AI agents are quickly becoming one of the most talked-about innovations in industry. From analyzing bills of materials and checking compliance requirements to optimizing inventory levels, they promise major productivity gains. But does that mean companies should trust them without question? In this op-ed, Laurent Germain, Product Development Specialist and AI Expert at PTC answers: Absolutely not!
The biggest risk does not lie in the intelligence or autonomy of the agents themselves. It lies in the quality of the data they depend on.
For AI agents to deliver on their promise, companies first need solid foundations. That starts with the business systems that support day-to-day operations: PLM, ERP, inventory management, CRM, and others. Each system typically has its own data model, terminology, and governance rules. As a result, the same information can be represented differently across different platforms.
Many organizations are tempted to rely on AI to reconcile these inconsistencies automatically. That is a mistake. When data lacks context or a clear definition, an AI agent can only produce unreliable conclusions. And if one agent can make a wrong decision, imagine what happens when several agents are working from the same flawed information. The takeaway is clear: before deploying AI agents, companies must ensure that their data is properly qualified, contextualized, and connected across systems.
Another challenge arises when organizations deploy multiple specialized agents, each focused on its own domain, but without any overarching coordination. Taken individually, each agent may perform its task effectively. Yet when several agents operate simultaneously across interconnected processes, inconsistencies and conflicting outcomes become much more likely.
Consider a product change:
- One agent evaluates the technical impact.
- Another checks regulatory requirements.
- A third assesses the impact on costs and inventory.
Without coordination, each agent optimizes its own objective without considering the overall impact. Fifty agents making mistakes together do not create an intelligent system; they create an industrial risk.
Bringing AI Agents Under Control
A specialized AI agent represents only the first stage of maturity. The next step is enabling multiple agents to work together toward a shared objective within a tightly governed environment.
That evolution relies on two fundamental concepts:
- Planning, which breaks a project into clearly defined steps.
- Orchestration, which makes it possible to call upon the right agents, in the right systems, at the right time while consolidating their outputs into a coherent result.
The MCP protocol helps make this possible by providing a common framework that simplifies interoperability between AI agents and enterprise systems.
Through a conversational interface, an engineer could ask whether a change to an electronic board complies with a new regulation and what its downstream consequences might be. One agent analyzes the modification within the PLM system, another checks the relevant regulatory requirements, while a third evaluates the impact on costs and inventory in the ERP system. The engineer receives a single answer, but behind that answer lies coordinated collaboration across multiple agents and multiple business systems. That is exactly what orchestration means: planning actions, engaging the right agents in the right sequence, and consolidating their findings to support better decisions.
As an open standard, MCP has the potential to become a key enabler of AI adoption in industry, making it easier for AI agents to interact with industrial systems without locking organizations into a single technology stack, vendor, or operating model.
However, effective AI orchestration inevitably requires a more open ecosystem. And openness only delivers value when supported by the right safeguards:
- Data governance and access control: users should only have access to information relevant to their responsibilities and role.
- Traceability: organizations must be able to determine which decision was approved, by whom, and when.
- Infrastructure control: whether environments are cloud-based, on-premises, or hybrid.
- Human oversight: AI should never be allowed to launch a manufacturing order or make large-scale file changes without supervision. In industrial environments even when assisted by AI, accountability ultimately remains with people.
So, should companies fear specialized AI agents? No.
What they should fear are poorly governed agents operating on poorly managed data and working without coordination. AI does not fix organizational disorder. It amplifies it.
Before asking how many agents to deploy, manufacturers should first ask what data those agents will rely on. The organizations that gain the most value from AI will not necessarily be the ones with the most advanced agents. They will be the ones that built the strongest foundations.
