BARCELONA, Spain — In 2026, AI in business faces a turning point. Recently, large language models (LLMs) led to double-digit drops in stock prices for major software companies such as Salesforce, Intuit, and ServiceNow.
The main reason for the sellout was the launch of new features in Anthropic’s Claude Cowork. This AI assistant lets users assign agents to handle various computer tasks via natural-language prompts.
At Integrated Systems Europe (ISE) 2026, Sol Rashidi, an experienced executive in data infrastructure for global supply chains, gave a keynote that focused more on the industry’s challenges than its successes.
Rashidi, whose career has followed the growth of commercial AI from IBM Watson in 2011 to executive roles at Merck, Sony Music, and Estée Lauder, focused her talk on practical solutions.
She argued that although the industrial and logistics sectors have quickly adopted AI, turning these technologies into real automation is still held back by governance issues, energy constraints, and confusion about what machines can actually do.
Rashidi highlighted a key problem: LLMs are advanced, but they still struggle in complex real-world settings. She described this as “perpetual POC purgatory,” where pilot projects never become fully working systems.
According to data she presented during the keynote, between 74% and 88% of AI initiatives are paused, stopped, or canceled at the proof of concept (POC) stage. “Not all my babies graduated, went to college,” she noted, revealing that of over 200 initiatives, only approximately 63 reached production, and only 39 remain alive today.
For supply chain leaders and factory operators, this shows capital is being used inefficiently. Many companies have rushed to adopt new AI tools without first ensuring their master data management (MDM) and enterprise resource planning (ERP) systems are robust.
Rashidi, who started her career as an MDM specialist leading large ERP projects, believes these problems come from ignoring the basics. “Context matters,” she said, stressing that without reliable data, AI in logistics can become a risk instead of a benefit.
To fix this, Rashidi called for a return to the “4 D’s” utility model, which matches the original goals of industrial automation. She said AI should mostly handle tasks that are “dull, dirty, and dangerous,” or require “massive data processing.”
For factory automation, this model gives a simple way to decide where to invest. Rashidi mentioned “janitorial cleaning” and “sending robots with hazmat materials” as good examples of ways AI protects people rather than replacing them. Using autonomous systems for hazardous waste or dangerous environments fits this idea well.
However, Rashidi warned that the market is moving away from these practical uses and is now trying to replace creative and thinking jobs. “It was not meant to do the reverse,” she said, asking leaders: “Are we still maintaining the components that make us unique?” She believes supply chains need human judgment in unusual situations, and that automation should handle only repetitive or dangerous work.
A big part of her talk focused on the move from generative AI, which creates content, to agentic AI, which carries out tasks. This change brings serious risks for the security systems that protect global supply chains. “Agents break a lot of the things that we do now,” Rashidi said, especially when it comes to data access and governance.
In a traditional secure facility, granting a human data scientist access to a sensitive database is a rigorous process. Rashidi recalled that it would take “three to four months to beg, borrow, and steal access” for a human employee who has legal accountability.
On the other hand, as companies rush to automate, they are “opening the kimono for agents,” giving software broad access to OT and IT systems without the same careful checks.
This discrepancy creates a vulnerability in the supply chain where autonomous agents—lacking moral responsibility or legal liability—operate with high-level privileges. Rashidi predicts that the speed of agentic transactions will soon render human oversight impossible for real-time operations. “The notion of the human in the loop will fundamentally evaporate when agents hit critical mass,” she stated.
To mitigate this, Rashidi forecasts the emergence of a new layer of automated governance. To address this, she expects a new layer of automated oversight: security agents. She predicted that soon, companies will need to use AI to monitor other AI, since humans “cannot keep pace with the speed of agents.”
During the Q&A, EE Times raised the issue of a “flywheel” effect, where AI is now helping the design of the chips that run AI, and AI could eventually write the code that executes AI.
Rashidi acknowledged this closed-loop dependency as a systemic risk. She validated the observation that “software architects are already using AI to design the code that is actually doing AI,” noting that this accelerates the industry into a “bubble” of dependency. The risk for the global supply chain is that the opacity of these systems increases as the human element is no longer in the design and verification loops.
The so-called “flywheel effect” is powered by market forces that value speed over long-term stability. Rashidi compared it to the invention of the “infinite scroll,” a feature that led to unexpected social effects.
In the same way, quickly adding AI to semiconductor design and logistics is a “shiny new toy” that could hide long-term problems with system control and auditability.” I don’t think individuals know the long-term consequences,” she said.
Beyond software and hardware, Rashidi often talked about the physical limits of automation. She pointed out that agentic AI requires significant computing power, which clashes with the industry’s energy limits. She noted that one AI prompt uses as much energy as recycling 47 plastic bottles.
For global manufacturers trying to balance ESG (environmental, social, and governance) goals with digital change, this is a real conflict. “It’s not by accident that you don’t hear companies talking about their ESG initiatives anymore,” Rashidi said, noting that rapid AI growth doesn’t fit with carbon neutrality targets.
The energy gap is huge. Rashidi shared data showing that if every Fortune 1,000 company fully used AI, it would take as much power as the entire U.S. electrical grid. This means that the real limit for factory automation may be the power supply, not software.
As automation grows, the industrial workforce is being reorganized. Rashidi described a future where work shifts from individuals to “pods.” In this setup, managers will oversee groups of people and AI agents, managing workflows rather than performing tasks themselves.
However, Rashidi is worried about how the next generation of logistics and operations leaders will gain experience. As entry-level analysis and data work go to algorithms, junior staff miss out on the learning that comes from doing these tasks. “If we’re not teaching the younger generation, where are they going to gain the experience?” she asked.
She warned about “intellectual atrophy,” in which workers lose the skills for critical thinking and problem-solving because they rely too heavily on automation. This is especially risky in supply chain management, where being able to handle new problems—what Rashidi calls “ingenuity” and “prudence”—is key. “AI fails at prudence,” she said. “Only we can read a room. Only we understand nuances that are unspoken.”
To protect human skills, Rashidi suggested measuring “effectiveness” instead of just productivity or speed. She said that saving time is not a good measure if it leads to a weaker workforce or supply chain. “So what if you’re doing more of the wrong things?” she asked.
She introduced the “human amplification index” (HAI) as a new way to measure technology investments. This model looks at whether technology helps people do more or makes them unnecessary. “Outsource your tasks. Do not outsource your critical thinking,” she implored.
In industry, using HAI means choosing systems that give operators better data to make decisions, instead of removing them from the process. Rashidi said the goal is to create “force multipliers” in which technology supports employees’ expertise rather than replacing them.
Rashidi’s talk at ISE 2026 was a warning against the industry’s rapid pace. For supply chain and manufacturing, the message is clear: the time for experimental pilots is ending. The future will need a strong focus on the “4 D’s,” solving the energy problem and treating autonomous agents with the same caution as outside contractors.
Rashidi called for active governance, urging the audience to make sure AI is integrated “with us, not to us.” In a field focused on moving goods, she said the real advantage is human judgment. “Not doing anything is the quickest way to become irrelevant,” she warned, but acting without care is the fastest way to lose control.
