Digit heads toward AGCO: a factory pilot planned for 2027
A humanoid in a farm-equipment factory will first sort parts and position containers. AGCO and Agility Robotics announced a Digit pilot agreement on 20 October 2026 for Hesston, Kansas, with an expected start in 2027. Two days later, Agility described its plan to move Digit 5 onboard computing to NVIDIA IGX Thor. Together, the announcements cover two different parts of commercialisation: a customer with a job and a developing hardware platform. They do not establish an operating Hesston fleet.
Agility Robotics official press kit, Digit 4. Product photograph with a tote; not the planned AGCO pilot and not confirmation of its robot generation.
What AGCO agreed
The release names parts sorting and stacking, plus positioning totes for storage. It identifies a factory and an expected starting year. It does not disclose robot count, contract price, pilot duration or productivity results. The release body says Digit; we do not automatically treat a photograph or next-generation announcement as confirmation of the variant for the entire agreement.
For European readers, the application boundary matters. This concerns manufacturing agricultural machinery, rather than a humanoid working in a field. Moving a tote inside a plant is different from working in an orchard, livestock building or unpaved outdoor environment. The shared issue is physical-work automation; the environment and success criteria differ.
Why legs for a container-handling task?
Our interpretation is that value must be tested where transport meets manipulation in human-designed space. Before a humanoid earns an advantage, it should demonstrate that it reduces layout changes or covers several operations within one deployment. A human shape is not an economic justification.
A useful counterfactual is the same operation performed by an AMR with a prepared station, a conveyor or a simpler parts-feeding improvement. If the simpler option meets throughput requirements, the humanoid must justify its flexibility. If changing the production layout would be expensive, mobile manipulation may become more attractive. These are questions to test, not AGCO-reported outcomes.
IGX Thor: a Digit 5 plan, not completed deployment certification
On 22 October, Agility said it plans to consolidate Digit 5 onboard computing around NVIDIA IGX Thor. The collaboration includes NVIDIA Halos elements and planned participation in a systems inspection lab. Agility links that work to future external certification. A technology collaboration should not be represented as completed certification of the whole robot.
Agility continues to expect early access to Digit 5 in the first half of 2027, followed by increasing availability over six to nine months. That is the supplier timetable, not a delivery guarantee for every European customer. Compute architecture may support perception and safety-system development, but a module name does not disclose task reliability.
What the pilot needs to report
| Area | Evidence needed after launch |
|---|---|
| Throughput | Correctly handled parts or totes per hour across the complete process |
| Human assistance | Intervention count, cause and employee time |
| Quality | Sorting errors, damage, retries and rejected operations |
| Availability | Working, charging, waiting and recovery time |
| Economics | Integration and support cost per correctly completed unit of work |
Start a site pilot with one parts flow and one accountable owner. Define success before the robot arrives. Separate the speed of an individual movement from station throughput. Agree when an operator should stop the experiment and revert to the existing process. This keeps the result comparable even when initial commissioning requires substantial changes.
What this changes in the market picture
A named customer and a defined task are a stronger signal than a general future-factory video. They still precede demonstrated results for this application. A new compute plan describes a product path to scale, not completed shifts. Combining the two announcements does not create proof that the entire system is ready.
RoboMorrow follows these distinctions in our Digit 5 coverage and humanoid total-cost analysis. Our humanoid availability comparison also separates buying hardware from reserving access. The next useful Hesston evidence will be real trial results, rather than another retelling of the signed agreement.
Sources and methodology
Redakcja RoboMorrow. Verified 26 October 2026. Analysis of documentation and source material; no physical hardware test. Manufacturer statements are not independent measurements.