Kodiak starts autonomous trucking on I-5 produce lane after California DMV permit

Let me be straight about why I care about this one. A truck running semi-automatically between two California points does not, on its own, change the rate you pay for a container out of Shanghai. What it does is plant a flag on a specific problem that has been quietly eating importer margins on the US side for three years, which is inland trucking. Drivers are short, hours-of-service rules are tight, and any lane that combines a fixed origin, a fixed destination and a predictable schedule is exactly where automation gets traction first. Fresno to Los Angeles is that lane. Agricultural production clusters in the Central Valley, a single dense consumption market at the other end, roughly 225 miles apart, all highway. If you were going to pick a place to prove this works commercially, that is the place.

The reason this matters to an importer is the structure of your US inland cost. On a typical ocean booking to Los Angeles or Long Beach, the inland leg is not a rounding error. If your all-in door-to-door rate is around $9,000 per forty-foot container, the port-to-door trucking portion on a 200-plus mile move is often 12% to 18% of that, so call it $1,100 to $1,600. That number has been sticky in a bad way: it does not fall when ocean rates fall, because it is driven by driver availability, equipment, chassis and terminal turn times, not by vessel capacity. On the agricultural side, where produce moves on tight windows and a late truck means rejected or discounted cargo, the cost of a missed pickup is much higher than the truck rate itself. That is why carriers talk about reliability first and rate second on these lanes.

So what should you actually do with this news today? Not much on the contract. What you should do is start building the two things that will decide whether you can use autonomous capacity when it becomes available: data on your own lanes, and a position in the queue.

Start with the lane audit. Take your top ten US inland lanes by spend and write down, for each one, four numbers: annual spend, average miles, average transit time, and a measure of schedule variability. You can get variability from your own records - how often did the truck arrive outside the pickup window over the last twelve months. The lanes that score high on miles, high on volume, low on variability and high on the pain of a late truck are the ones an autonomous carrier will want, and they are also the ones where you have the most to gain. Do that audit before the end of October. It is an afternoon of work and it is the input to every conversation that follows.

Next, the contract conversation, and I would have it now rather than later because the leverage sits with you this quarter. When you renew contracts for 2027, ask your US truckload and drayage providers one direct question: what is your plan for autonomous capacity, on which lanes, and when. You are not asking for a commitment you can enforce. You are finding out which of your providers is building the capability and which is hoping it never arrives. The ones with a plan will want volume commitments in exchange. That is a real trade, and it is worth making a limited one - say, a portion of one or two lanes, with a rate review in twelve months but no long-term lock. Do not sign a five-year dedicated agreement to get access to a technology that is still running with a safety driver. A rule I would apply: no autonomous-linked commitment beyond 18 months until driverless operations have run for at least six consecutive months on the same lane.

After that, and this is the unglamorous one, look at whether you can make your own cargo easier to automate. Automated trucking economics depend on dwell and handling. If your receiver keeps a truck waiting three hours at the dock because the appointment system is weak, no autonomous or semi-autonomous service will price your lane attractively. Fixing dock scheduling, getting your delivery windows accurate, and making sure your pallets are load-ready has more immediate value than any technology watching, and it costs almost nothing.

Now let me put numbers on the inventory side, because for produce and perishables the driver is not the truck rate, it is spoilage and rejection. Assume you are an importer or a buyer moving fresh produce on this kind of lane, 60 truckloads a season, average cargo value $60,000 per load, and assume a two-percentage-point reduction in spoilage and rejection from more consistent transit times. That is $1,200 per load, $72,000 a season. Modest-sounding, and it is real money. Add the labour saving on the driver side - assume one driver shift per run at a fully loaded cost of roughly $0.85 per mile, which on 225 miles is about $190, and assume automation eventually removes 30% of the driving hours on eligible runs, which is $57 per load. On 60 loads that is $3,420 a season. So the direct labour saving is trivial and the schedule-consistency saving is the actual prize. Anyone selling you autonomous trucking on driver cost alone is selling you the wrong number. Productivity per truck per year is $150,000 to $200,000 in revenue terms, much of it at night. Ask the operator what their utilisation rate actually is. If they cannot give you a credible number, they are early-stage and you should treat their rate card as aspirational.

