Back to blogs

    Full Autonomy in Freight Is a Setting and Not a Product

    By TransportOne Editorial Team•October 10, 2026•7 min read
    Featured image for an article on AI agent autonomy levels in freight back-office work

    Someone has asked you to approve an autonomous agent for a freight workflow, and the approval form has one box: yes or no. The demo ran end to end without a human touching anything. That's impressive, and it's also the wrong thing to evaluate when you're buying AI automation in logistics.

    AI agent autonomy levels describe how much an agent may do on its own in a given workflow, from showing you information up to acting and reporting afterwards. The useful question is which of your workflows an agent should run unattended, inside what bounds, and what happens the first time one of those bounds is hit. You set the level per workflow, the way you'd set a dial, and you can turn it back down.

    What Are the Different Levels of AI Autonomy?

    Everyone in freight knows the automation levels; the trucking press has spent a decade explaining them. SAE J3016, currently in its April 2021 revision, defines six, running from no automation up to full. It's also the model most discussions of AI autonomy levels borrow from. Most summaries stop at that list, and in stopping there they drop the part that matters.

    The level belongs to the active feature

    The standard's own abstract says the levels apply to the driving automation feature or features engaged in a given instance of on-road operation. So the level describes whatever is switched on at that moment. One vehicle can show two levels in the same hour depending on which feature is active. SAE also asks that its levels carry the prefix, as "SAE Level 2". Treat the L0 to L4 shorthand below as structure borrowed from the standard, and don't cite it as SAE's own.

    One lane went driverless

    The highest-autonomy freight deployment in the United States makes the same point physically. Aurora began commercial driverless runs between Dallas and Houston on May 1, 2025, with Hirschbach Motor Lines and Uber Freight as launch customers. TechCrunch reported that one truck had covered 1,200 miles without a driver at that stage, while more than 30 supervised autonomous trucks hauled over 100 commercial loads a week. Aurora's autonomy arrived one lane at a time, with supervised trucks running alongside.

    Levels of Automation in Freight Operations

    Out of the cab, freight automation tiers describe who acts and who checks. The examples are where a procurement conversation actually gets decided.

    TierWho actsExample
    L0 ManualYour analyst, start to finishInvoice and contract shown side by side. A person reads both.
    L1 AssistedAgent surfaces, person decidesA week of invoices ranked by variance against contract. The analyst picks what to work.
    L2 DraftedAgent prepares, person approvesA dispute written with charge code, contract clause and evidence attached. A human sends it.
    L3 SupervisedAgent acts, person can reverseA missed delivery appointment rebooked, the planner holding an hour to undo it.
    L4 BoundedAgent acts, person reads reportingLine-haul charges matching contract rate exactly, under a stated ceiling, on carriers with no open dispute.

    There's no L5 on that list, and the omission is deliberate. A condition-free tier would mean a workflow with no edge: no dollar ceiling, no carrier exclusions, no lane where the rules differ. Freight operations don't contain one. Naming the edge is most of the work.

    How Do Automation Levels Vary Across Freight Workflows?

    Ask what tier your AI runs at and you won't get an answer, because the question points at the wrong subject. Two things decide the tier for a workflow: how reversible a mistake is, and how cleanly evidence settles the question.

    Same agent, two tiers

    Take automated freight invoice audit. Penny matching a line-haul charge to a contracted rate is arithmetic against documents you already hold, with the rate table and the signed delivery receipt in evidence. That can sit at L4 under a ceiling in the first month. Penny deciding whether four days of detention at a consignee's dock are the carrier's fault or yours is a causality judgment worth a few thousand dollars. It belongs at L2 for a long time.

    The pattern repeats across agents

    With shipment exception tracking, Chase pushing a revised ETA into a customer portal is high-frequency and low-consequence, which puts it in L4 territory from the start. Chase choosing to reroute a load around a closed terminal moves money and reorganises somebody's afternoon, so it sits lower.

    In freight rate negotiation, Miles assembling a carrier shortlist produces a draft. Miles saying a number out loud to a dispatcher produces a commitment, a separate problem covered in how to stop a voice agent committing you on a carrier call.

    Hard rules sit above the tier

    A tier is a default. A hard rule is a veto, evaluated before the tier is consulted at all. That ordering decides whether your guardrails survive a genuine exception. The stack has three layers:

    • Hard rules. Never approve payment without a signed delivery receipt. Never release a dispute against a carrier under legal review. Never commit above the authority the lane owner holds. These apply at every tier, L4 included.
    • Tier default. What the agent may do on this workflow when nothing above has fired.
    • Agent judgment. How it decides inside that permission.

