Troy Lester is chief revenue officer and co-founder of Warp, a software platform for optimizing the routing of “middle-mile” freight. Despite ambitious plans for growth, he doesn’t expect to need more than another 10 full-time employees on staff — ever.

The reason, he says, is artificial intelligence.

While others in the logistics sphere make tentative inroads into AI for data analysis and decision-making, Warp is going all-in on technology for managing transportation between stores, warehouses, consolidation facilities and sort centers.

“We’ve always dreamed of this world where AI can make better decisions than humans on what needs to happen within a supply chain,” Lester says. Robots have been operating in warehouses for years, he adds, but what’s been missing up to now is a “super-intelligence component” that directs their movement in the most efficient manner. And that requirement extends to the broader world of managing domestic transportation.

Thanks to recent advances in AI, Lester says, the technology is finally able to analyze, plan and facilitate the movement of goods between various modes of trucking: truckload, less-than-truckload, parcel and van, as well as third-party crossdocking.

The biggest expense for Warp in its 10 years of operation has been headcount. Now, Lester says, decisions that used to require human intuition are being made by AI.

Given the complexity of transportation networks, it hasn’t been an easy transition, he acknowledges. “It took a lot of time to achieve interconnectivity, but once the workflows started learning on their own, it got better — and freaked us out.”

That today’s AI can handle brutally complex routing puzzles seems plausible. But doesn’t a service-oriented venture like Warp need large teams of human beings on the sales and customer-relations end? Lester challenges even that assumption.

Sales, he says, is thought of as “the last bastion” of business activity requiring a human touch. And the need for people is supposed to get even more pressing as a company grows. With each funding round comes the further expansion of a bureaucratic corporate structure, often consisting of multiple vice presidents drawing salaries well into the six figures. But instead of embarking on a hiring spree, Warp leaned into technology to perform tasks that previously were assumed to be the exclusive province of humans.

“AI eliminated a lot of the specialty needs,” Lester notes, such as department managers and customer-service reps. Agentic AI — discrete modules performing specific tasks autonomously — stepped in to take their place.

The company began by automating the simpler workflows, designing a system whereby multiple AI agents collaborate with one another, as well as with the large language models that underlie generative AI applications such as ChatGPT. “How we got there was by playing around — seeing what’s possible,” Lester says. “We were able to refine and upload everything that was in our heads.”

Applying AI to traditionally human-centric sales was “scary,” he acknowledges. The model had to contain a treasure trove of knowledge about sales contacts and techniques. But it wasn’t all about cutting-edge technology. Having AI agents push out emails was no different from relying on standard sales-automation tools for the same purpose. And when it comes to the need for more “personal” customer contact, an AI-created avatar can mimic human speech to a large extent, Lester says.

Issue resolution is another area where automation efforts can easily go astray. Lester admits to early concern about whether an automated system can cope with the endless varieties of disruption that logistics networks encounter on a regular basis. To make it work, Warp programmed into the model as many scenarios as it could envision, along with proactive updates and notifications for which customers could opt in.

In the end, Lester says, the AI chatbot turned out to interact with customers “better than what you’d get from somebody overseas.”

Lester concedes that AI in its current incarnation is far from perfect. “Hallucinations,” in which the AI model delivers outright falsities and terrible advice, continue to crop up in even the most powerful and sophisticated applications. Humans are still needed to double-check and correct those errors, which, with any luck, should become less common with experience. Still, he says, “I don’t have a great answer as to whether they’re going to exist in future.”

What he’s sure of is the continuing need for at least some humans in the loop — hence those 10 flesh-and-blood entities — especially for tasks that require deep interaction with customers. At some level, their experience and knowledge base will continue to be essential to running the operation — even if their numbers are minimal.