Physical AI and the skilled trades: what's real in 2026

What counts as physical AI in the trades?#

Physical AI in the trades means systems that sense and act on the physical world: robots, drones, smart tools and machines guided by models trained on real-world data. It is different from the software AI many contractors already use, such as AI receptionists, dispatch optimizers and estimating assistants, which work on calls, schedules and documents rather than on pipe and wire.

The distinction matters for planning. Software AI can change how an office runs once it is set up; physical AI changes a job site only when a machine can handle the space, the materials, the safety rules and the person paying for the work. That bar sits far higher in a homeowner's crawlspace than on an empty concrete slab.

Here now, emerging and distant: a task-by-task view#

Sorting trade tasks into here now, emerging and distant gives a clearer picture than headlines about humanoid robots. The status column reflects the general state of the market in 2026 rather than any single product, and adoption is uneven even where a tool is available.

Why the residential service call stays out of reach#

The residential service call stays out of reach because it stacks nearly every hard problem in robotics into one visit. A no-heat call might mean crawling into an attic, reading an old furnace's wiring by flashlight, judging a cracked heat exchanger and explaining options to a worried homeowner, all in a space no machine has seen before.

Every home is a new environment, with pets, children, clutter and stairs. The materials are flexible and inconsistent: old copper, PEX, corroded fittings, stiff wire in a crowded panel. Much of the skill is judgment by feel and sound, and a mistake can mean water damage, a gas leak or an electrical fire.

Licensing and code add another layer. Inspectors, insurers and licensing boards expect a licensed person to answer for the work, and that framework has not been rewritten for machines.

What robotics teams are missing, and where job records come in#

Robotics teams are missing real examples of skilled manual work in messy settings, which is why trade footage and job records have become relevant to them. Models that control robots learn from demonstrations, and tidy lab demonstrations do not show how a technician recovers when a fitting will not seat.

Developers are already seeking real-world capture at scale. In September 2025, robotics company Figure announced a partnership with property owner Brookfield to capture human video across Brookfield's residential, office and logistics environments as training data for its Helix model. Trades work in occupied buildings is a harder, less documented setting than most of that footage.

• 1First-person video of complete tasks, with both hands visible, including mistakes and recovery.
• 2Job photos before, during and after the work, linked to the work order.
• 3Work order sequences showing which tasks happen together and in what order.
• 4Technician notes that explain the diagnosis behind each action.
• 5Outcome records such as callbacks and warranty claims, which show whether the work held up.

Planning moves for trades owners in 2026#

The planning moves that make sense for trades owners in 2026 concern hiring, training and records rather than waiting for robots. Physical AI will not ease the technician shortage in residential service work any time soon, while software tools and prefabrication can already help.

Linkage is the thread through every move. A clip tied to a work order, equipment model and outcome can be checked and labeled later; a folder of unlabeled videos on an old phone usually cannot.

Illustrative: a plumbing and mechanical contractor looks at its prefab shop#

Illustrative: a fictional plumbing and mechanical contractor builds pipe racks and fixture assemblies in its own prefab shop, and its service technicians already attach job photos to work orders in ServiceTitan. The owner reads that robotics developers want video of skilled manual work and asks what the company already holds.

The owner sorts the material by consent burden. Shop footage involves only employees in the company's own building, so it needs a technician consent process and a check for drawings that belong to clients. Residential photos need homeowner notice, removal of faces and house numbers, and exclusion of private rooms such as bedrooms.

The company starts with shop footage and linked fabrication records, and adds consent language at booking before any residential capture. The result is a narrow, documented package from the shop and a capture policy for future service work, instead of a rushed upload of old phone videos.

How SourceX approaches physical-world data from the trades#

SourceX helps trades companies license footage and job records to AI developers through the SourceX five-step transaction: Supply, Rights, Preparation, Approval and Delivery. SourceX does not build robots. The Rights step covers technician consent, homeowner notices and client-owned drawings; Preparation removes faces, addresses and documents; and the company approves every release.

Video archives are large, so they stay in the company's own storage or ship on encrypted drives rather than being hosted by SourceX. For footage, the privacy record in the SourceX Evidence Packet lists the consent forms, notices and blurring steps behind each clip, alongside provenance, licensing rights, permitted use and release authorization.