    An L4 workflow that trips a hard rule doesn't get clever about it. It stops and routes to a named person, and the restriction holds at full strength however high the tier is. That's what makes a tier safe to raise.

    When a human must approve, in regulatory terms

    Regulators have begun writing the same shape down. Article 14 of the EU AI Act requires that high-risk systems let the people overseeing them disregard, override or reverse an output. They must also be able to interrupt the system through a stop button or equivalent so it halts in a safe state. Article 14(3) splits delivery between measures the provider builds in and measures the provider specifies for the deployer to operate. Whether or not your use falls in scope, it's a sound checklist for any tier you switch on.

    Watch L3, where approval stops being a decision

    Article 14(4)(b) names a failure worth taking personally: the tendency to rely or over-rely on an output, especially where a system recommends decisions to a person. In a freight back office it shows up as a reviewer who clears an entire L3 queue without opening a single item in it. We've watched this happen with approval queues on invoice work: once the agent's drafts are usually right, the review shrinks to a glance and then to a click.

    If your team approves 99 percent of what the agent produces, you're running L4 already. You're paying latency for a signature that carries no scrutiny, and your audit trail records a review that didn't happen. You have two honest options. Promote the workflow and spend the recovered attention on exceptions, or work out why the check went hollow and repair it. Leaving it as it stands buys nothing.

    Moving a workflow up, and back down

    This is how you implement autonomous AI agents without betting a workflow on a demo: tiers ratchet on evidence, with the evidence agreed in advance.

    1. Run the workflow in shadow for a fixed period. The agent decides nothing; you compare what it would have done against what your team did.
    2. Write the promotion threshold before that run starts, as a number: agreement rate across a stated sample, every disagreement reviewed individually.
    3. Step to L2, then to L3 with a reversal window long enough to catch a real error and short enough to mean something.
    4. Step to L4 inside a bound narrower than feels necessary. Widening later is cheap. An unwind isn't.

    Demotion has to be as easy as promotion, and it needs an owner who can act without convening anyone. If the dial only turns one way, you've built a ceremony.

    This matters for budgets as much as for governance. Gartner forecast in June 2025 that more than 40 percent of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls. Programmes die of missing controls far more often than missing models.

    Your autonomy page, written before the next demo

    Pick your three highest-volume freight workflows. A plant shipping on contracted lanes will often start with line-haul invoice approval, and the freight audit approach for manufacturers follows the same order described here.

    For each workflow, write the hard rules first, as plain sentences built around the word never. Add the starting tier, which for most teams is L2. Finish with the evidence that would justify a move up and the name of the person who can move it back. It takes an afternoon, and you'll get more from that page than from any demo.

    Bring one of those three workflows when you book a walkthrough, and you'll see its tiers running against real invoices and real exceptions.

    Frequently asked questions

    What does human-in-the-loop mean in freight audit?
    It means a person sits in the decision path. At L1 and L2 the person decides or approves before anything goes out, such as a human sending the dispute Penny drafted. At L3 the agent acts and a person can reverse it. If your reviewers approve 99 percent of the queue without opening items, the loop has gone hollow and you're running L4 already.
    How can businesses determine the right level of AI autonomy for a workflow?
    Look at two things: how reversible a mistake is, and how cleanly evidence settles the question. Matching a line-haul charge to a contracted rate is arithmetic against documents you hold, so it can sit at L4 under a ceiling. Deciding who caused four days of detention is a causality judgment, so it belongs at L2 for a long time. Most teams start a new workflow at L2.
    How can businesses implement autonomous AI agents?
    Write the hard rules first, as plain sentences built around the word never. Run the workflow in shadow and set the promotion threshold as a number before that run starts. Step up one tier at a time, ending at L4 inside a bound narrower than feels necessary. Make demotion as easy as promotion, with a named owner who can act alone.
    What are AI agent autonomy levels?
    They describe how much an agent may do on its own in a specific workflow. In a freight back office the tiers run from L0, where your analyst does everything, through assisted, drafted and supervised work, up to L4, where the agent acts inside stated bounds and a person reads the reporting.
    Why is there no fully autonomous level for freight AI agents?
    A condition-free tier would need a workflow with no edge: no dollar ceiling, no carrier exclusions and no lane where the rules differ. Freight operations don't have one, so L4 is always bounded and hard rules still sit above it.

    See it on your freight

    Stop Overpaying On Freight Invoices

    Penny checks every invoice against your contracts and what the shipment actually did, then disputes what doesn’t hold up. You pay 30% of what she saves you, and nothing if she saves nothing.

    See how Penny works

    Found this useful? Share it with your team.

    Share on LinkedIn