Why ChatGPT and other LLMs aren’t a winning strategy
For this experiment, I used the professional version of Google Gemini because it is already integrated into a lot of the software tools I use.
Prompt: Review my interview with Jake Dettmer, SVP of product for Optimal Dynamics. Explain why a fleet shouldn't just buy a ChatGPT enterprise license, dump their data into it, and call it a day.
Output: While Large Language Models like ChatGPT are excellent at processing text and writing articulately, they lack true decision logic and the ability to do complex math. As Dettmer noted, ChatGPT is highly "agreeable"—if you tell it that its answer is wrong, it immediately agrees with you rather than defending a calculated strategy. If a fleet deploys a generic agent to automate a task like freight procurement based on simple rules, it could drive marginal profitability to the floor faster than a human could. Instead, fleets require a purpose-built "decision engine" that uses stochastic optimization to evaluate real-time variables and prioritize actual profitability.
This is a critical lesson. Technology has made deploying AI agents incredibly accessible. You can build one in minutes to automate a phone call. But automating the wrong task just speeds up your mistakes.
From reporting data to executing actions
The real operational transition is from a “system of record” to a “system of action.” As Beyond Trucks CEO Hans Galland pointed out during TCA, traditional transportation management systems were built strictly to report data. By unifying these fragmented data streams, modern platforms can inject decision intelligence directly into fleet workflows.
Consider asset management. Instead of a dispatcher using a clipboard during yard checks, an AI-powered system tracks nuanced trailer statuses—such as the commodity previously hauled and specific wash requirements. It then uses natural language processing to adjust load planning on the fly. You can tell your AI agent to “avoid the Des Moines area for the next four hours due to a blizzard,” and the algorithm updates instantly.
But getting an AI to actually do something requires deep connectivity.
Prompt: Take the explanation from Dwayne Lazarre, Trimble’s North American VP of business development for telematics, about AI moving from “suggesting” a fix to actually “executing” an action plan. Why is integration the biggest hurdle for standalone AI platforms?
Output: An AI agent cannot execute a real-world action without API access to the fleet's entire ecosystem. When a truck breaks down, a fully integrated agent can read the fault code, cross-reference the truck's location, book an appointment at a preferred vendor, and dispatch the driver. A standalone AI cannot perform these actions without deep, prebuilt integrations with the ELD, navigation, and backend workflows.
That level of automation naturally makes some industry veterans nervous. Are we just replacing dispatchers with robots? Not exactly. It is about keeping the humans in the loop while replacing some of that human busywork that has defined transportation for decades.
