- The early adopters: Industrial companies including Borouge, Aramco and Eneos Materials are exploring or deploying different forms of AI-driven industrial operations, with varying degrees of autonomy.
- The suppliers: Schneider Electric, Siemens, Honeywell, ABB and other automation specialists are developing the software and control systems that could underpin the next generation of intelligent factories.
- The missing infrastructure: Nokia, Ericsson, Cisco and Huawei are helping build the industrial networks, edge computing and operational platforms required to connect these systems. Fully autonomous factories, however, remain some distance away.
Industrial artificial intelligence is beginning to find its place on the factory floor. But before declaring the arrival of the autonomous robot factory, it is worth establishing precisely what the machines are doing and how much human supervision they still require.
There are at least three distinct stages of industrial AI development. The first involves systems already operating in production, using machine learning, predictive analytics and optimization to improve established industrial processes. So far, so ho-hum.
The second introduces AI agents capable of interpreting operational information, recommending actions and performing defined tasks under human supervision. Now we're getting somewhere.
The third, and most ambitious, involves autonomous systems capable of making and executing physical control decisions within established safety limits (hopefully). This is the future, but it's not quite here on anything approaching a universal basis.
The distinctions are more than academic. A software agent recommending changes to a production schedule is not doing the same job as an AI system controlling the temperature of a blast furnace. Both may improve productivity, but the consequences of a mistake are rather different. One is fancy spreadsheet calculus. The other is either part of the new industrial revolution or people running screaming from a giant fireball if you get it wrong.
Schneider lays the groundwork
FNTV recently reported on Schneider Electric collaborating with Hewlett Packard Enterprise and Nvidia to introduce industrial AI agents at the edge.
The companies are targeting operational technology monitoring, cybersecurity and energy optimization, with agents operating close to the industrial systems they are intended to support.
Their announced collaboration identifies Chilean pulp and paper producer CMPC as an example of the operational foundation required for these systems.
CMPC has used Schneider's Operational Technology Monitoring System in production for approximately three years, monitoring the health and availability of control systems across its manufacturing operations.
That is an established production deployment, but not evidence that the newly announced industrial AI agents are already operating autonomously at CMPC. Rather, it demonstrates the importance of reliable operational data and monitoring infrastructure before more advanced agents can be introduced.
Who is actually deploying industrial AI?
“Beyond CMPC Chile, we are seeing significant movement across other major industrial players,” Wayne M. Butner, president and CEO of Altan Technologies, a software engineering services firm, told FNTV.
“Enterprises such as Borouge, Aramco, ENEOS Materials and even Siemens (within their own manufacturing facilities) have started laying the groundwork for similar physical AI control layers that prioritize strict cybersecurity and operational data sovereignty,” he wrote.
Published customer examples support the broader direction of his argument, although the deployments are at substantially different stages.
At Borouge, the Abu Dhabi petrochemicals producer, Honeywell has been developing AI-assisted autonomous control-room technology. In January 2026, Borouge announced completion of a proof of concept at its Ruwais facility, identifying potential improvements in efficiency, reliability and operating costs.
Honeywell subsequently introduced its Experion Cognition platform, designed to support AI-assisted decisions and more autonomous industrial operations. The results are encouraging, but the companies are still working toward wider deployment rather than claiming a fully autonomous petrochemical complex.
Aramco provides another example. The Saudi company describes industrial AI agents deployed across selected operations to support energy optimization, operational decisions and defined workflows. These applications move beyond passive analytics, but they do not establish plant-wide autonomous control.
Japan's Eneos Materials offers an especially interesting contrast. In March 2023, the company and Yokogawa Electric announced the commercial adoption of reinforcement-learning-based autonomous control at a chemical plant, following an extended trial controlling distillation operations.
This is a documented example of AI exercising direct physical process control, rather than merely recommending an action to a human operator.
But it should not be confused with a general-purpose or multipurpose AI agent. The system was designed for a defined industrial control problem, operating within a specific process environment. Eneos Materials has separately developed an AI agent for materials research, which entered internal verification in October 2025.
The lesson is that meaningful industrial autonomy can already exist without the sort of conversational agents currently attracting so much attention.
Industrial automation vendors join the race
The established automation suppliers are also moving rapidly.
“Other major industrial automation and control incumbents actively building similar physical AI layers include Siemens, ABB, Rockwell Automation, Yokogawa ... and Emerson (alongside AspenTech),” Butner added.
Siemens provides a useful example of the difference between AI-assisted engineering and autonomous manufacturing.
Its Eigen Engineering Agent has been piloted with more than 100 companies across 19 countries, including Andritz, CASMT and Prism Systems.
