Software AI and embodied AI hit different jobs
Concern about AI and employment has mostly centered on software: drafting, coding, customer support, analysis — tasks performed at a keyboard. A humanoid robot changes a different part of the labor market. It has to move through a physical space, lift and place real objects, and operate reliably alongside people and machinery. That makes its economic effect land somewhere else: warehouses, factory floors, and distribution centers, not open-plan offices.
It matters, then, to be specific about where deployment is actually happening today, rather than where it is marketed as eventually happening. The honest picture: real, paid, multi-year humanoid deployment currently clusters in logistics and light manufacturing — not in general household work, which remains substantially teleoperation-dependent rather than autonomous.
What is actually deployed, at what scale
Agility Robotics’ Digit is the clearest example of humanoid robots doing paid, sustained physical work today. GXO Logistics announced what it called the industry’s first multi-year humanoid-robot agreement with Agility in mid-2024, describing a Robots-as-a-Service model in which GXO pays for completed work rather than owning the hardware (GXO). By November 2025, Agility reported that Digit had moved more than 100,000 totes at a single GXO facility in Flowery Branch, Georgia (Agility Robotics). By mid-2026, in filings tied to its public listing, Agility disclosed a broader figure: Digit had logged more than 65,000 hours of operation across nine customer facilities, including Schaeffler, GXO, Toyota Motor Manufacturing Canada, and Mercado Libre.
That is a meaningful, verifiable deployment record — hours logged, named customers, a public filing behind the numbers — but it is also narrow. These are tote-moving, machine-tending, and sortation tasks: structured, repetitive, high-volume physical work in facilities designed for automation. It is not evidence that humanoid robots are ready for the variability of a home, a retail floor, or unstructured outdoor work, and Agility does not claim otherwise.
What forecasters expect, and how fast
Wall Street’s own numbers on this sector have moved quickly and mostly upward. Goldman Sachs, in a September 2026 research note, raised its humanoid-robot shipment forecast to 890,000 units by 2030 and 6.5 million units by 2035, sizing the resulting market at roughly $138 billion, with logistics, warehousing, and automotive cited as the primary drivers (247wallst.com, reporting Goldman Sachs research). Morgan Stanley’s China-specific forecast, updated in June 2026, put that country’s humanoid-robot market at roughly $2 billion in 2026, growing to $15 billion by 2030 at a 106% compound annual growth rate, with unit shipments reaching about 446,000 by 2030 (CNBC, reporting Morgan Stanley research).
These are bank research projections, not delivered results, and should be read with the same caution applied to any forward-looking market forecast: analysts have revised these numbers upward repeatedly over a short period, which says as much about how fast sentiment is moving as it does about underlying deployment reality. Still, the direction both banks describe — sharp, sector-specific growth concentrated in logistics, warehousing, and manufacturing — lines up with where Agility’s actual contracts sit.
A labor-market data point worth taking seriously
The World Economic Forum’s Future of Jobs Report 2025 adds a complicating, and arguably confirming, data point. Surveying employers on skills they expect to need through 2030, the WEF found that demand for “manual dexterity, endurance and precision” is expected to see a net decline of 24% — the first time this recurring survey has shown a net-negative outlook for a physical-skill category. That is a broad macrotrend finding, not an AI-specific or robotics-specific one, and the WEF’s report attributes shifting skill demand to several forces at once, automation among them. But it is consistent with what the Agility deployment record shows on the ground: industrial and warehouse physical labor — the very work most driven by “manual dexterity, endurance and precision” — is the segment where employers already expect the least future demand, and where humanoid robots are already being paid to work today.
That should not be read as a claim that all manual work is being automated at once, or that this trend is driven solely, or even mostly, by humanoid robots specifically as opposed to conveyor automation, machine vision, and other non-humanoid systems already common in these facilities. The WEF figure captures a broader employer sentiment shift; humanoid robots are one visible, well-funded piece of a longer automation trajectory in this segment.
The trades are a different story, for structural reasons
Skilled trades — electricians, plumbers, HVAC technicians — sit in a different part of the physical-labor economy, and the forces bearing on them are largely not the ones described above. That work requires diagnosing unpredictable problems in unstructured, often irregular built environments (an unfamiliar basement, a nonstandard wiring job, a building code variance), a domain where the foundation-model generalization gains described in how AGI-era models change robot capability remain, by the developers’ own account, well short of practical reliability. These trades also face structural labor shortages driven by retirement and years-long licensing pipelines that have nothing to do with AI. See automation-resistant trades for that argument in full; the short version is that warehouse and factory floor work, and skilled trades work, face genuinely different automation exposure over the next decade, and treating “physical labor” as one undifferentiated category obscures that.
What this means for a household weighing the risk
For a family assessing how AI-driven job disruption might reach a household’s income, the distinction matters directly. A warehouse associate, machine tender, or sortation worker in a facility resembling GXO’s is in the segment where paid humanoid deployment already exists, forecasters expect the fastest unit growth, and employers already report declining demand for the underlying physical skill set. A licensed tradesperson doing variable, judgment-heavy physical work in unstructured environments faces a materially different and, on current evidence, slower-moving exposure. Both deserve honest planning; neither deserves the same timeline.
The reasonable, source-grounded position sits between two extremes. It is not true that humanoid robots are about to replace all manual labor — deployment today is narrow, concentrated, and still expensive per unit. It is also not true that this is a distant, speculative concern with no present-tense evidence: named companies are paying for tens of thousands of hours of humanoid robot labor today, in exactly the segment of physical work employers already say they need less of.