How AI Is Shaping Manufacturing’s Future

For Connecticut’s biggest manufacturers, adopting artificial intelligence isn’t a question of if—it’s a question of how fast, how far, and how to prove the technology’s value.

“We want to be able to leverage AI to learn, to become smarter, but also to be more efficient and to move faster as a business,” Paul Link, senior program manager, data and artificial intelligence at Sikorsky, told an audience of more than 550 at CBIA’s Oct. 1 Made in Connecticut: 2026 Manufacturing Summit.

Link was part of a summit discussion with Pete Bradley, principal fellow in digital tools and data science at RTX Pratt & Whitney, and Gary Cooper, director of attack submarine electrical engineering and design at General Dynamics Electric Boat, that was moderated by Connecticut Center for Advanced Technology chief technology officer Amy Thompson.

The panelists all described how AI is already reshaping the way their companies measure productivity, train workers, and compete for talent.

All three cautioned that building the proper foundation for AI implementation—including protecting sensitive information—is a critical first step.

Assigning Value

Thompson opened the discussion by asking how the companies assign value to their AI investments.

“What we’re looking for in our use of AI is really to accelerate every aspect of our business,” Cooper said, noting that Electric Boat’s Pentagon customers are demanding more product, faster.

He said isolating AI’s specific return on investment is difficult because it’s often intertwined with other process improvements, so his team instead tracks throughput and rework—“the killer of productivity.”

Bradley described AI’s return in more varied terms, pointing to an inspection process where the technology now measures small part features in seconds rather than minutes.

“That really has a couple of different returns,” he said.

“First of all, we’re able to do things in seconds that used to take minutes. But the other thing is, the way they were doing it before was really painful for the operator.”

Link said Sikorsky evaluates ROI through “two big buckets: speed and intelligence,” comparing process speed and accuracy before and after AI implementation, along with employee adoption rates.

Technology Sourcing

Asked how their companies source AI capability, all three said the answer is some combination of building, buying, and partnering.

“We build, we buy, we partner,” said Cooper.

He noted that Electric Boat draws on AI expertise already built within General Dynamics, adding that “the real power” comes from pairing subject matter experts with AI tools to simplify complex work orders for mechanics.

Bradley noted Pratt & Whitney’s recent acquisition of Aiir, a Netherlands-based borescope inspection company, to strengthen its predictive maintenance capabilities.

“It’s going to allow us to once again predict challenges before they occur, go faster, and serve our customers better,” he said.

Sikorsky leans on frontier AI model providers while building its own applications through data integration, Link said.

“AI is evolving so rapidly, so as new capabilities become available, we want to make sure we’re able to leverage them and then build off of that base technology,” he told the audience.

Vetting AI vendors, the panelists agreed, is its own challenge.

Bradley warned against startups that lead with questions about a company’s problems rather than concrete capabilities.

“There are a lot of these folks out there,” he said.

“You’ve really got to focus on what they’re bringing, because that’s going to be a key element to whether you’re successful or not.”

Cooper said he looks for proof of past results and strong data protections.

“We got to make sure that our data doesn’t train the world how to build submarines,” he said.

‘Data Is Foundational’

All three panelists pushed back on the notion that AI can simply be pointed at messy data and produce reliable answers.

“The data side of the house we certainly view as being foundational,” Link said, describing years of investment in structuring and centralizing Sikorsky’s data before layering AI on top.

“Once you have that digital connectivity, and you have high-powered models that that can actually reason across your data in the context of your business, it’s rocket fuel in the AI space,” he added.

Bradley agreed, calling it “the grind” behind the more visible, “flashy” uses of AI.

“It’s really important that you focus on it, that you recognize and understand the quality of your data and how you’re making it more accessible,” he said.

Cooper said Electric Boat tested the risk directly.

“We purposely injected bad data to see if AI would still work with it,” he said. “A lot of times, the answer is yes, it will.”

Bradley added that AI “will enthusiastically give you the wrong answer” when fed bad information—a line that drew laughs but underscored the panel’s shared emphasis on keeping “a human in the loop.”

Shop Floor Adoption

The panelists were candid about how far AI has—and hasn’t—reached the factory floor.

Cooper said Electric Boat’s white-collar workforce is currently focused on building tools to help mechanics, welders, machinists, and pipefitters, rather than having frontline workers generate AI outputs themselves.

“It’s going to be an interesting thing we evolve through over the next couple of years,” he said.

Bradley said Pratt & Whitney is tailoring AI training by role, while Link said Sikorsky treats AI adoption like any technology change—requiring strong change management and reassurance that AI is “an assistant” rather than a replacement.

On hiring, Cooper said Electric Boat is now recruiting data scientists to help structure “125 years worth of data,” while Bradley noted that Pratt & Whitney increasingly looks for engineers who combine deep domain expertise with AI knowledge.

“Those people are very much in demand,” Bradley said.

Asked where AI will have the biggest impact in the next two to four years, Cooper pointed to more predictable supply chains and manufacturing processes that catch bottlenecks before they slow production.

Bradley said he’s watching AI-enabled generative engineering design, while cautioning that AI is evolving “quarter to quarter” rather than year to year.

Link predicted faster gains in production yield and a narrowing gap between AI and robotics.

Supply Chain Adoption

Thompson asked the three panelists what advice they would give companies in their supply chains that are exploring AI adoption.

Cooper said the priority was to “get your data in order, get it structured, get it validated, make sure you know what the AI is working with.”

“The next thing I think that’s important is is to train your workforce, help your workforce understand, like ‘what are the pitfalls with AI? What are the what are things it can do for you?'” he added.

Pratt & Whitney surveyed 1,500 of its supply chain companiers about AI adoption, and Bradley noted that “about 50% of our suppliers are actively exploring AI or deploying AI at this point, and that’s exciting.”

“You need to think about three things,” he said. “You need to think about fear, hype, and opportunity.

“It’s really important to recognize that that done right, this is an opportunity to go further to build more.

“Opportunity is reall where the rubber meets the road.”

Pratt & Whitney’s Pete Bradley

“The second thing is is hype, and there’s nothing more hyped than AI right now. It’s really important to understand that this is not magic. It’s not even intelligence, right?

“It’s really math and software, and if you can understand that it’s just a very powerful math and software tool—that I think is a really key thing.

“Finally, opportunity is really where the where the rubber meets the road. What are your challenges? What are the things that you think you could go after, and how do you capitalize on that?”

Link advises suppliers to “start with understanding your pain points, where you have challenges, bottlenecks, deficiencies, and then evaluating AI as a potential solution.”

“Start from the opportunity side, and then it’s just being digital, right?” he said.

“As much as you can move and progress in the digital space, that’s going to help you prepare for AI adoption and utilization in the years to come.”

Resources

During audience questions, Reza Gottsi, head of the engineering department at Central Connecticut State University, asked whether industry could help universities afford the computing infrastructure needed to teach AI.

Bradley, a former supercomputing specialist, acknowledged the strain GPU demand has put on the market, while Cooper suggested smaller, lower-cost systems—“a supercharged desktop on steroids” costing tens of thousands rather than millions—could let developers build AI capabilities before scaling to the cloud.

“When things scale, things get more affordable.”

CCAT’s Amy Thompson

Thompson pointed to state-level efforts, including QuantumCT and CTNext-backed initiatives, aimed at pooling computing resources for smaller organizations and universities.

“When things scale and grow over that technology curve, those prices come down and things get more affordable,” she said.

“But that’s where investment from the state and other folks is really helpful in the beginning.”