Will robots take warehouse picker jobs?

Threat timeline: Already happening at GXO, Toyota Canada, Schaeffler and Mercado Libre — nine facilities and 65,000 logged hours is no longer a pilot programme, it is a small deployed workforce. Over 1-2 years the pattern replicates at the next tier of 3PL operators, the same crowd already running Locus AMRs. Over 2-4 years the gate is Digit v5: fenceless operation is what removes the safety cage and the designated zone, and that is the difference between a robot that needs its own bay and one that can be dropped into an existing aisle. December 2026 is the date to watch. The interesting structural question has not changed: whether Digit-class humanoids displace the AMR category by handling the tote-on-tote-off step AMRs cannot, and whether Toyota proving the automotive case opens the same door at every other OEM. What this re-check adds is a caution that belongs on the record. The $300 million order book that underpins a $2.5 billion valuation rests on one contract, from one customer nobody has named, for a robot that has not yet been delivered to anyone, with payment tied to milestones. Against that sits roughly $100 million a year of cash burn and no disclosed revenue. None of this means Digit fails — it is the best-evidenced humanoid deployment in this catalog by some distance. It means the timeline above is contingent on a company that is still spending far more than it earns successfully completing a public listing. Anyone using this entry to judge their own exposure should weigh the deployment evidence and the balance sheet together, not just the first one.

Threat timeline: Unchanged; nothing in this re-check justifies a revision. Already deployed at scale. Over 1-3 years, the next wave of large retailers and 3PLs that have not built yet. Over 3-5 years, the long tail of mid-sized operators, as Pio lowers the floor for the SMB segment. The one observation worth adding comes from the pricing structure rather than the technology. AutoStore's pay-per-pick model compresses the upfront grid infrastructure payment to 20-40% of total cost and converts the rest into a subscription that scales with order volume. That has nothing to do with robot performance and everything to do with how the system gets compared against people: once the billing unit changes from a one-time multi-million-dollar capital expenditure into a price per pick, it lands on the same page as a picker's hourly wage. That is this category's real diffusion mechanism, and it is also why a 26-week go-live against competitors' 52-78 weeks determines who gets displaced first far more than any hardware specification does.

Threat timeline: 1-2 years for automotive assembly line tasks where the BMW playbook can be replicated. 2-4 years for broader light-industrial deployment. The economics work at scale when one Figure runs 10-hour shifts replacing a wage that compounds — automotive is the natural first market because the integrator cost is amortized across high-volume production lines.

Threat timeline: The 'already happening' verdict stands, but this re-check adds a directional change worth writing into the timeline. The original assessment holds: a 2-3x productivity multiplier does not usually show up as layoffs in the short term. It shows up as the same headcount absorbing more volume during peak. The real long-run effect is the rate of new picker hiring falling, particularly for the contract 3PL roles that are the entry point into warehouse work. Over 1-3 years the compression is in picker-hours, especially seasonal hiring. Over 3-5 years, as Array scales, pick labour in the largest facilities moves from augmented-by-robots to mostly absent. The new variable is the May 2026 Nexera acquisition and its NeuraGrasp technology, which deserves its own flag because it moves the boundary of what gets displaced. Assistive picking AMRs like Origin have only ever eliminated walking and searching; the grasp itself stayed with the human, and that is the last technical reason warehouses still need people. Reliably pulling one specified item out of a tote full of mixed shapes, materials and weights remains unsolved. Once grasping is folded into the Array platform, that reason starts to erode. It will not happen in a year or two — general-purpose grasping has been persistently underestimated — but the acquisition is an unambiguous statement of direction, and readers should treat 'can a robot pull an arbitrary item out of a mixed tote at human success rates' as the single most informative indicator for the next three to five years.

Threat timeline: 3-5 years. RB-Y1 is a development platform, not a finished worker — today's buyers are labs, not factories. The variable is Samsung: once the consolidated manufacturing and distribution muscle engages, light-assembly and machine-tending applications built on this platform can scale faster than any independent startup's path. The wheeled base skips bipedal locomotion's hardest problems, which may pull deployment earlier, not later.

Threat timeline: 1-3 years for greenfield warehouse picking and light assembly, where the environment can be designed around the robot. 3-5 years for retrofit deployments in existing facilities. Generalized household work is further out and is not where the labor displacement happens at scale anyway.