The Last Mile Goes Driverless

A parcel reaches a person through more than a stretch of road. Someone loads it, resolves the address, finds a place to stop and completes the handoff. A vehicle can become driverless while some of that work remains. This is the distinction the automation story needs if its labor consequences are to be discussed seriously. The important unit is the service that a customer receives and the work required to provide it, rather than a photograph of an empty driver’s seat.

Amazon’s current Vulcan explainer provides a bounded example of warehouse automation. It describes picking and stowing items in fulfillment-center storage, using touch and force feedback, and calling a human partner when an item cannot be handled. Its introduction dates to 2025; this page was checked on October 10, 2026. These are company descriptions of a specific task and assistance arrangement. They do not establish that a delivery station runs without people or that a vehicle can complete the handoff at an unfamiliar door.

Street operation has a stated permission

NHTSA’s August 4, 2025 closing record for investigation AQ23001 states that Zoox obtained a demonstration exemption after an inquiry into vehicle certification. It describes conditions including removal of claims that the vehicles conform to all applicable federal motor-vehicle safety standards, and says the covered vehicles then operating on public roads did so under an exemption. Closing this certification inquiry is not a comparative finding about crash risk, a guarantee of all future operations or evidence that the same vehicle delivers parcels.

A later NHTSA announcement, dated July 30, 2026, describes a temporary commercial exemption allowing up to 2,500 Zoox robotaxis annually for two years with oversight conditions. This is an authorization ceiling in an agency announcement, not a count of vehicles deployed or completed paid trips. Demonstration authority and commercial authority have different scopes. Neither record establishes Tesla’s launch timing, which this essay does not repeat as a verified fact.

The distinction matters beyond accurate terminology. Passenger transport can end when a rider leaves a vehicle; a parcel service must still define a successful handoff. My proposed deployment review would identify operating territory, weather and road limits, remote support, accessibility arrangements and what happens when the system cannot complete the trip. Those are questions to ask of a specific service. This review does not supply a universal operating envelope or infer one from another operator’s permission.

Count tasks before predicting jobs

BLS’s Occupational Outlook Handbook, modified August 27, 2026, estimates 1,068,500 light-truck-driver jobs in 2025 under occupation 53-3033. It separately counts driver/sales workers. That identifies an occupational baseline, not jobs already lost or technically removable. The inherited figure of 1.7 million and its historical ranking are withdrawn because this review did not establish a matching definition and date. A deployment announcement cannot convert an occupation count into a displacement estimate.

The BLS duties description includes loading and unloading, communicating with customers, reporting problems and handling delivery documentation. Driving is one task in that account. The relevant labor question is how a particular employer reorganizes the complete assignment when a vehicle can perform part of it. This cannot be answered merely by giving a system a new name. It requires a comparison of work, staffing and service requirements before and after the change.

An empty driver’s seat is evidence about a vehicle; a changed livelihood needs an employment record.

Consider a constructed eight-hour paid route serving 120 stops. Assume three hours are spent driving, four on package handling and handoffs, and one on paperwork. Now let the vehicle perform the driving, while the same employee still accompanies it to reach each stop and complete the remaining tasks. Assume the route and paid shift remain eight hours. Steering time falls from three hours to zero; paid staffing time in this example does not fall. These invented inputs are a counterexample to a simple inference, not a claim about a real delivery operator.

Savings could require a different arrangement: someone else completes handoffs, the customer collects packages, staff cover several vehicles, or the service changes its promises. Each arrangement moves work or responsibility somewhere. A complete comparison would account for remote assistance, maintenance and failed deliveries as well as the onboard role. It should also show whether workers gain safer tasks, lose hours, receive different pay or leave employment. Task automation creates possibilities; the staffing decision and its consequences require their own evidence.

A transition is an arrangement, not a slogan

The Department of Labor’s WARN overview describes advance-notice requirements for covered employers and qualifying plant closings or mass layoffs, with exceptions, and notes that some states have their own laws. It does not establish that every automation-related change triggers notice. Nor does it support calling WARN the only relevant federal protection or declaring all jurisdictions devoid of transition obligations. This essay has not reviewed every contract, worker classification or state requirement, so it withdraws those universal claims.

My proposed standard for an employer begins before deployment. Explain which tasks may change, what training is available, whether training time is paid and how workers can ask for review of a reassignment. Then publish enough outcome information to compare planned transitions with completed placements, hours and pay. This is an editorial recommendation rather than an account of present duties everywhere. Counting training invitations alone would not show that displaced workers found usable alternatives.

There is a complication: a change can benefit a worker in one dimension and worsen another. Less physically demanding work can arrive with less predictable scheduling, or a technical role can require training that some workers cannot attend. Those are possibilities to examine, not outcomes measured here. A useful transition record needs the affected workers’ account alongside the operator’s efficiency measures. An invented driver watching her former route cannot stand in for that account.

The civic test follows the complete service

Cities and customers have related questions. Can a person with limited mobility receive the service? Who handles a blocked curb or an unsuccessful handoff? What evidence supports a reliability or safety claim, and what comparison was used? A sporting record by a humanoid or a general decline in component prices would not answer these operational questions. This essay does not verify those inherited claims or use them to forecast widespread adoption.

The last mile goes driverless one defined task and operating arrangement at a time. Its public significance comes from how those arrangements change service, street use and livelihoods. The strongest case would connect vehicle capability to completed deliveries and an accountable employment transition. Until that record exists for a particular deployment, the right judgment is conditional. There is room for useful automation, and a continuing need to ask who completes the work and who carries the consequences when its organization changes.