AI agents turn industrial AI into trusted industrial action

Applied effectively, AI in manufacturing can help engineers, planners and operators evaluate more information, explore alternatives faster, reuse expertise and automate routine work. The opportunity extends beyond productivity. It can improve decision speed, increase responsiveness to change and help organizations make better use of the knowledge they have accumulated over years of industrial experience. 

As industrial operations become more complex, this ability becomes increasingly valuable. Products are more configurable, production networks more distributed and supply chains more dynamic. Critical knowledge is often spread across experienced employees, enterprise systems, documents, spreadsheets and plant-specific processes.

This is where agentic AI in manufacturing becomes particularly important. Agentic AI can understand intent, reason across context and coordinate multi-step workflows across applications and processes. Rather than simply returning information, an agent can help determine what needs to happen next and initiate or coordinate actions within a governed workflow. 

Generative AI is an important related technology. It is particularly suited to creating and adapting content, processes and proposed outputs. The two can work together. An AI agent may call on generative AI when something needs to be created, just as it may draw on simulation, optimization or predictive intelligence when a task requires those capabilities. The distinction is important: generative AI primarily creates, while agentic AI coordinates capabilities around an objective.

The opportunity is therefore not simply to give people more data or more AI-generated content. It is to bring different forms of intelligence together so people can understand complexity, explore more possibilities and act on trusted information with greater confidence.

Why does general generative AI fail in critical manufacturing environments?

By general-purpose generative AI, we mean broadly trained tools such as ChatGPT or Microsoft Copilot that can answer questions, summarize information and generate content across many topics. They have demonstrated significant value for everyday productivity, but moving from experimentation to dependable operational use has proved more difficult. Gartner reported that at least half of GenAI projects had been abandoned after proof of concept by the end of 2025, citing poor data quality, inadequate risk controls, escalating costs and unclear business value. That helps explain why general-purpose models alone are not enough for industrial environments, where AI must operate within specific products, processes, constraints and operating conditions.

In manufacturing, an answer that merely sounds plausible is not enough. Decisions may affect product quality, worker safety, compliance, machine utilization, material availability, production schedules, cost and customer commitments.

That means industrial AI must be grounded in more than text. It needs access to governed information, validated engineering knowledge, operational constraints and models that represent how industrial systems actually behave. Generative AI still plays an important role, but its outputs become far more valuable when they are based on trusted industrial context and connected to the systems in which work actually takes place.

Generative AI can create new content, proposed processes, instructions or alternatives. Agentic AI can interpret an objective, determine which capabilities are required and coordinate a workflow that may include generative AI alongside predictive models, optimisation, simulation and other technologies.

Human supervision remains essential. AI may analyse, propose, generate or coordinate actions, but people need visibility into the information being used, the assumptions being made and the potential consequences before critical decisions are approved.

How can manufacturers use agentic AI to mitigate the loss of tribal knowledge?

One of the most immediate industrial opportunities for agentic AI involves knowledge. Many manufacturing processes still depend heavily on experienced individuals who understand why certain decisions were made, which exceptions matter and how formal processes need to be interpreted in practice. That creates a growing challenge as experienced workers retire; organisations struggle to recruit specialised skills and newer employees need to become productive more quickly. AI agents can help make enterprise knowledge easier to access in the context of the work being performed.

Instead of searching through documents or trying to find the person who remembers a particular decision, an engineer might ask why a manufacturing process was configured a certain way, what standards apply or what previous decisions are relevant to the current situation.An agent can connect that request to available enterprise knowledge, operational context and historical information, helping surface relevant guidance directly within the workflow.

The important point is that AI does not replace industrial expertise. It helps organizations make that expertise easier to discover, apply and reuse. Knowledge that was once difficult to find or concentrated in a few individuals can become more accessible across people, programs and locations.

Agentic approaches can therefore help organizations move beyond using AI simply to retrieve information. They provide the potential to put accumulated knowledge to work by combining context, reasoning and actions within industrial workflows. AI-Powered Virtual Companions and Industrial Knowledge

Shifting from disconnected data to integrated industrial world models

Industrial AI becomes significantly more valuable when it understands the environment in which decisions are being made. General AI systems may begin with documents, databases or individual data sources. Industrial operations require a richer representation that includes products, production systems, supply chains, resources, processes, constraints and real-world operating conditions. This is where industrial world models become important.

Rather than treating information as isolated data points, a world model represents relationships between the different elements of an industrial system. It provides context for understanding not only what something is, but how it relates to everything around it.

Virtual models of products, production environments and supply chains can provide an important part of that foundation. Connected to operational information, they allow organisations to experiment, simulate behaviour and evaluate the potential consequences of decisions before implementing them in the real world. Agentic AI provides a practical way for people to interact with this richer industrial context. For example, a planner responding to a production disruption might begin with a simple question. Predictive intelligence could identify the likely impact, optimisation could evaluate alternative schedules and simulation could help assess the consequences. The agent can coordinate the workflow and bring the relevant information back to the user. The agent does not replace those capabilities. It helps bring them together around the user's objective.

As industrial AI develops further, this relationship between AI agents, structured industrial knowledge and richer models of the physical world could become increasingly important. Industry world models offer a way to ground AI in relationships, constraints and behavior rather than relying on disconnected information alone.

What are the key performance indicators (KPIs) and ROI of industrial AI?

The value of industrial AI ultimately comes from improving how quickly and effectively organizations can turn information into decisions and decisions into action.
At an operational level, that can mean less manual work, faster response to disruption, better use of scarce expertise, improved quality, more consistent execution and stronger planning and resource utilisation.

The benefits can be significant. Representative examples from manufacturing engineering and machining include:

• Up to 50% improvement in manufacturing efficiency
• Up to 25% reduction in industrialisation lead time
• 90% to 95% reduction in MBOM creation and update effort
• Up to 50% reduction in machine setup time.

These examples illustrate the potential at the task and process level, but the greater value comes from what those improvements enable across the business.
Faster engineering and planning cycles can help companies bring products to market sooner. More reliable decisions can improve delivery performance and customer responsiveness. Better access to expertise can reduce dependence on scarce skills.

At the corporate level, the cumulative effect is the ability to operate with greater speed, resilience and flexibility. That can strengthen competitiveness by helping companies innovate faster, respond to customers more effectively, protect margins and make better use of their existing knowledge and resources.
The value of industrial AI is therefore not simply measured in minutes saved on individual tasks. It is measured in how effectively the organization can convert knowledge into better business outcomes.

Shifting from simple AI assistance to trusted industrial transformation

Agentic AI represents an important next step in the evolution of industrial AI. Rather than simply answering questions or generating content, AI agents can help users understand intent, navigate complexity and coordinate actions across industrial workflows. Their potential comes from being able to bring together the right information, applications and specialised capabilities around a specific objective. That may include generative AI to create or adapt content, predictive AI to anticipate risks, optimisation to evaluate complex trade-offs and simulation to test potential outcomes.

In this sense, agentic AI can become the connective layer between people and an increasingly sophisticated industrial technology environment. But the value of that capability depends on trust. Agents need to operate within governed workflows, draw on reliable industrial context and provide people with appropriate visibility and control over important decisions.

The goal is not autonomous AI for its own sake. It is to make industrial knowledge and advanced capabilities easier to access, coordinate and apply. When that happens, agentic AI can help move industrial organisations beyond isolated AI assistance toward something more significant: a trusted way to turn knowledge into decisions, decisions into action and industrial intelligence into measurable business value.

Adrian Wood is director of strategy and marketing at DELMIA

www.3ds.com/products/delmia