By Amy Thompson, Ph.D., Chief Technology Officer, CCAT
Manufacturers want to know where AI fits on their shop floor. It is a reasonable question, but it may not be the right place to start.
AI in manufacturing is a technology issue, certainly. But it is also an economic development issue, a workforce issue, and ultimately a competitiveness issue. It will influence where companies invest, where they place R&D and production, and which regions lead the next generation of industrial innovation.
These technologies are advancing whether we act or not. If other regions create environments that enable responsible AI adoption more quickly, some regions may risk losing investment, projects, and eventually parts of the supply chain that support our manufacturing sector.
My own experience spans working as a manufacturing engineer in industry, researching and teaching advanced modeling and simulation of manufacturing and complex systems in academia, and technology research. I took my first AI course in 2002 thanks to a recommendation from Bryan Reimer, (“How to Make AI Useful”, 2025 - which I highly recommend), who at the time had great insight on how AI would affect autonomous manufacturing and vehicles over the next two decades. Before joining the Connecticut Center for Advanced Technology (CCAT) as Chief Technology Officer, I taught and conducted research in systems and digital engineering at the University of Connecticut, where I served as Associate Director of the Pratt & Whitney Institute for Advanced Systems Engineering and led its graduate systems engineering programs. Today I oversee CCAT’s Advanced Technology Center and its technology, talent development, and training teams, helping manufacturers move from evaluating advanced technologies to applying them in real-world operations. Working across these domains shaped and continues to inform how I think about AI in practice.
In my view, success with AI over the next five years will depend less on how quickly manufacturers adopt it and more on whether they build the AI foundation that makes it effective. That foundation is connected, secure, and reliable data. Workforce capabilities and an organization prepared to keep learning about AI and manage change for its adoption are equally important, and both the workforce and organizations depend on data that can be trusted.
Ask a Better Question
There is encouraging progress in how industry is approaching AI. A few years ago, the conversation was largely, “Should manufacturers be using AI?” Today it is increasingly, “How do we use it effectively, safely, and at scale?” That shift is important, because the value of AI is realized when it becomes part of manufacturing workflows rather than remaining as a pilot.
If I could change one thing about how manufacturing leaders think about AI, it would be the question they start with. Most ask, “Where can we use AI?” The better question is, “How do we build an organization that can continuously identify, evaluate, and implement AI opportunities?”
The first question tends to lead to individual projects. The second leads to organizational capability. Companies that develop the ability to evaluate, validate, and operationalize AI will be positioned to compete more effectively in both domestic and global markets.
The next five years will not be defined by any single tool or application. AI capabilities will continue to evolve rapidly. Each new capability will have to integrate with the processes, equipment, data systems, security practices, and people you already have.
Connected Data Is the Foundation
One thing we see consistently is that the barrier is not a lack of interest in AI. It is readiness, and readiness begins with data. You cannot simply add AI to a manufacturing shop floor and expect it to create sustainable value. Many manufacturers are still working to get operational data into usable digital form from paper and to build the connected infrastructure that allows AI to deliver results.
Before AI can make a good recommendation, it needs data it can trust. In manufacturing, that means structured, connected, and secure data flowing through a digital thread across the product lifecycle.
This is something that can be harder to see from inside the day-to-day of production. On the shop floor, data tends to live within a single stage, and the connections there are already complex. Step back across the full lifecycle of a complex system, from design through inspection, operation and sustainment, and those relationships multiply. When that data is fragmented in parts of the lifecycle, AI inherits the gaps and may lack the full data to make a good decision. The result is recommendation errors, subpar optimization, and findings people do not trust to deliver optimized lifecycle solutions.
That is why I suggest manufacturers pursue model-based definition and digital thread practices before or in parallel to adopting AI in manufacturing and inspection. The digital thread serves two purposes. It supports the automation of technical workflows (locally), and it forms the foundation across the lifecycle for faster, better AI applications that result in more reliable solutions (globally).
