5 industrial AI trends to watch in 2025

As many as 2.1 million manufacturing jobs could go unfilled by 2030, according to an influential 2021 projection from Deloitte and The Manufacturing Institute. Meanwhile, the Bureau of Labor Statistics projects a growth rate of 0.1% from 2023 to 2030, yielding only 110,000 projected new jobs, resulting in a potential overall change from 12.9 million jobs in 2023 to about 13 million by 2033. Exacerbating the challenge is a mismatch between the skills required and those available in the labor force.

I. The looming skills crisis (and the AI opportunity)

In 2025, AI will remain a critical bridge to address the skills gap in industrial sectors ranging from oil refineries and petrochemical plants to specialty chemical facilities and offshore platforms. But the transition will involve aligning the deterministic approaches that are the bedrock of industrial control systems, which rely on mathematical certainty, with the probabilistic nature of many AI/ML algorithms. “In our world of controls, we have worked in this deterministic space for ages, where a set of inputs always give the same outputs. And you can mathematically prove it, you can empirically prove it,” explained Jason Urso, VP & CTO, Honeywell Industrial Automation. “In the world of AI, it becomes more probabilistic.”

This paradigm shift necessitates a structured implementation framework. Analysis of successful industrial AI deployments reveals several dimensions that organizations must address simultaneously:

The above infographic illustrates four interconnected dimensions of AI integration in industrial settings:

1. Knowledge Capture: Urso emphasized the criticality of preserving institutional expertise, noting, “We want to build a massive knowledge repository from experienced workers so that enterprise knowledge isn’t lost when someone retires.” This systematic approach to knowledge preservation addresses the increasing rate of veteran worker retirement in industrial sectors.
2. AI Advisory (Pattern Recognition and Troubleshooting): Urso stressed AI’s role in equipment diagnostics and problem-solving: “If a two-year employee goes and looks at a piece of equipment and says, ‘It’s vibrating, it’s making a noise,’ the system can reach into the repository and say, ‘Actually, that has happened before.'” This capability transforms decades of maintenance records into actionable insights.
3. Skills Transfer: With industrial sectors facing significant demographic shifts, Urso highlighted AI’s potential to accelerate expertise development: “How do we make someone who is more junior have access to the same level of expertise that a 30-year veteran has?”
4. Safety Oversight: While advocating for AI integration, Urso maintained that human judgment remains essential: “There are still humans needed for safety reasons.” This is especially important in safety-critical operations such as in fixed process facilities where complete automation isn’t feasible.

II. Knowledge capture: Diverging paths to autonomy

Industrial autonomy is evolving along two distinct paths: greenfield and brownfield trajectories. “With new projects, you can design them for autonomy,” Urso said. “Take offshore platforms being built today—they’re designed to be normally unmanned, with no living staff on board.” This approach requires designing equipment to be resilient to various faults right from the start.

For established facilities like refineries and fixed process sites, autonomy follows a structured maturity model with quantifiable implementation stages. “Looking at the categories of control and automation, reliability, and human actions, each has a stack of tasks that can be addressed,” Urso notes. The process begins with automating manual procedures through strategic equipment installations and sensor deployments. Following that come implementing analytics for fault detection.

The nature of the maturity model itself is beginning to evolve with the emergence of AI agents. Deloitte projects 25% of companies across industries using generative AI will launch agentic AI pilots in 2025, growing to 50% by 2027. This adoption curve reflects both technological maturity and operational requirements across different facility types.

In the industrial sector, however, the emergence of more sophisticated AI can prove valuable by strategically reducing human presence in hazardous areas through three core capabilities: enhanced control automation, predictive reliability systems, and AI-augmented human performance.

For brownfield sites, implementation focuses on systematic sensor deployment that replicates human sensory capabilities—sight, hearing, smell—enabling remote inspection and monitoring. “It’s about gradually taking people out of hazardous areas,” Urso explains. “Do you need to send a human for a physical inspection if you can use acoustic sensors or video?”

