detection a decade ago. It has arrived in eliminate-driven operations now. It will arrive on factory floors by the end of this decade.

The leaders who benefit will not be the ones who wait for the technology to mature. They will be the ones who redesign the work while it does. Three questions are worth asking now:

  • What does your operational history already say should be eliminated, not automated?
  • Where will your people sit in the loop, and have you designed that role on purpose?
  • Is your operating model changing as fast as your technology?

If the answer to the last one is no, the gap is not in your AI. It is in how you run the business around it.

This article is part of the Enterprise Intelligence series on building intelligence into every enterprise process, application and operation to drive peak business performance. Earlier: why the model doesn’t know why you want a rain monitor and how we built the platform.