The agent assists engineers with programmable logic controller coding, device configuration, human-machine interface design and other automation tasks. Siemens has published customer examples describing accelerated engineering workflows.
These are substantive industrial applications, but they primarily automate engineering work rather than independently operate production machinery.
Other major suppliers pursuing industrial automation and AI include ABB, Rockwell Automation and Emerson.
The competitive question is whether these incumbents can extend their existing control-system expertise into increasingly intelligent, agent-driven operations without compromising the reliability on which industrial customers depend.
Agents are already producing measurable results
Ben Weaver, an analyst at ABI Research, told FNTV that industrial simulation agents are further along than broadly autonomous physical systems. These agents operate within digital twins and other virtual environments, allowing manufacturers to evaluate decisions before applying them to physical processes.
Manufacturers including BMW and Tata Steel have already adopted simulation-backed digital manufacturing technologies.
There are also increasingly substantial examples of AI agents operating in production-related workflows.
GE Appliances, owned by Haier Smart Home, announced in April 2026 that it had deployed more than 800 AI agents across manufacturing, logistics and supply-chain operations using Google Cloud, part of Alphabet.
These agents support activities including production analysis, quality improvement and supplier communications. Their scale is notable, but their deployment should not be interpreted as 800 autonomous machines controlling physical production.
Another example comes from Sight Machine, working with Microsoft.
In a published customer case study, the companies describe an AI-assisted production scheduling system that helped a beverage manufacturer reduce non-value-added production time by 75% and increase production capacity by more than 5%.
The system combines manufacturing data with optimization models to generate updated schedules when conditions change. It is a practical application of AI to industrial productivity, without requiring the AI to assume direct control of the production machinery.
“Many live production agents are less involved in machining and live closer to production planning and quality,” Weaver noted.
Where the network companies fit
There is another important group of suppliers in this developing market: the telecommunications and networking companies responsible for connecting industrial machines, moving operational data and providing the computing infrastructure on which increasingly sophisticated AI systems depend.
Nokia is positioning private wireless networks, industrial edge computing and AI as components of more flexible manufacturing systems. At MWC Barcelona 2026, Nokia and Verizon demonstrated how private 5G and edge AI could support production lines that adapt dynamically to different product configurations. That was a demonstration of an industrial architecture, not proof of an autonomous factory operating at commercial scale.
Ericsson has more established industrial connectivity deployments. At ArcelorMittal facilities in France, private 5G supports connected workers, industrial monitoring and AI-assisted vision applications. Ericsson's case study also describes testing of autonomous vehicles. Separately, Ericsson and NTT Data have announced a partnership to combine managed private 5G with edge and physical AI.
But FNTV has also documented an example of industrial autonomy already operating at enormous scale.
At Newmont's Boddington gold mine in Western Australia, our documentary series, "The Autonomous Mine," followed the deployment of Ericsson private cellular technology supporting fleets of autonomous haul trucks and drilling machines.
These are not laboratory demonstrations or simulated production environments. They are enormous industrial machines operating in a working mine, supervised from centralized control rooms and supported by an industrial network designed to remain reliable as the physical environment changes.
The mine had previously relied on Wi-Fi, but the movement of equipment, changing terrain and demands of autonomous operations exposed limitations in coverage, handovers and reliability. Private 4G and 5G provided a more suitable communications foundation.
The implications extend beyond productivity. Autonomous operations can remove workers from hazardous areas while creating new technical and supervisory roles. They also illustrate why autonomy is not simply a matter of adding an AI agent to an existing machine. It requires sensors, control systems, communications, operational procedures and human oversight to work together.
The FNTV documentary offers a useful counterpoint to predictions that industrial autonomy is still entirely a future prospect. In mining, certain forms of it are already here.
Cisco approaches the opportunity through industrial networking, cybersecurity, edge infrastructure and machine-vision integration. A published Cisco and Invisible AI case study describes how an automotive supplier used edge-based visual intelligence to identify production bottlenecks and improve throughput. The example demonstrates AI-assisted industrial optimization, rather than autonomous machine control.
Huawei, which is privately held, offers a particularly interesting example of AI moving deeper into industrial operations. Working with Nanjing Iron & Steel (NISCO), Huawei has supported deployment of the YuanYe industrial AI system across ironmaking, rolling, quality inspection and energy and environmental control.
The companies report more than 300 AI models and 25 business agents, combining probabilistic AI with mathematical and physical process models. They claim annual cost reductions of more than $15 million (CNY100 million).
FNTV has previously examined the NISCO deployment, which illustrates how AI can be integrated into established industrial control environments rather than simply layered onto them as a conversational interface.
FNTV has also documented Huawei's private 5G deployment at the East-West Gate intermodal terminal in Hungary, where automated cranes and industrial logistics systems operate across a vast railway facility near the Ukrainian border. It is another example of industrial digital infrastructure being deployed in a real operational environment, although connectivity and automation should not be confused with AI-directed autonomous control.
