Freight/Signal Subscribe
Issue 8 · Voices ·

A year in, freight split in half on AI — augment vs. autonomy

A year into freight's AI wave, brokers split 41% deploying, 48% refusing. Every loud voice sells software; the skeptics just kept their money. The line that sorts it: augment vs. autonomy.

Start with the number, because it’s the honest one. In Bloomberg Intelligence and Truckstop’s 2026 broker survey, 41% of brokers say they’re deploying AI tools and 48% say they’re not. A year into the “AI agent” wave — a year of demos, funding rounds, and “the future of freight is here” — the industry didn’t stampede. It split almost exactly down the middle and stopped. That’s not an adoption curve. That’s a standoff, and it’s the most interesting thing happening in freight technology right now.

Listen to who’s talking

Here’s the part the trade press mostly misses: the conversation about AI in freight is loud on one side and silent on the other, and the loud side is entirely made of people selling it.

Chain launched a booking agent in June and a customer described it as “an Iron Man suit” for their reps. Numeo is pitching “trust infrastructure.” C.H. Robinson says its AI now handles 95% of its missed-pickup checks and saves hundreds of hours a day. Every one of those is real, and every one of those is a company whose revenue depends on you believing the wave is inevitable. That doesn’t make them wrong. It makes them interested.

The 48% who passed have no chorus. They didn’t put out a release that said “we evaluated three agents and kept our money.” They just kept their money. And because silence doesn’t trend, that half of the industry gets read as behind — the laggards who’ll wake up when it’s too late.

I don’t buy that read, and neither should you. Because when you actually talk to the operators in that 48%, they’re not confused about what AI is. They’ve done a different piece of math.

The math the skeptics already ran

For a large brokerage running dozens of reps on repeatable lanes, an AI booking agent that automates 20–40% of routine loads is a straightforward win — you’re spreading a subscription across a lot of transactions and buying back real hours. The ROI is obvious, which is why the big shops are in the 41%.

For a one-truck owner-operator or a three-person dispatch office, the same tool is a different equation. A $300-to-$500-a-month “agent” has to replace enough labor to pay for itself, and when the labor it replaces is you, sitting in the same seat either way, it often doesn’t. The honest answer for a lot of small operations is that the standalone AI-dispatch subscription doesn’t clear its own cost yet — and passing on it isn’t a failure of imagination. It’s reading your own P&L correctly.

There’s a second, quieter reason. The operators most skeptical of the next AI black box are frequently the ones who most recently got a piece of software forced on them — the carrier-vetting platforms that started demanding ELD logins and personal data to book a brokered load. One owner-operator, refusing to hand over an ELD password, put it bluntly on an industry forum: “I refuse to give Highway my personal login ID and password.” That’s not anti-technology. That’s someone who’s been asked to trade control and data for access before, and learned to read the fine print. When the next tool shows up asking for deeper access in exchange for a promise, that operator’s skepticism is earned, not ignorant.

The AI that’s working augments a human. The AI that stumbles is the AI you asked to decide.

— The line that sorts the whole 41/48 split.

The line that actually sorts it

So drop the pro-AI / anti-AI frame. It’s the wrong axis. Here’s the one that works.

The AI that’s quietly succeeding in freight augments a human. It drafts the carrier email you still send. It catches the missed pickup you still resolve. It pre-books the load that was already going to move on a lane you already trust. The human stays in the loop and keeps the judgment; the machine takes the typing.

The AI that stumbles is the AI you ask to decide. Set the rate with no one checking. Pick the carrier with no one accountable. Book the load and move on. That’s where the confident-but-wrong errors live — the made-up rate, the carrier that looked clean and wasn’t — and it’s exactly where the trust problem underneath all of this surfaces. Notice that the freshest AI pitches (Numeo’s “trust infrastructure,” the vetting integrations) are all trying to solve trust precisely because autonomy without trust is where the money gets stolen. This week’s cargo-theft headlines — criminals calling from inside carriers’ own phone systems — are the reminder that “let the system decide” has a failure mode measured in six figures.

Where to draw it

The move isn’t to adopt or abstain. It’s to place the augment-versus-autonomy line deliberately, for your operation, before a vendor’s demo places it for you.

Let AI touch the low-stakes, high-volume, reversible work first: summarizing documents, drafting communications, cleaning up data, flagging the exception for a human to judge. Keep three things human for now — the price, the carrier, and any compliance answer. Then move the line one notch at a time as you build trust in the tool, not as the sales cycle pressures you to.

A 40-truck brokerage and a solo owner-operator will draw that line in different places, and both can be right. What neither should do is let the loudest voice in the room — a voice that is, almost always, selling something — draw it for them.

The move

This isn’t really a story about AI. It’s a story about who gets to decide how you run your business, and the discipline travels: when anyone puts an inevitable-sounding tool in front of you, ask what does it actually do, what does it cost me at my size, and does it augment my judgment or replace it? Answer those three honestly and you’ll know which side of the 41/48 split is yours — and, more useful, you’ll have gotten there on your own math instead of someone else’s press release.

Issue #8 of Freight/Signal · back to all issues · subscribe to the Tuesday newsletter

Sources