Flywheel AI Is Putting Excavator Operators in an Office Instead of a Ditch

The Macro: Nobody Wants to Operate an Excavator Anymore

The construction industry has a labor problem that numbers alone do not capture. The Associated Builders and Contractors estimated 500,000 unfilled construction positions in the United States in 2024. The average age of a heavy equipment operator is climbing every year. Young workers are not lining up to sit in a machine cab in 110-degree heat for twelve hours.

This is not just a hiring problem. It is a safety problem. OSHA data shows that heavy equipment incidents account for a significant portion of construction fatalities annually. Excavators specifically are involved in struck-by incidents, cave-ins, and rollovers. Many of these accidents happen because an operator has a blind spot, makes a fatigue-related error, or is working in conditions that limit visibility.

The industry knows this. The solutions so far have been incremental. Caterpillar has added GPS-guided grading to some equipment lines. Komatsu has its intelligent Machine Control system for dozers. Built Robotics has pursued autonomous operation for specific tasks. Teleo has gone after remote operation for loading and hauling. SafeAI has focused on autonomous mining equipment.

But here is the gap. Most of these approaches require new equipment or work only with specific machine models. The existing fleet of excavators in the field is massive, and contractors are not going to scrap functioning machines to buy new ones with built-in autonomy. The economics do not work. A new excavator costs $100,000 to $500,000 depending on the size. Contractors finance that over years. They are not replacing a machine that runs fine just to get remote operation capabilities.

What the industry needs is a retrofit solution that works on existing equipment, does not require extended downtime for installation, and does not interfere with manual operation when an operator is in the cab. And ideally, one that gets smarter over time.

The Micro: Retrofit First, Autonomy Later

Jash Mota and Mahimana Bhatt founded Flywheel AI in San Francisco out of Y Combinator’s Summer 2025 batch. Mota is the CEO. The team is two people. Their pitch is direct: Waymo for excavators.

That analogy is more precise than it sounds. Waymo did not start by building fully autonomous vehicles from scratch. It started with instrumented vehicles and human safety drivers, collected millions of miles of driving data, and used that data to train increasingly autonomous systems. Flywheel is running the same playbook for construction equipment. Start with remote operation. Collect data on every dig, dump, trench, and grade. Train models on that data. Gradually introduce autonomous policies for specific tasks.

The hardware is machine-agnostic. Flywheel says their retrofit system installs in hours without disrupting manual operations. That means a contractor can install the system on a Monday morning and have the machine back on the job site Monday afternoon, operable by either a person in the cab or a remote operator at a desk. The installer does not need to rip out the existing control system or modify the hydraulics in ways that void the manufacturer warranty.

The remote operation platform runs with low latency and includes supervisor override capabilities. An operator sitting at a workstation can run the machine as if they were in the cab. A supervisor can watch multiple machines and take over if something goes wrong. The safety story is compelling: the operator is not in the trench, not near the swing radius, not exposed to the physical hazards that cause the injuries and fatalities that the industry has failed to reduce.

But the real value is the data flywheel. Every hour of remote operation generates training data. Every trench, every dump, every grading pass. Flywheel is using that data to train machine learning models that can automate specific tasks. The vision is a gradual handoff: remote operators handle the complex parts, autonomous policies handle the repetitive parts, and over time the balance shifts.

The business model aligns incentives correctly. Contractors do not want to buy software. They want to operate equipment more efficiently with fewer people. Flywheel is selling that outcome. One operator managing multiple machines. More hours of operation per day. Fewer safety incidents. Those are metrics that construction companies already track and care about.

The Verdict

I think the teleoperation-first approach is the correct way to build autonomous construction equipment, and I think Flywheel’s retrofit model gives them a distribution advantage that purpose-built solutions lack. There are hundreds of thousands of excavators already in the field. If you can make them remotely operable without replacing them, you skip the biggest adoption barrier in the industry.

The competition is real but fragmented. Built Robotics is further along on autonomy but requires specific equipment configurations. Teleo is in the remote operation space but focused on different equipment types. Caterpillar and Komatsu have OEM advantages but move slowly and price their technology into new equipment only. None of these players have a machine-agnostic retrofit that installs in hours.

Thirty days, I want to see how many machines are actively running with Flywheel’s retrofit in the field and what the operator-to-machine ratio looks like. Sixty days, whether the autonomous policies are handling any real tasks in production or are still in the training phase. Ninety days, the question is whether contractors are seeing enough productivity gains to justify the cost and whether the data flywheel is generating models that actually reduce the need for human input. If one operator can reliably manage three machines, the ROI sells itself. If the ratio stays close to one-to-one, this is a safety play but not a productivity play, and the market for that is smaller.