Where this hurts is if you are not ready, meaning you are still using manual paper-based drayage dispatch, you do not know your own lane variability, and your receiver's dock is chaos. Your competitor with the same cargo and better dock discipline gets the first shot at the cheaper middle mile, and if you are an agricultural exporter in the regions where human drivers are genuinely scarce, you are the one who suffers most.

Go back to Fresno to LA for a moment. It moves about 225 miles, mostly on State Route 99 and Interstate 5, with a safety driver. That safety driver is the tell. The service is not cheaper yet, because the driver is still there and the monitoring infrastructure, remote operations people and insurance loading all sit on top of the same fuel and equipment cost. Honest reading: this is a capability demonstration that gets a carrier to the front of the queue for permits and customer trust, not a cost-reduction event. The numbers that will matter are the ones that come after the driver comes out, and the one Kodiak has put on the calendar is driverless Texas runs by the end of 2026. Watch that date. If it slips past the first quarter of 2027, discount everything else you hear about the timeline by a year.

There is also a compliance dimension people are not ready for. When you are importing into the US and you buy inland transport from a carrier using autonomous equipment, your liability chain changes. Your broker and your insurer will want to know who the motor carrier of record is and whether the equipment was operating under permit. Ask for the permit reference and the operating authority on the lane, and get it before the first load runs, not after. If your cargo is high-value produce, your insurer will also want to know whether the service is running with a safety driver, because that changes the risk profile. Put a clause in your transport agreement that requires notification if the operating mode changes from supervised to driverless. That single clause protects you from discovering a coverage gap at the worst possible moment.

One more thing on labour, because this is where importers get caught out politically and contractually. If your US receiver or your distribution centre has a unionised workforce or an existing driver agreement, the introduction of autonomous capacity on an adjacent lane can trigger consultation obligations that have nothing to do with your contract. Nobody plans for it, and it delays the benefit by quarters. Worth asking your 3PL whether any of the DCs you use have that exposure.

My view after watching transport technologies come and go: the money in autonomous trucking is not in replacing a driver at $0.85 a mile. It is in running assets at higher utilisation and cutting variability on lanes where cargo is expensive and time-sensitive, and in covering peak weeks when no amount of money buys you a truck. Plan for that, and you will get value even if the driverless milestone slips. Plan for a labour-cost story and you will be disappointed, because the labour saving on the California lanes is around $57 per load at best, while the schedule-consistency saving on perishables is in the thousands per load. The trade is about reliability, not wages.

I want to make one more point about how to read the regulatory sequence, because it is the part that tells you what to expect and when. California approved heavy-duty autonomous truck rules on 28 April, the DMV issued the permit on 13 August, and the first commercial run happened on 22 September. That is roughly five months from rule to revenue. Compare that to how long a new port terminal or a new rail intermodal yard takes to go from approval to throughput, and the pace is unusual. It means the constraint on this technology is no longer primarily regulatory. It is commercial: insurance, customer acceptance, and whether the economics work with a safety driver still in the cab. When a constraint shifts from government to insurers, timelines get shorter and less predictable, and the announcements start coming faster than your planning cycle. That is a reason to build the lane data now rather than after the rate cards change.

Also worth understanding what this does not do. It does not help you at the port. The drayage move from Los Angeles or Long Beach to a warehouse 20 miles inland is not the target of this technology, because short moves are dominated by terminal turn times and appointment slot availability rather than driving hours. It does not help you on a lane with irregular volume, because the operator needs utilisation, and a lane that runs three loads one week and thirty the next is hard to schedule for an automated fleet. It does not help you if your cargo requires handling that a driver normally does at the stop, and quite a lot of less-than-truckload freight falls into that bucket. What it does help is the long, boring, repeatable, high-value middle mile, and Fresno to Los Angeles is precisely that. If you do not have any lanes of that shape, this story is interesting but not actionable for you yet, and I would rather say that plainly than sell you a plan you cannot use.

