AI has reached forwarders' P&L - now the arguments begin
AI appears poised to transform freight forwarding. And the industry's biggest players are finally starting to put numbers on it.
Kuehne+Nagel told investors at its Q2 earnings call that AI initiatives were expected to deliver savings of up to Sfr150m ($184m) in 2027, while improving productivity by around 5%.
It is not alone. CH Robinson also told investors AI was already translating into financial performance, claiming productivity improvements of more than 60% since the end of2022 had helped drive a 20% increase in adjusted operating income in the second quarter.
For an industry that has spent the past two years experimenting with new technologies, it is a clear sign that AI has moved beyond pilot projects, and onto the P&L.
But while there is growing agreement that AI will reshape freight forwarding, there is little consensus on what that will ultimately look like.
The debate has intensified following Gartner's prediction that, by 2030, 60% of supply chain management software will incorporate agentic AI, with enterprises increasingly deploying clusters of specialist AI agents that work together to complete complex tasks.
But, of course, not everyone agrees.
FreightSuite says the industry's fascination with multiple specialist agents risks solving the wrong problem.
"We're able to do a lot more with agents that are operating inside the TMS than... daisy-chaining a lot of agents outside of the TMS together," co-founder Sam Moore told The Loadstar. "The internal agents have all of the context to the TMS... external ones can't actually get the quantum of data they need to achieve the levels of automation people are setting out to get."
Rather than building thousands of specialist AI workers, FreightSuite argues that the future lies with a handful of far more capable "super agents", each able to understand an entire shipment, access every piece of operational data, and reason across multiple workflows.
"We see super agents being where the industry is going to go," said co-founder William Jacobs. "Our more controversial part of that is the thousands of small agents versus... a handful of larger agents."
Mr Jacobs argued that dividing intelligence across numerous specialist agents inevitably fragmented decision-making.
"These larger super agents... can have holistic reasoning," he said. "Whereas if you're bolting lots of agents onto fragmented data... that orchestration layer just becomes a patch on top, rather than actually solving the problem."
The company believes context is the defining advantage.
"AI is no different. It needs as much context as it can have to be able to make the right decision," Mr Jacobs said.
Not everyone shares that vision.
Former Magaya executive Kristjan Lillemets believes the industry should move in almost the opposite direction, telling The Loadstar: "I'm currently in the clusters of specialist AI agents camp, mainly due to trust.
"Building harnesses around many agents, each with a smaller set of responsibilities, allows more control and evaluations around which agent produced a faulty result. This will allow for faster troubleshooting, human intervention and fixing compared to a 'super agent', which might appear too much of a black box.
"For AI to succeed in vertical industries with deep operator expertise, trust is the key issue. The more transparent the system, the easier it is to trust it."
Mr Lillemets says the real challenge is not the sophistication of AI models but the quality of logistics data.
"The models are probably smart enough already for what we need," he said. "Data is the biggest limitation."
Robert Petti, founder of Prompt Global, believes the entire debate risks missing the point.
"I believe focusing on multi-agent versus super-agent architecture is the wrong approach," he said. "The real question is about capabilities, not structure.
"If a single agent can handle everything effectively, that is a positive outcome. The primary goal is that the agent enhances the user's or company's performance and improves the customer experience."
Instead, he argues that clean, structured data, governance and business context will prove far more important than the underlying AI architecture.
Project44 chief executive Jett McCandless also cautions against organisations accumulating ever more standalone AI tools.
"If I was advising [a freight forwarder], I'd say 'don't buy another point agent'," he told The Loadstar. Instead, he argues AI should combine APIs, specialist agents, and a central AI "brain" operating across a unified logistics data graph.
Despite their differing technical philosophies, there is one point of agreement.
None believes the future lies in AI chatbots that simply answer questions or draft emails. Instead, they expect AI to execute operational work - processing bookings, handling documentation, responding to customers, and managing exceptions - while humans increasingly focus on commercial relationships and genuinely complex problems.
FreightSuite says it has already demonstrated what that future could look like, recently completing an end-to-end shipment that was more than 90% agentic, with human involvement largely limited to customs compliance.
Whether the industry ultimately converges around specialist agents, super agents, or hybrid architectures remains an open question.
What is less disputed is that, as CH Robinson's strategy demonstrates, AI is no longer an experiment - it is a competitive advantage measured in margins, not promise.
