Which jobs humanoid robots can actually do today
Strip away the demo videos and the answer is specific. A humanoid in 2026 does a short, repeatable loop inside a defined space, with objects it has seen before. That rules in about five of the twelve physical jobs in a typical $1-50M company, puts two in a rent-only band, and leaves five for later. Here is each job, where it lands, and why.
The one constraint that explains everything: the work cell
A work cell is a box on the floor. Inside the box: known objects, known positions, a sequence that repeats. A conveyor on one side, a rack on the other, totes that always look like totes. The humanoid learns the loop and runs it for a shift.
Outside the box, the robot's reliability falls off fast. New objects, people walking through, a layout that changes week to week, a decision that depends on context the robot does not have. That is why a humanoid can pick a known SKU all day and cannot plate a dish. It is not that the kitchen is harder in some abstract sense. It is that the kitchen is not a cell.
So the test for every job below is the same: can you draw the box, and is the loop inside it short? If yes, it is a candidate. If no, it waits, and no vendor price changes that.
Where humanoids are actually deployed in 2026
Per the vendor, Agility Digit is the one in paid pilots at scale, in logistics, offered as Robots-as-a-Service at roughly $2,000 to $4,000 a month. Its deployed jobs are exactly what you would expect from the work cell rule: tote handling and line-side moves in warehouses. 1X NEO (per the vendor, $20,000 or $499 a month) is positioned home-first, so its public deployments are household tasks rather than shifts. Figure 03 is targeting $20,000 to $25,000 for pilot customers with a manufacturing focus. Tesla Optimus is on 2027 consumer guidance, so there is nothing for a business to deploy yet. If you want the cost side of those numbers, it is in how much it costs to rent a robot worker.
The twelve jobs, by band
These are the rows on the Robot Org Chart. For each: what a humanoid does in 2026, and the AI employee that has to exist first. A humanoid is a body. The AI employee is the brain, a software role with rules you can read running inside your company.
Pilot now
| Job | What it does in 2026 | What has to exist first |
|---|---|---|
| Receiving | Line-side tote move, QR scan, put-away on known SKUs | Receiving AI employee reading POs and reconciling counts |
| Picking | Single-item picks from a bin with a known pose. Digit does this in paid pilots | WMS integration plus an AI employee that picks the picks |
| Inventory counts | Scheduled aisle walks with a camera and a counter. Monotonous, high-value | Cycle-count AI employee that owns the schedule and the variance report |
| Security walks | Fixed-route patrol with cameras and a radio. Boring, high-coverage | Security AI employee reading the feeds, dispatching the human for anything real |
| Lab sample handling | Repetitive, predictable, in-cell, high-value. The sweet spot for first-wave pilots | Lab-ops AI employee that owns chain-of-custody and the exception report |
What these five share: a box you can draw, a loop you can time, and a high cost of boredom. Humans doing these jobs make errors because the job is dull, and they leave because the job is dull. That is the robot's opening.
Rent, don't buy
| Job | What it does in 2026 | What has to exist first |
|---|---|---|
| Packing | Fixed-SKU boxing on a predictable line. Variety still breaks it | A packing-rules AI employee that handles the exceptions a human handles today |
| Cleaning rounds | Known routes, known surfaces. Cobot vacuums already do the floor; humanoids are overkill until the ladder jobs land | A facilities AI employee owning the SOP and the checklist |
These two are real but fragile. The loop exists, but the exceptions are frequent enough that you want the vendor's support contract and the freedom to send the robot back. Do not buy hardware for a job where "variety still breaks it" is the honest status.
Wait
| Job | What it does in 2026 | What has to exist first |
|---|---|---|
| Maintenance rounds | Visual inspection on a route is close. Actual repair is still a human | A maintenance AI employee that logs the finding and dispatches the human |
| Kitchen prep and plating | Vendor demos exist (Figure, 1X). Real kitchens still have too much variety | A kitchen-ops AI employee that writes the standards tight enough for a machine |
| Front desk cover | A screen does this better in 2026. A humanoid is for the moment the customer is in the room and nobody else is | An AI receptionist already handling the inbound; the humanoid is the physical hand-off |
| Local delivery runs | Humanoid-in-vehicle demos only. Driving plus stairs plus dogs is still humans only | Dispatch AI employee first; the vehicle layer comes before the humanoid |
| Event setup and teardown | Too much variety and too many people in the space. Watch, don't buy | An event-ops AI employee writing the run-of-show tight enough that any helper can execute it |
Every job in this band fails the work cell test. The space changes, the objects change, or the public is standing in it. Note what the "what has to exist first" column says for all five, though: there is an AI employee you can run today for each. The wait band is not "do nothing." It is "build the brain, skip the body."
How to read your own business against this list
Count your pilot-band jobs. A warehouse-heavy company might have four. A restaurant group might have zero. A lab has at least one. That count tells you whether 2026 is a hardware year for you or a software year.
Then, for the job you would pilot first, ask whether the AI employee in the right-hand column exists yet. If it does not, that is your first project, and it costs a fraction of a rental. If it does, you are ready to talk to a vendor. The full pre-flight is in what to do before buying a robot for your business.
Five jobs pilot. Two rent. Five wait. All twelve have an AI employee you can run this quarter.
A $1-50M business owner on an Optimus mastermind call summed up the right posture: pick the dullest job in the building with the clearest box around it, and start there. That is the pilot band in one sentence.
FAQ
What jobs can humanoid robots do right now?
Narrow, repetitive physical tasks inside a defined work cell: tote moves and put-away on known SKUs, single-item picks from a bin with a known pose, scheduled inventory count walks, fixed-route security walks, and sample transfers in a lab cell. Those five are the pilot band on the Robot Org Chart.
What is a work cell and why does it matter for robots?
A work cell is a defined space with known objects, known positions and a repeatable sequence. Humanoids in 2026 work reliably inside one and poorly outside one. If a job cannot be drawn as a box on the floor with a short loop inside it, it is not a 2026 robot job.
Can a humanoid robot work in a kitchen or restaurant?
Vendor demos exist from Figure and 1X, but real kitchens have too much variety for a 2026 humanoid. Kitchen prep and plating, and front desk cover, are both in the wait band. The useful 2026 move is a kitchen-ops AI employee that writes the standards tight enough for a machine later.
Which humanoid robots are actually deployed in businesses?
Per the vendor, Agility Digit is the one in paid logistics pilots at scale, offered as Robots-as-a-Service at roughly $2,000 to $4,000 a month. 1X NEO ($20,000 or $499 a month) is positioned home-first. Figure 03 is targeting $20,000 to $25,000 for pilot customers. Tesla Optimus is on 2027 consumer guidance.