One more thing about the geography. State Route 99 through the Central Valley and the I-5 crossing at the Tejon Pass are two of the busiest agricultural corridors in the country, and they are also corridors with known weather and fire-season disruption. An automated fleet running a fixed lane has to handle the same closures a human driver does, and when a closure forces a five-hour detour on a produce load, the recovery plan matters more than the technology. Ask an operator what their diversion protocol is before you assume a fixed lane means a fixed transit time. I have had loads stuck on a mountain pass closure and, believe me, nobody was asking about autonomy that day.

What I would also flag for anyone with 2027 budget planning in front of them is that this is a technology where the cost curve will probably be driven by utilisation rather than by hardware. That means the early commercial offers will not look cheap on a per-mile basis compared with a conventional carrier, and the correct comparison is not rate against rate. It is total landed cost per unit of product delivered on time. If automated capacity lifts on-time delivery from, say, 92% to 97% on a perishable lane, the avoided penalties, reshipments and markdowns are worth far more than the freight differential. Build that case internally with your own numbers before you talk to a carrier. A procurement team that walks in with a total-cost model and one pilot lane will get a much better arrangement than one that walks in asking for a discount.

And keep the scale in perspective. Kodiak is running one commercial lane with a safety driver, across roughly 225 miles of California highway. The US truckload market moves billions of tons a year and the vast majority of it is still driven by a person in the seat. Nothing about one produce lane in September 2026 resets your freight budget for next year. What it does is tell you which direction the middle mile is going and roughly how fast, and that is enough to justify spending an afternoon on your own lane data and one clause in your next transport contract.

There is a version of this story that matters much less than people think, and I would rather name it. One lane, one carrier, one commodity group, and a safety driver on board. If you move machinery, furniture, apparel or chemicals, nothing in this announcement touches your 2027 landed cost this week, and any consultant telling you otherwise is selling a deck rather than a plan. The reason to read it anyway is that California is the largest single import gateway into the United States, the central and eastern US is where a large share of that cargo ends up, and the middle mile between them is where your cost is least controllable and least transparent. Anything that puts a floor under how bad that leg can get is worth tracking, even at this stage. Track it, prepare your lane data, write the notification clause, and then get back to the dozen things on your desk that actually move margin this quarter. When the no-safety-driver rate card shows up, you will be ready in a week instead of a year, and that is the whole return on reading a story like this today.

If you want a single number to hold in your head as the trigger to act rather than watch, use this one. The moment an operator quotes you a lane with no safety driver, at a rate below the prevailing conventional truckload rate for the same lane, the technology has crossed from pilot to product and you should be in the market within two quarters. Until that happens, your job is preparation, not procurement. Watch the two dates that matter: the driverless Texas commitment for the end of 2026, and whatever insurance product your broker can actually place for unsupervised heavy-duty operations. Insurance availability, not the permit, will be the real gate, because nobody runs a fleet of this size uninsured, and the underwriting market moves on its own clock.

Leo

• By 31 October, complete a US inland lane audit covering your top ten lanes by spend, recording annual spend, miles, transit time and pickup-window performance over the past twelve months.
• In 2027 contract renewals, ask every US drayage and truckload provider for their written autonomous-capacity plan, and limit any autonomous-linked commitment to 18 months until driverless operations have run six consecutive months on the same lane.
• Add a clause requiring notification if a lane changes from supervised to driverless operation, and collect the motor carrier of record, permit reference and operating authority before the first load.
• Budget an assumed 2 percentage point reduction in spoilage and rejection on perishable lanes, worth about $1,200 per load on a $60,000 load, and treat schedule consistency rather than driver cost as the real saving.
• Before 31 December, tighten receiver dock appointment accuracy and load-ready pallet discipline on your two worst-performing lanes, since variability blocks access to cheaper automated middle-mile capacity.
• Track the driverless Texas commitment announced for end-2026, and re-plan if it slips beyond the first quarter of 2027.