[Image from Sora]
More than nine out of ten manufacturers (92%) say smart-manufacturing technology will be their chief competitive edge within three years, according to Deloitte’s 2025 Smart Manufacturing and Operations survey of 600 U.S. executives. Nearly 15 years after “Industrie 4.0” promised a literal factory revolution with software and sensing advances reshaping workflows. While the benefits of “smart” manufacturing are no longer merely aspirational for many facilities, a good number have trouble finding employees to fill key production and operations management roles. The same Deloitte study found close to half (48%) of manufacturers report significant challenges in that area.
Inside the lab, momentum is just as strong. A 2024 Pistoia Alliance “Lab of the Future” survey of 200 life-science R&D execs found that 68% already run AI/ML workflows at the bench, and 62% say it will command their biggest tech spend over the next two years. Similar to in the manufacturing space, automation is becoming a must and burnout is forcing the issue. For instance, 89% of lab professionals in clinical settings say their facilities need automation just to meet demand, and 91% see AI tools as the fix, according to a Siemens Healthineers/Harris Poll of 408 U.S. lab workers.
When fallback isn’t an option anymore
According to McKinsey’s January 2025 cross-industry report, “Superagency in the workplace,”‘ only 1% of surveyed C-suite leaders describe their companies’ AI initiatives as “mature.”Yet 92% of companies plan to increase their AI investments over the next three years. On some floors operators still log machine data by hand, clipboard in hand. “You’d be surprised,” said Ivan Madera, CEO of Adaptiv AI. The fallback, he adds, is “hiring more people to do some of that work,” which is tough when Deloitte also reports 48% of manufacturers struggle to fill key production roles. A similar problem pervades much of the STEM landscape.
Ivan Madera
Madera says the broader challenge of giving process optimization short shrift is most common at firms “experiencing hyper-growth and not worrying about systems. Some people just brute-force, figure it out,” he explains. But that stop-gap is more likely to primarily scale payroll rather than productivity. “If you’re on this trajectory continuously, you have to put an intelligence system in play,” he adds. The alternative is to risk turning every output gain into a hiring spree the talent market can’t supply.
Who’s steering the brain?
Over the past decade plus, predictive maintenance has emerged as a poster child for data-driven optimization. Yet here too, there is a spectrum of maturity levels. Old-school predictive systems still rely on, say, rule-based alarms and quarterly trend reports. That’s not exactly smart, said Madera said of such old-school predictive maintenance approaches. “The problem is that people stopped using it in real time… We’re often still in the infancy of this and what’s possible.”
A factory’s AI champion doesn’t have to be a coder or a data scientist. “It doesn’t need to be a highly technical person; it needs to be someone that understands the business or the manufacturing operations,” Madera explains. “Generally speaking, you’re partnering. I’ve got the business process know-how, and [my data science team] obviously has the tech stack know-how. They know how to do this (the tech), but they don’t necessarily know how to do this (the business application). They don’t understand the business, or the workflow, or the desired outcomes of what they’re programming.”
It’s a team effort: one party supplies the process and broader domain expertise, and the data scientists of software developers, for instance, might handle the technical heavy lifting. “When we do program an agent, we’re training it the right way, so it doesn’t just say, ‘Hey, I’m an expert in manufacturing operations,’ and then suddenly start spewing things out to you that aren’t right.”
Our system is going to be SME-led, a subject matter expert that understands a specific domain very deeply, so that we can partner with our data science team and our programming team.” – Ivan Madera
Augmenting human skills with AI co-pilots
Once grounded by subject matter expertise, the AI can then serve various team members according to their specific needs and technical aptitude. “I wouldn’t see the shop floor necessarily programming… most of the shop floor is going to use it as an LLM, just asking their AI agent questions like, ‘Hey, tell me, plan my work today,’ or ‘What bottlenecks do I have that I need to address?'” Madera says.
“For example, on the engineering side: ‘Hey, provide different design considerations for part X.’ …it’s taking historical context based on prior drawings or quality issues… how do I augment the level one engineer and create a Ph.D.-level engineer that has all these different insights with all these different available datasets to improve their design function?”
How do I augment the level one engineer and create a Ph.D.-level engineer? —Ivan Madera
Towards supervisor agents and 24/7 factories
Looking to the future and the role of “humanoids,” as Madera puts it. While it’s early days for humanoids, Madera sees long-term potential in them. He paints a scenario of agentic AI in action: “Imagine [an operator is] unloading a truck, but he’s behind schedule, and we already know that. The humanoid walks over and helps you unload the truck… Because I’m predicting that, ‘Hey, [the operator] hasn’t clocked in…’ So, it knows that [the operator] needs help; go send help to [the operator] so that he can complete that task faster, so that my production line doesn’t shut down.”
Even without humanoids, the basic principle could hold true for automated systems more broadly. Madera offers an example: he asked the agent to “develop a strategic plan to improve operational efficiency.” The model instantly surfaced bottlenecks: “prioritize repairs on the idle packaging line,” “reroute parts around the paint booth,” insights that would have taken a human team days to assemble. “That kind of speed feels almost superhuman,” he says.
The agent then pulls in production, materials, cost and revenue data to give a horizontal view of the entire operation, not just a single value-chain slice. Seeing the whole factory in one dashboard is what makes this compelling, Madera adds.
When envisioning a continuous factory workflow, Madera pinpoints the workforce as the central challenge. “What type of workforce is required to operate that factory 24/7? Because that’s the biggest challenge today,” he states. This consideration runs alongside the goal of determining “how much of that factory can be fully automated.” This line of thinking, and corresponding execution, enables more dynamic operations, he explains: “So I could plan and schedule my operations based on available material, not necessarily on this pre-canned plan.”
Thread all those strands together, from skills shortages to agentic scheduling and 24/7 orchestration through a single factory, and it moves from being conceptual to becoming a math problem. The aim, Madera suggests, is a system where ‘the tool, the brain, will know that it has automatically rescheduled and re-shifted your production plan based on available capacity and resources. The firms that nail this kind of round-the-clock automation will sit atop that 92% who, as Deloitte found, view smart manufacturing as their chief competitive edge.
