How AI Operating Systems Will Rewrite Last-Mile Logistics
Last-mile logistics is no longer mainly about who owns the vans and drivers. It is increasingly about who owns the software that decides which asset moves, when, and under which brand. The sector is shifting from carrier-centric delivery to AI-driven operating systems that coordinate mixed human and autonomous fleets in real time – a quiet migration of power from metal to code.
Photo: Jan van der Wolf
For years, innovation focused on assets and incremental software: bigger hubs, denser networks, better route planning inside traditional transport management systems. Those systems were built for a world where one carrier or integrator controlled most of the chain and could plan once or twice a day, then live with the consequences.
That world is fading. Retailers and platforms now combine in-house fleets, gig workers, local delivery firms, parcel carriers, and specialist partners. Autonomous vehicles, robots, and drones are moving from pilots into early commercial use, adding new options and more complexity. The challenge is no longer just finding capacity, but coordinating this mix of assets as conditions change by the minute.
Many leaders remain trapped in the vehicle illusion: believing the decisive question in the last mile is which trucks to buy or which carrier to pick. The more decisive question is which operating system will orchestrate all of those assets – and on whose terms. Static rules cannot cope with traffic, weather, labour shortages, demand swings, regulation, and customer expectations at once, nor can they decide in real time whether an order should go to an in-house driver, a gig courier, a regional provider, or an autonomous unit.
A crowded field of delivery management platforms has emerged to tackle parts of this problem. Tools such as Onfleet, Tookan, Shipday and Cartwheel help restaurants, grocers and local retailers orchestrate their own fleets and third‑party couriers with auto‑dispatch and routing. Fleet and visibility solutions like project44 or Descartes optimise assets and flows within specific networks. These are important building blocks, but they generally optimise in a single fleet or channel and rely on rules‑based automation.
A new class of platforms, AI operating systems for logistics, is emerging to sit above this layer. They bring together order data, fleet information, carrier performance, service levels and customer interactions into one decision layer, then automate dispatch, routing, communication, returns, fraud checks and reconciliation across connected providers. Nash is one of the clearest early examples in the last mile. According to the company and trade coverage, it positions itself not as a delivery company, but as the platform powering logistics for major retailers and platforms, handling millions of deliveries each year across multiple countries and a broad provider network.
The key point is that this is not simply better route optimisation. It is a different architecture. Instead of treating each fleet or carrier as a silo, the software treats all available assets as resources in a single, live trade-off among cost, speed, reliability and service commitments. Plans stop being static. They become living plans that adjust as reality changes. The last mile starts to look less like a line of trucks and more like an internet of moving things, held together by a common protocol: the operating system.
A similar pattern is visible elsewhere in the supply chain. Platforms such as Altana or Manhattan’s AI‑enabled suites apply computational intelligence to multi-tier visibility, compliance, warehousing and network design. Marketplaces and white-label delivery providers like DoorDash Drive and Uber Direct aggregate capacity and offer APIs, but remain providers inside the network, not neutral brains above it. Together, these developments suggest AI is permeating the stack, with OS‑like layers emerging at different nodes – Nash and peers in the last mile, others in planning and visibility.
As autonomous delivery scales, the last‑mile operating layer matters even more. Autonomous vehicles and robots do not remove coordination problems; they multiply them. Someone, or something, must decide when a robot should handle a job, when a human should take over, how curbside hand-offs should work, and what happens when weather, regulation or technical faults disrupt an autonomous asset. The same software that balances cost and speed can also be designed to account for resilience and carbon, routing around fragile corridors, favouring lower-emission options, and making those trade‑offs visible to boards and regulators.
That is why the operating system is becoming the last mile’s real chokepoint. The platform that sits between merchants and fragmented delivery networks will increasingly influence which provider gets which volume, on what terms, and with what visibility into customer data. It will shape the balance between cost, resilience, service and emissions. For carriers, this brings opportunity and risk. For retailers and platforms, the strategic question is no longer just which carrier is cheapest, but who will orchestrate the network of humans and machines that defines the last mile and whether that software layer reflects their own priorities.