Cario has released a new paper examining how artificial intelligence is beginning to rewire freight management for large shippers, with implications for the brokers, 4PLs and software platforms that support them.
One morning earlier this year, the freight team at a large Australian retailer arrived at work to find an alert waiting for them.
Australia Post had increased its fuel surcharge. A competing carrier on the same routes had not, and was not due to for another two months because of its hedging strategy. For 60 days, the retailer could shift a large share of its parcel volume from one carrier to the other, saving the business nearly half a million dollars.
The alert did not come from an analyst. A system had observed both carriers’ rate movements overnight, modelled the cost impact across the retailer’s freight profile, and surfaced the opportunity before the team’s first coffee.
In the old model, no one would have spotted it. Or someone would have noticed eight weeks later in a quarterly review, when much of the window had already closed.
This is what is actually changing in freight, and it has little to do with the version of AI most vendors are trying to sell.
For most of its history, freight management has been a rear-view exercise. Businesses looked at what happened last month, asked why a route underperformed, ran a tender, and adjusted. By the time the analysis was complete, the conditions that produced it had often changed.
That loop is now collapsing. Exception reports that once took a team of people hours to assemble can now run dozens of times a day with one person supervising. Proactive customer calls about deliveries likely to slip, once a differentiator, are quickly becoming the floor.
The real shift is not that freight management is getting better dashboards. It is becoming a different kind of activity. The work is moving from retrospective to real-time, from real-time to predictive, and from predictive to autonomous. Each step changes who needs to be involved in decisions, and how many decisions a single person can reasonably oversee.
Talk to senior supply chain leaders about AI for long enough and they often stop talking about AI. What they describe instead is a wish for freight to stop being a problem they have to constantly manage. They do not want another dashboard. They want the right exceptions surfaced only when human judgement is actually needed.
The emerging label for this is the “autonomous freight orchestration layer”. Strip away the jargon and it means three things: real-time exception management, proactive intervention, and eventually full orchestration, where agents observe, decide and execute, with humans involved at the points where judgement matters.
The leading edge is closer than many assume. Large freight operators are already deploying AI agents to automate shipping tasks, driver follow-ups, appointment scheduling and other repetitive workflows. The opening example is notable today, but by 2028 it may be ordinary.
However, this is where the industry’s public story often becomes too neat.
Most AI sold into freight today will not deliver what is promised. Some of it is simply existing automation rebranded as agentic AI. Some is a chatbot placed on top of messy operations and marketed as transformation. Some may be genuinely capable technology, but deployed into environments that were never designed for autonomous systems.
Freight remains one of the messiest operational domains in the enterprise stack. Carrier data arrives in different formats and at different levels of quality. Critical workflows still sit in inboxes, spreadsheets, PDFs and tribal knowledge. Many shippers are still trying to reconcile invoices accurately, standardise master data, and gain visibility over freight they have already booked.
Layering sophisticated AI on top of fragmented operational plumbing does not automatically create autonomy. In some cases, it simply creates faster confusion.
This is the market tension. Generative AI has likely been overestimated in the short term and underestimated in the long term. Many organisations have become caught up in the immediate excitement without appreciating that long-term value depends on governed, accessible and operationally useful data.
The tools themselves are often more capable than businesses realise. Many companies already have AI functionality inside existing platforms that remains largely unused. The bottleneck is less about model intelligence and more about operational readiness, workflow design, governance and data maturity.
For freight buyers, this makes vendor evaluation more important. A genuine agent should be able to observe, decide and act. It should know what to do when uncertain, including when to escalate to a human. It should leave an audit trail. It should have clear data requirements. And it should have a plan for what happens when it gets something wrong.
If a vendor cannot explain those points clearly, it may be selling a demo rather than a deployment.
The same caution applies on the demand side. One Australian freight executive recently described a call with a major retail client where she walked through a sophisticated AI use case, only to be interrupted. “That’s all great,” the client said, “but I just need my carriers to actually turn up. Right now, I just need the basics.”
That gap matters. A meaningful portion of large shippers are still working through carrier management, invoice reconciliation and basic visibility. AI does not solve those problems by itself.
The shippers getting value today are not the ones trying to automate everything at once. They are picking narrow, painful problems and applying AI where the risk is low and the outcome is measurable. The highest-return use cases are often unglamorous: invoice anomaly detection, carrier allocation, proactive ETA intervention, appointment scheduling, proof-of-delivery matching and exception triage.
These are not science-fiction outcomes. They are operational friction points being reduced incrementally.
If autonomous orchestration is where freight is heading, the next question is who delivers it.
The simplest answer is that AI cuts brokers and 4PLs out of the equation. Direct shipper-to-carrier matching becomes faster and cheaper, and the broker’s margin is captured by the algorithm. Some startups are already built around that premise.
The more realistic answer is more complicated. Routine brokerage work, including matching, quoting, document handling and milestone tracking, will continue to be automated. But brokers and 4PLs will not disappear if they move up the stack.
Their value will sit in network design, complex commercial decisions, crisis response, dangerous goods compliance, multi-modal exception handling and the moments where judgement matters more than speed. They may also become the implementation layer that helps shippers get value from platforms they do not have the time or capability to fully configure themselves.
For shippers, this changes how partners should be assessed. Carrier coverage, rates and account management still matter. But two new questions now sit alongside them: how will your AI capability compound my advantage, and what happens when something goes wrong that the AI cannot handle?
There is also a deeper platform question. Freight systems are shifting from being databases with interfaces to becoming the plumbing beneath agent-led workflows. Some agents will be built into freight platforms. Some will sit in ERP systems and call freight data from other applications. Some will be built in-house by sophisticated shippers.
Most businesses will use a combination of all three. But that means the choice of freight platform now carries more weight than it did when the platform’s job was simply to store bookings, rates and milestones.
If a shipper changes freight platforms in 2028, it may not just be replacing software. It may be replacing the foundation on which years of agent development have been built. That makes the plumbing decision more important than many buyers realise.
The practical advice is simple: try something.
Not a multi-year transformation program. Not a board-approved AI roadmap. Start with one narrowly scoped agent applied to one painful problem, with low risk, low cost and a clear way to measure whether it worked. Then try another.
The companies best positioned for AI in freight may not be the ones with the most advanced models. They will be the ones with the cleanest foundations and the fastest experimentation cycles.
The path to a transformed freight operation in 2028 is not necessarily a transformation project in 2026. It is a portfolio of small, working agents, tested honestly, expanded selectively and stitched together over time into something that eventually looks like transformation.
Freight has always been a problem most leaders wanted to forget. For the first time, that may be becoming possible. The companies that get there will be the ones that started.
For more information, or to see a demo of the AI features in Cario, click here.
.png){kind=small}
.png){kind=small}