This is the focus of CCAT’s Digital Thread Lab (DTL), offering hands-on training in model-based definition and digital thread practices, beginning with awareness-level training for everyone from leadership to technical teams and continuing into pathways for digital design, digital manufacturing, and digital quality and inspection. It also includes a change management workshop for senior leaders and frontline workers developed in partnership with Quinnipiac University, because changing from paper-based methods to adopting digital, connected data is an organizational shift as much as a technical shift.
Connected data should not be viewed as a preliminary step to complete on the way to AI. It is the foundation on which effective AI applications are built. Skipping or glossing over this step may appear to save time or seem unnecessary with new and “powerful” AI, but the consequences tend to surface later, in results that are harder to validate and more costly to correct.
Why This Matters Most for Smaller Manufacturers
Connecticut has more than 4,500 manufacturers, and 61 percent of them employ fewer than 50 people. A large manufacturer may have teams dedicated to data, cybersecurity, and automation. A 50-person supplier generally does not.
Today, 37 percent of Connecticut manufacturers report that they have integrated AI into their business practices. Of those who have not, 63% say they don’t know how. At the same time, 82 percent report difficulty finding and retaining workers. (CBIA 2026 CT Manufacturing Report) Together, these figures point to a significant opportunity. AI can help these companies raise productivity and make better use of the skilled people they already have.
Small and midsized manufacturers often face the greatest challenges in adopting new technology, attracting AI and data talent, and absorbing the costs and risks of experimentation. A practical pathway must start with data and digital infrastructure, then account for workforce capabilities, cybersecurity, safety, business value, and implementation risk.
Here is the question I would ask every manufacturing leader to keep in mind: how can a 50-person manufacturer go from hearing about an AI technology to safely applying it on the shop floor with trusted data and solutions? That is the adoption challenge we need to solve.
Automate Tasks, Not People
Manufacturing is fundamentally a human-machine environment. We should not think that humans are going to be out of the loop in this domain. The human-technology relationship matters as much as the technology itself. NIST explicitly recognizes that human–AI configurations can range from fully manual to fully autonomous and that some AI applications do not require continuous human oversight. At the same time, NIST’s 2026 smart-manufacturing AI roadmap emphasizes trustworthy, explainable, reliable operation because manufacturing AI increasingly operates in high-stakes industrial environments. Human oversight of AI needs to correspond to risk and the consequence of failure.
Human-in-the-loop can be the wrong architecture for some manufacturing systems. A human cannot meaningfully approve thousands of sensor-based control decisions per second. Instead, the AI/control system operates inside a validated operating envelope while humans have supervisory authority, exception handling, and the ability to override or safely stop the process.
With workforce shortages persistent in manufacturing, AI is less about replacing workers and more about helping companies do more with the skilled people they already have. A useful guiding principle is to automate tasks, not people. The goal is people working with technology.
For that reason, workforce development should be considered from the outset of AI adoption. Education, training, assessment, technology adoption, and implementation are not separate initiatives. They are connected parts of the same pathway, and their outcomes are interdependent. Connecticut and the nation will need a pipeline of AI-capable technicians, engineers, and manufacturing technologists to make it work. A good example training program is the U.S. IBAS-funded Massachusetts Institute of Technology (MIT) Tech AMP program, which CCAT collaborates in to deliver in Connecticut. It trains future Technologists needed to plan and implement advanced manufacturing technologies, like smart manufacturing on shop floors. Not just technical skills are needed, but also critical thinking and systems thinking skills to implement effective projects.
A manufacturing technician and engineer who understands what an AI model is indicating, and when to question it, and what their role is working with it, is essential to turning an AI output into a sound decision – manufacturing personnel need understanding of manufacturing fundamentals and good critical thinking and systems thinking skills to supervise and validate AI.
A Foundation for Responsible Adoption
Manufacturing brings a valuable perspective to the responsible AI conversation. Our industry already operates under rigorous safety standards, regulatory requirements, and process controls. We test processes, qualify equipment, validate models, monitor systems, and set acceptable operating limits.