III. AI advisory systems enhance operational efficiency

As industrial sectors navigate demographic shifts and technological evolution, the implementation of intelligent decision support systems emerges as a bridge for operational knowledge transfer. “Time to expertise is something that’s on everyone’s mind,” Urso explains. The traditional path—where new employees spend extensive time training and observing others before working independently—is becoming ever-more difficult as experienced workers retire at a faster rate than new workers can be trained.

Urso describes how AI could help newer employees access the enterprise’s accumulated knowledge. He gives an example: “If a two-year employee goes and looks at a piece of equipment and says, ‘It’s vibrating, it’s making a noise, I’m not sure what that is,’ they can enter that into the system.” An AI system can then search through historical service records to find similar cases and suggest solutions based on how these issues were resolved in the past.

IV. AI systems accelerate the transfer of industrial expertise

This approach isn’t about removing human judgment but rather about giving workers better access to the organization’s collective experience. “Being able to draw on the enterprise’s expertise or an enterprise’s knowledge base is much like having the 30-year expert standing right next to you,” Urso says.

The World Economic Forum notes that while organizations report that generative AI enables employees to do more enjoyable, creative, and value-adding work, successful deployment requires effective change management and leadership from both top executives and middle managers who understand day-to-day operations. As a result, training programs that demystify the technology while building new skills are key for building trust and enabling workers to grow into enhanced roles.

V. Savvy industrial organizations will know when to keep humans in the loop

While automation is a trend, as Urso suggests, maintaining human oversight remains critical in many industrial settings. “We work in industries that are safety critical; you can’t make mistakes,” he notes. He suggests starting with AI in an advisory role, where it provides information to humans who make the final decisions. This allows organizations to validate the AI’s effectiveness while maintaining safety protocols. “Give it a try in areas where you’re giving advice to the human, but you still have the human to make a final decision based on their own experience and training, especially when it comes to critical operations,” Urso said. This approach parallels medical validation protocols, where AI recommendations run alongside human expertise: “Much like a doctor saying, ‘I’m going to try the AI in parallel to the doctor making the recommendation.’ When the doctor notes that the AI was wrong with the recommendation, well, give that feedback into the model, help it get better over time.”

This parallel validation approach aligns with emerging industry consensus. Early AI adopters surveyed by the World Economic Forum consistently warned that “removing humans from the loop is still considered to be a mistake.” Savvy organizations are taking a cautious approach, conducting experiments and implementing pilots within controlled environments to prevent reputational damage and regulatory conflicts.

The cultural dimension of AI adoption requires careful navigation. “There’s a spectrum of people that want to be the early adopters and prove it out… and there’s the other end of the spectrum that are concerned,” Urso notes. Rather than forcing adoption, he advocates for letting skeptics engage on their own terms while recognizing their vital input on safety and operational integrity: Early adopters and skeptics all have valid points to share. “We need to try this to advance the state of the art… And at the same time, we also need to recognize that we need to try it in areas that don’t have such a negative consequence,” Urso said. He emphasizes that even skeptical teams can begin by implementing AI in non-critical areas: “Try it in some non critical areas, and see how it does, and continue to train up those models and get them better.” This measured approach doesn’t impede innovation — early adopters can “try to use this more comprehensively,” pushing boundaries while maintaining safety protocols. Those ready to move faster can “move more aggressively,” while others can implement AI with appropriate safeguards, creating a balanced adoption pathway that accommodates varying risk tolerances and operational requirements.

Similarly, the WEF research reinforces this point, highlighting how effective change management and leadership from both executives and middle managers are crucial for successful implementation. The key lies in demonstrating concrete value through measurable outcomes while maintaining operational safety standards.

Ultimately, trust builds through demonstrated value and transparent reasoning. Urso illustrates this with a practical example: “Would you like to see the service records that I identified that were similar to yours?” This approach allows technicians to verify AI recommendations against historical data. “What if anyone in my organization had access to the entire knowledge base of all my service records that have been carried out across all sites for the last 25 years?” Urso asks. “We’re finally unleashing the knowledge that always existed within our enterprise, but it was just never evenly distributed.”

Early adopters are seeing this value materialize. The WEF reports that tasks that once took weeks now take minutes, and employees are freed to do more creative, meaningful work. However, the human element remains central—as one WEF interviewee put it, “The biggest mistake you can make is to remove humans from your processes.”