These examples demonstrate the role network suppliers can play in connecting sensors, controllers, robots, edge computing and industrial AI applications. They also highlight an important boundary: providing the infrastructure for autonomous operations is not the same as supplying the autonomous control system itself.
As AI moves closer to machinery, network reliability, latency, cybersecurity and operational sovereignty become increasingly important. The question is not simply whether an agent can make a decision, but whether the entire industrial system can execute that decision safely, predictably and under the control of its operator.
How far away is full autonomy?
For all the progress, the transition to broadly autonomous industrial operations remains gradual. But it would be a mistake to assume that industrial autonomy itself is still waiting to be invented.
The Newmont Boddington operation provides an obvious example of machines already performing complex physical tasks autonomously, with humans supervising the operation. Eneos Materials provides another, rather different example, involving reinforcement-learning-based control of a defined chemical process.
What remains much less mature is the prospect of general-purpose AI agents coordinating and controlling multiple industrial processes, potentially across entire factories, with minimal human intervention.
Weaver expects initial deployments of these more broadly autonomous industrial agents to remain tightly constrained.
“In terms of timeline toward full autonomy, I would expect customer zero announcements within the next 2 years that are limited in scope (single machines in noncritical lines to start),” he told FNTV.
He believes wider external adoption will take substantially longer.
“Predicated on the success and expansion of initial internal pilots, external customers will begin cautious testing within the next 7 years,” Weaver concluded.
Those forecasts concern the emergence of more broadly autonomous industrial agents, not the first use of AI for physical control.
The more difficult challenge is extending that capability across multiple machines, processes and operational domains while maintaining the deterministic behavior, safety assurances and accountability that industrial production requires.
For now, industrial AI is advancing through a combination of proven machine learning, increasingly capable software agents and carefully bounded autonomous control.
The factory of the future is beginning to take shape. But it will probably arrive one production line, one control system and one verified decision at a time.
Related Fierce Network and FNTV coverage
FNTV documentaries and industrial case studies
- The Autonomous Mine — How Ericsson private 5G is reshaping autonomous mining in Australia — Stephen M. Saunders MBE. Autonomous haul trucks, industrial connectivity and machine operations at Newmont's Boddington mine in Western Australia.
- Ericsson private 5G powers safer, smarter mining in Western Australia — FNTV documentary coverage of the communications infrastructure supporting industrial autonomy.
- Huawei and NISCO put AI to work in the steelworks — Industrial AI models and intelligent agents deployed across steelmaking, quality inspection and process optimization.
- Huawei powers Industry 4.0 in EV manufacturing — Stephen M. Saunders MBE. The integration of robotics, AI and 5G in Chinese electric-vehicle manufacturing.
- Watch Huawei's private 5G solution in action — Stephen M. Saunders MBE. Industrial automation and logistics at Hungary's East-West Gate intermodal terminal.
Further analysis by Stephen M. Saunders MBE
- What a gold mine in Australia teaches us about the real future of AI — June 23, 2026. Why physical AI and machine productivity may prove more consequential than conversational AI.
- Ericsson's AI bet: The intelligent fabric vs. the AI factory — Feb. 17, 2026. The economic and architectural distinction between centralized AI factories and distributed, deterministic network intelligence.
- Sweden to the world: The revolution will be digitized — July 9, 2024. Ericsson, Telia and Sweden's NorthStar initiative as examples of industrial cyber-physical integration.
- China is quietly crushing America in the race to industrial digitalization — Sept. 26, 2025. The geopolitical and economic consequences of differing national approaches to industrial modernization.
- 5G's true potential: Why analysts are missing the mark on industrial digitalization — April 7, 2025. Why consumer-centric telecom forecasts miss much of 5G's potential contribution to the industrial economy.
- The AI bubble won't be the end of the world — Dec. 10, 2025. The distinction between speculative AI valuations and the underlying transformation of the physical economy.
Related reporting by Dan Jones
- Schneider teams with HPE and Nvidia on industrial AI agents at the edge — The original report examining Schneider's industrial AI collaboration, early applications and the emerging market for industrial AI agents.
Stephen M. Saunders MBE is a communications analyst and USPTO-registered inventor examining how digital infrastructure — 5G, cloud and AI — is reshaping industry, power and society, as well as underpinning the emerging, ubiquitous global digital economy. As anchor of FNTV and a longtime industry insider, he focuses less on growth narratives and more on execution, risk and how hyperscale technology is distorting markets, governance and society at scale.
Opinion pieces from industry experts, analysts or our editorial staff do not represent the opinions of Fierce Network.