AI is making steady inroads into manufacturing systems like ERP, enterprise asset management and supply chain, where new technologies like agentic and generative AI are finding some of their most compelling enterprise use cases.
It's no secret that there's a heap of hype around enterprise AI, but it's impossible to deny that the enterprise has already entered an AI age and will not go back to a pre-AI state.
The questions that organizations need to ask are how they can safely implement AI, how quickly can they get a return on the investment and, crucially, where can they derive real organizational value.
For many, that answer is in the systems and processes that are integral to manufacturing, like design and production, supply chain, logistics and fulfillment.
ERP moves to system of action
AI, particularly agentic AI, enables ERP and related applications to evolve from back-office systems of record to real systems of action, according to Vaibhav Vohra, chief product and technology officer at Epicor, a manufacturing-focused ERP provider. However, this evolution is not without significant growing pains.
"There's millions of transactions created every day, which means ERP is a fantastic opportunity to generate AI [productivity]," Vohra said. "Most manufacturers, distributors and retailers are experimenting with AI, but unless they're coupled with value statements, it sometimes feels like a science project."
Most manufacturers, distributors and retailers are experimenting with AI, but unless they're coupled with value statements, it sometimes feels like a science project.
Vaibhav VohraChief product and technology officer, Epicor
This science project phase can lead to negative outcomes, as Gartner reports that 40% of enterprise agentic AI initiatives will be cancelled by the end of 2027. Additionally, S&P Global Market Intelligence reports that 42% of organizations have abandoned AI initiatives this year, according to a survey of 1000 organizations in North America and Europe -- an increase of 17% from last year.
This is not putting the brakes on enterprise investments in AI, however. MarketsandMarkets estimates that the market for AI agents is pegged at $7.84 billion in 2025 and will grow to $52.62 billion by 2030. Garter reports that 33% of enterprise software applications will include agentic AI by 2028, up from less than 1% in 2024, and that 15% of day-to-day work decisions to be made autonomously by then.
Therefore, it's inevitable that AI in generative and agentic forms will make its way into manufacturing and supply chains, Vohra said. In turn, manufacturing processes will prove to be the areas that gain ROI the fastest.
One reason is that manufacturers and distributors are facing a serious labor shortage and new employees can't be expected to learn one – or in many cases – several different ERP and associated systems. This is requiring the enterprise vendors to step up efforts to make the systems more conversational via generative AI (genAI) prompts.
Vohra is bullish about the possibilities for agentic AI in ERP, even as he acknowledged that most ERP vendors are still in the very early stages of including true agentic AI.
The use cases are there, he said. For example, a distributor might experience delivery delays because of weather conditions and want to tell its manufacturing customers to make alternative supply chain decisions. This can be handled with AI agents if they are equipped with the context-aware system data and connections through what Vohra calls "cognitive ERP."
"The idea is that the agents that are working with distributors and manufacturers are coming together to make optimal supply chain decisions for the community," Vohra said. "That's an example of agentic AI that's worthy of ERP. You are no longer in this copilot mode of talking to the ERP, now you're starting to make decisions on the fly."
Digital labor enters the workforce
Ultimo, an IFS company, is an EAM software provider that's moving full speed ahead into the AI-centric world.
The combination of genAI and agentic AI is enabling ERP systems to move from being systems of record to systems of recommendation or action, said Steven Elsham, CEO at Ultimo.
Rather than just making insights, the systems are now able to take recommended actions to the employees where they are, Elsham said. This is becoming critical now, as experienced employees are leaving the workforce and new ones are entering.
"In the old days, even before we had computer systems, you had highly skilled people who made recommendations and talked to one another," he said. "Now we're kind of going back to that [with AI], but with a rock-solid foundation, that's informed by strong data that's able to drive the optimal insight and recommendations from that data."
Ultimo has added AI agents to its platform that are designed to help workers perform tasks more efficiently and safely, Elsham said. These provide Ultimo's customers with what it calls digital labor.
One recently released agent is tasked with alerting workers on potential environmental, health and safety (EHS) issues, by operating as a specialized EHS expert that's directly embedded in organizational workflows.
Traditional EHS management relies heavily on manual incident reporting, a process that's fraught with gaps and errors in reporting, making it difficult to implement effective preventive measures or safety awareness programs, according to Elsham.
The Ultimo AI agent continuously monitors work requests, automatically identifies EHS-related content and generates incident reports that can recommend mitigation measures, he said. The comprehensiveness and automation of the AI agent is the key to creating a safer work environment.
"We're enabling the health and safety manager to fully report all incidents without having to manually draw out all of the data and look through it and try and determine what may or may not have been a health and safety incident that occurred in past," he said.
Nevertheless, Elsham acknowledged that manufacturing has not traditionally been an area that's known for innovation, and agentic AI will need to prove its worth before it's widely accepted.
Workers are often eager to adopt emerging technologies that can improve efficiency, while those tasked with implementing and managing those technologies are concerned that they are embedded into the business safely, he said.
"At the top of the organization, there are people who say yes, but they subordinate the decision that a platform can be trusted to the people who are experts in their business," Elsham said. "They want to use something that is going to enable the organization to operate most efficiently and effectively, and gives the flexibility to be able to outperform their competitors."
Different AI forms for different jobs
Agentic and genAI can be useful in manufacturing processes, and the various AI forms have different strengths and different roles, said Simon Ellis, practice director at IDC.
GenAI, for example, is good ingesting large documents like repair manuals or procurement service-level agreements and can help organizations make sure that they meet their procurement contracts or order goods from approved vendors, Ellis said. Most organizations lack the manpower that can devote the time needed for the tedious work involved.
"But genAI is really good at things like factory maintenance through ingesting operating manuals," he said. "GenAI is not helpful in things like demand planning -- that's more for the more traditional prescriptive AI tools."
Agentic AI is more of a decision-making entity that essentially acts as a control tower across other forms of AI, Ellis said, taking in traditional metrics and making recommendations or performing actions.
"It remains to be seen the degree to which companies are going to want agents to make decisions or just consult with people and defer the decision to the person who then becomes the responsible or accountable part," he said.
Data quality is a major issue for AI-led decision making and is a reason why organizations may be reluctant to allow agents to make decisions without at least providing a mechanism for human review, Ellis said.
But this can also depend on the importance of the decision, he said.
If, for example, a truck breakdown delays a shipment, an AI agent can approve to move the load to another trucking company, as long as it's an approved vendor, Ellis said. However, if there's no other trucking company available or urgency calls for switching the load to air freight, this dramatically changes the costs.
"Now, maybe you don't want the agent making that call," he said. "You might want the agent making a recommendation, but have a person make the call because of the cost implication."
There's much hype around the implementation of AI in manufacturing, which may be holding back results, at least in these early stages, said Gaurav Malhotra partner at EY. However, AI differs from other hyped emerging technologies like IoT and blockchain.
"AI is more like semi-hype because it carries a lot more potential than some of the others," Malhotra said. "There's more relevance and reality to this technology."
Organizations will need to have better strategies to ensure that AI is more successful than technologies like blockchain, which were often implemented -- at least in pilot projects -- without the technical and organizational foundations to make it work, he said.
For example, blockchain requires an orchestrated network to take full advantage and has limited use as a technology within an enterprise, Malhotra said. But organizations can do a lot with AI, even if it's just incorporated within the enterprise.
"There are many use cases around logistics and distribution, inventory management, demand planning, EAM or predictive maintenance that leverages [AI with organizational data], versus having to execute based on information that is being provided," he said. "That's the potential."
Jim O'Donnell is a news director for Informa TechTarget who covers ERP and other enterprise applications.
