On July 7, 2026, Ars Technica published a feature in which leading researchers and founders from Boston Dynamics, Agility Robotics, Physical Intelligence and Intuitive explain how modern AI is moving robots from simple navigation toward independent work in warehouses and factories. The conclusion is cautious — robots already work in logistics and manufacturing, but general-purpose machines for homes remain decades away.
Key takeaways
- Agility Robotics has logged more than 65,000 hours of Digit robot operation across nine customer facilities, including GXO, Schaeffler, Toyota Motor Manufacturing Canada and Mercado Libre
- On June 24, 2026 Agility announced a merger with Churchill Capital Corp XI to go public at a $2.5 billion valuation under the ticker AGLT
- The company secured over $300 million in orders for Digit v5, billed as the first "cooperatively safe" humanoid
- Boston Dynamics and Hyundai target capacity of 30,000 Atlas humanoids per year by 2028
- In surgery, most commercial robots remain at the first of five autonomy levels, with decisions left to humans
From the Stanford Cart to the warehouse
The scale of the shift is clearest against the starting point. In 1979 the experimental Stanford Cart needed five hours to travel 20 meters across a room full of obstacles. The first bipedal robot able to walk without losing its balance appeared only in 1996. Autonomy then meant getting from point A to point B.
Matt Malchano, vice president of software at Boston Dynamics, told Ars Technica that autonomy has always been a "moving target". Today it means not just navigation but sequences of actions and an understanding of tasks. The switch was unlocked by two waves of AI — reinforcement learning in the 2010s and large foundation models in the 2020s.
Sergey Levine of UC Berkeley, cofounder of Physical Intelligence, warns against overreach. Current methods can produce a robot very good at one task under narrow conditions, or roughly capable at many, but not at 99.99 percent reliability. "We really want something that's extremely good at all things, and that's still at the frontier of research," he said.
Who is already working
Three companies illustrate the range of approaches. Agility Robotics was the first to win a long-term commercial contract for humanoids, deploying its Digit robots at a GXO logistics warehouse in Atlanta from 2024, where Digit moves totes from order-picking areas to conveyors. It expanded to Toyota Motor Manufacturing Canada lines, a Schaeffler plant in South Carolina and a Mercado Libre facility in San Antonio, and has been tested in Amazon warehouses.
Boston Dynamics represents a different path. Its four-legged Spot has for several years conducted autonomous inspections in places hazardous to humans, such as National Grid converter stations in Massachusetts. Its wheeled Stretch robots handle packages in DHL warehouses. The Atlas humanoid is being trained by the company's South Korean parent, Hyundai Motor Group, aiming to work at an EV factory in Ellabell, Georgia, by 2028.
Physical Intelligence offers a third view. Levine does not believe in a single all-purpose humanoid, but in a general AI model powering many robots suited to their jobs — an arm hanging from a ceiling in a small apartment, or a heavy machine on a farm.
| Company | Approach | Where it works today |
|---|---|---|
| Agility Robotics | A single Digit humanoid | Warehouses and factories — GXO, Toyota, Schaeffler, Mercado Libre |
| Boston Dynamics | A fleet of specialized robots — Spot, Stretch, Atlas | Inspections, warehouses, eventually a car plant |
| Physical Intelligence | One AI model powering many different robots | From an apartment arm to a farm machine (concept) |
Safety as the blocker
The physical risk is not abstract. On January 25, 1979, 25-year-old Robert Williams was crushed and killed by a one-ton robot's arm inside a Ford factory in Flat Rock, Michigan. Jonathan Hurst, cofounder of Agility, points to safety as the main reason so few robots have been deployed. "That's the blocker for everybody," he said. Today Digit works in separated "work cells", isolated from people.
That is set to change. Within a year Agility plans the commercial launch of Digit v5, described as the first "cooperatively safe" humanoid, able to detect a person and sit on the ground before anyone can touch it. Agility and Boston Dynamics also work within an ISO group on ISO 25785-1, a standard for industrial mobile robots that, after committee sign-off, would go to a vote by 89 member nations.
Surgical robots are the counterpoint. Bhushan Patel of Intuitive, maker of the Da Vinci 5 system, stresses that they do not operate fully on their own. Most commercial systems sit at the first of five autonomy levels, some at the second. Inside the human body, tolerance for error is far lower than in a warehouse.
Why it matters
The Ars Technica feature matters because it separates marketing from reality in a segment absorbing billions of dollars. Instead of promising a household helper, it shows narrow but genuine deployments in logistics and manufacturing, and the economic test every robot must pass to justify its purchase and upkeep. The tone of the founders is notable — Hurst says plainly that companies promising safe home robots "are either lying or wrong". The analysis also surfaces a social tension — the Hyundai labor union approved a potential strike on June 25 over job protections tied to the planned Atlas deployment. Agility's public listing and the data gathered from every deployment suggest the advantage will go to whoever converts real operating hours into safer, more general machines fastest.
What's next
- Agility plans the commercial launch of Digit v5 within the next 12 months as the first "cooperatively safe" humanoid
- Boston Dynamics and Hyundai target trained Atlas robots at the Georgia factory by 2028, at 30,000 units per year
- The draft ISO 25785-1 standard awaits a vote by 89 member nations after committee approval
- Agility's merger with Churchill Capital Corp XI is expected to close in 2026, subject to shareholder and SEC approval
