The AI conversation in logistics has shifted. on s. In the latest installment of Lean Quarterly Dive with FreightWaves and the team at Lean Solutions Group (LSG), Thomas Wasson sat down with Alfonso Quijano, chief technology officer and co-founder at LSG, to talk about how AI and automation are actually playing out inside freight brokerages and logistics companies.
For much of 2025, frontier AI labs promised agentic workflows would remake entire job categories. Quijano said that promise didn’t land the way it was pitched, and companies without existing technology infrastructure are discovering the gap between access to AI and the ability to use it well.
“I’m seeing that the experimentation phase is mostly over,” Quijano said. “You tried it, you went into it without a lot of experience. Companies that didn’t have technology teams found themselves investing a ton of money into AI to see if it worked for them.”
Broad access to AI platforms didn’t eliminate the need for fundamentals, though.
“Even if the technology had democratized access and everybody could download OpenAI and Claude, it doesn’t mean that you’re an AI company,” Quijano said. “You actually need good project management. You need good change management. You need to have good control of your costs. Who knew?”
The rise and fallout of “vibe coding” (using AI to generate software with minimal traditional development) was a cautionary tale for the industry, according to Quijano.
“An independent publisher put out that 99% of vibe coded apps are in the garbage,” Quijano said. “[These apps] are not making any money. They get published and then get unpublished from the infrastructure platform.”
The lesson, he said, is that speed of creation doesn’t equate to business value.
“Garbage in, garbage out. Even if it’s faster to create technology, that doesn’t mean that you’re going to create more businesses,” Quijano said.
According to Quijano, applications built hastily during the height of the AI wave were often fragile once deployed. “It seems to work, but in reality it’s very brittle,” he said. “When you apply it in real life and in production, it can cause some pretty important damage within your business.”
Rather than chasing AI-first branding, Quijano argued the more mature approach treats AI as one component within a broader automation strategy, deployed only where it adds value.
“It’s better for AI to be present but not talked about than the other way around,” Quijano said. “You kind of have to make it invisible for it to be adopted as it should within organizations.”
Many companies have wrapped AI around problems that once had straightforward, rules-based solutions.
“You’re taking something that would have otherwise been pretty simple and now using AI for it, and the unpredictable nature of the outcome is causing issues,” Quijano said. “You’re kind of better off saying, ‘I’m going to bring automation into my company’ rather than being an AI-first or AI-native solution. That automation can have AI components when it’s necessary, but no more than that.”
According to Quijano, organizations should conceptually treat the technology not as software to configure, but as personnel to onboard. “I don’t think it’s a tool. I think it’s an employee,” he said. “What do you do with a junior employee that joins your company? You train them. You ensure that you give them a very defined job description so that they know exactly what they need to do.”
That means documented processes, exception handling and ongoing training.
“You need a very defined SOP,” Quijano said. “You need to ensure that your process is well-documented and that you account for potential errors in the process. Exception management. You need to have continuous training. You need to close the loop.”
Lean has been building that closed-loop infrastructure directly into its own technology stack. The company recently expanded its LeanTek platform with new AI governance, workforce intelligence and cost visibility tools, giving organizations execution-level cost tracking, workflow feedback loops and centralized governance dashboards designed to move AI deployments past experimentation and into accountable, measurable production use.
“Organizations are increasingly focused on how to manage, measure and scale AI responsibly,” Quijano said of the update. “These new capabilities give leaders the confidence to scale AI while ensuring transparency, accountability and measurable business outcomes.”
Skipping the groundwork produces the same outcome as a mismanaged hire.
“You’re eventually going to have to fire this new employee, which is another thing that I’m seeing very, very frequently now,” Quijano said. “There was a wave of AI projects that started earlier in ‘25 that companies are suddenly finding ways to get out of.”
Quijano also pushed back on the idea that AI can fix broken processes or underperforming teams. “You can actually increase the output of errors into your system, potentially,” he said.
LSG was founded when brokerages needed nearshore talent to fill roles they couldn’t staff domestically. The conditions are now such that there’s an increased demand for people who sit at the intersection of operations, business, and technology.
“You have to be able to move between knowing what the right metrics are that allow you to show that this really is providing value for the business, knowing how to implement it, and knowing what technology implementation really looks like,” Quijano said.
That shift requires rethinking existing roles, Quijano said, describing a “cross-pollination” approach borrowed from software development. “You have very strong technology teams that sometimes you want to convert into pods that have greater coverage capacity amongst other projects that are adjacent to your own,” he said.
Many companies deploying AI at scale are grappling with runaway token spend. According to leading AI companies, there are hundreds of thousands of agents deployed across various companies. Quijano says that many companies are building more than they need, and that they are quickly relying on token usage more than anticipated.
LSG is itself reliant on Claude for engineering work.
“We’re big proponents of an AI-first strategy towards coding, but every leading AI company is also saying you have to start thinking about your own AI models,” Quijano said. “I think that’s the next direction that this is going to go.”
It’s questionable whether the newest frontier models justify their added cost. “When you compare [the newest models] to what you have the capability to do today, it’s about six to seven percent better,” Quijano said. “But at the same time, it’s twice as expensive. In my eyes this seems more like a strategy to monetize and pump this AI narrative rather than actually a humongous stride in AI improvement. We’re definitely in the law of diminishing returns right now.”
That plateau raises a bigger question, he said. “Does that mean that we’re at a point where the open source community can provide me better bang for my buck for deploying AI models than what the frontier labs are going to do? It’s yet to be seen. We’re doing it, but it’s not easy. Now it requires technology knowledge, infrastructure. You require DevOps experience. Now you’re back to the basics.”
LSG’s strategy has centered from the start on augmenting people rather than displacing them. Quijano expects the rest of the industry to converge toward that model as the hype cycle settles.
“We decided to do that from the get-go,” he said. “We talk about human plus AI, or tech-powered humans. We talk about elevating the role of the human to have more responsibilities.” That elevation requires real investment in reskilling the existing workforce, he said. “New responsibilities involve training and the upskilling of these roles so that they can be more tech-focused. They can have the ability to not only use the software, but be a part of it, part of the solution.”
Click here to learn more about Lean Solutions Group.
The post The AI Experimentation Phase Is Over appeared first on FreightWaves.