AI is the next evolution of the modeling and analytics manufacturers have used for decades to improve quality, predict outcomes, and optimize performance. Industrial AI is also different from many consumer-facing applications. It focuses on operations, maintenance, scheduling, quality, and productivity inside controlled, often complex environments.
The goal should not be uncontrolled adoption. It should be disciplined, responsible adoption that still allows innovation to move forward in the manufacturing industry.
What Organizational AI Readiness Requires
Organizational AI readiness rests on five building blocks. Each can be assessed before any AI project begins, and leaders may find it useful to ask one question of each. Connected data comes first, because every other block depends on it.
1. Connected data and digital infrastructure. Product and process data that is structured, secure, and linked across the lifecycle through a digital thread, beginning with model-based definition. Can we trace a part from its design model through machining and inspection without re-entering data and use AI to plan and predict quality and performance across these phases?
2. Workforce skills. Operators, technicians, and engineers who can work alongside AI, interpret its outputs, and recognize when to question them. Do the people closest to the process understand how an AI recommendation is generated and how to validate its recommendation?
3. Governance and security. Defined practices for how AI tools are evaluated, validated, deployed, and protected, consistent with existing quality and cybersecurity requirements. Would this AI tool pass the same qualification process we apply to new equipment or software?
4. Use case assessment. A consistent method for weighing each opportunity on business value, implementation risk, and data availability rather than novelty. Is this the problem most worth solving, and do we have accurate data to solve it?
5. A pathway from experiment to implementation. A repeatable process for moving promising pilots onto the shop floor, with clear criteria at each stage. What would need to be true for this pilot to move into production?
The final building block often presents the greatest challenge. The path runs from awareness to assessment to pilots to production, and ultimately to scale. Each step needs its own support, and each one depends on the connected data beneath it.
Organizations that develop these capabilities will be better positioned to integrate new AI technologies as they emerge, rather than starting over with each one.
Readiness Starts Before AI
Across research, education, and industry, the lesson has been consistent. Understanding technology and applying it are two different problems.
At CCAT, we focus on the second. As a nonprofit, we provide a neutral sandbox where manufacturers can understand what is possible, weigh benefits and risks, test technologies, build implementation plans, and train their people, so they can adopt AI securely, responsibly, and with measurable impact.
In our autonomous manufacturing cell, we create a digital twin, a digital representation of what is happening on the shop floor that is built on the digital thread. AI and machine learning then analyze that data to identify patterns, detect anomalies, and support better operational decisions. This illustrates the pathway in practice.
Connecticut is well positioned to lead industrial AI. It already has a strong foundation: world-class manufacturers, a robust supply chain, strong higher education partners, and an innovation ecosystem that supports technology adoption. The opportunity now is to connect those assets and make responsible AI adoption accessible, especially for small and midsized manufacturers.
In my view, the advantages in industrial AI will belong less to the organizations that move first and more to those that are prepared. That preparation begins well before AI, with connected and secure data.
___________________________
About the Author
Dr. Amy Thompson is Chief Technology Officer at the Connecticut Center for Advanced Technology (CCAT), where she oversees the Advanced Technology Center and the teams guiding manufacturers through advanced manufacturing technology evaluation, talent development, training, and implementation. She joined CCAT in 2023 as Chief Talent and Training Officer. Previously, she was Associate Professor-In-Residence of Systems Engineering at the University of Connecticut and Associate Director of the Pratt & Whitney Institute for Advanced Systems Engineering, where she led its graduate systems engineering programs. Her expertise spans model-based systems engineering, digital engineering, smart and connected manufacturing systems, supply chain design, and engineering education. Her AI thought leadership speaking and moderator engagements include AI Day Hartford, CBIZ New England Manufacturing Forum 2026, and CBIA’s Made in Connecticut Manufacturing Summit.
CCAT is a nonprofit organization that helps Connecticut manufacturers evaluate, test, and adopt advanced technologies to stay competitive. Through hands-on training, technology assessment, and implementation support, CCAT gives small and midsized manufacturers a practical pathway from awareness to adoption.
