AI Agents Need More Than Access to Engineering Data

The conversation is shifting

Much of the discussion around AI agents still centers on the capabilities of the models themselves. Are they accurate enough? Can they reason effectively? Will they eventually work with minimal human oversight?

Those are important questions, but they are becoming less important than another one: what information are those agents working with?

In engineering, AI agents rarely operate in isolation. They are expected to retrieve requirements, interpret product structures, compare revisions, summarize engineering changes, recommend actions, and increasingly coordinate work across PLM, simulation, manufacturing, quality, and service systems. Their value depends on more than the sophistication of the underlying model. It depends on whether they can understand the engineering knowledge behind the data they retrieve.

That distinction matters.

Engineering information carries meaning that extends well beyond the document or record itself. A requirement may still be under review. A design may exist in multiple configurations. A manufacturing process may apply only to a specific product variant. Two revisions of the same part can represent very different stages of development.

Without that context, an AI agent can produce an answer that appears convincing while overlooking the details that determine whether the answer is actually correct.

As organizations begin deploying AI agents in engineering environments, the conversation is shifting from model performance to knowledge quality. Increasingly, the question is not whether an agent can retrieve information. It is whether the organization has created an environment in which that information can be interpreted correctly.

Engineering knowledge is more than data

Most organizations already manage enormous amounts of engineering information. CAD models, requirements, simulation results, bills of material, software artifacts, manufacturing plans, quality records, supplier documentation, and service data all contribute to the complete product definition.

The challenge is not simply bringing those sources together.

Each piece of information has relationships that give it meaning. Engineers need to know which requirement drove a design decision, which revision was approved for production, which test validated a change, and which manufacturing process applies to a specific configuration. Those relationships allow teams to understand what changed, why it changed, and what the downstream impact will be.

This becomes even more important for AI.

Consider a simple engineering change. An agent may quickly locate the latest CAD model and identify the associated requirement. On the surface, everything appears complete. But what if the requirement has not yet been approved? What if manufacturing is still building from the previous revision? What if a supplier qualification is still pending, or a related software update has not been released?

The agent has not failed because it misunderstood the language. It has failed because it lacked the engineering context needed to evaluate the information it found.

This is one of the reasons AI governance has become such an important topic in enterprise discussions. Recent research from McKinsey emphasizes that organizations scaling agentic AI need more than high-quality data. They need consistent definitions, clear ownership, controlled access, and the ability to trace how information is created and used across the enterprise. Gartner has reached a similar conclusion, warning that many organizations will struggle to operationalize autonomous agents because governance weaknesses often become visible only after deployment.

Engineering organizations face those challenges earlier than most because their information is highly interconnected and continuously changing.

Governance gives engineering knowledge meaning

Governance is often associated with security policies or regulatory compliance. Those responsibilities remain important, but they represent only part of the picture.

For engineering organizations, governance also establishes the context that allows information to be trusted.

It defines which systems are authoritative, who can modify information, how revisions are managed, what lifecycle state an object is in, and how decisions remain connected as products evolve. It preserves the relationships between requirements, designs, manufacturing processes, quality records, software, and service information so that changes can be understood throughout the product lifecycle.

Without that structure, AI agents inherit the same uncertainty that engineers already struggle with when information becomes fragmented across disconnected systems.

This is why simply giving an AI agent access to more data rarely improves the outcome. Access alone does not explain whether information is current, approved, complete, or applicable to a particular product configuration. Those distinctions come from governance.

As AI agents move beyond answering questions and begin recommending actions or automating portions of engineering workflows, those distinctions become increasingly important. An agent that retrieves incomplete or outdated information may still produce a fluent response, but confidence in the answer is not the same as confidence in the engineering decision.

Why the digital thread becomes the foundation

This is where the digital thread becomes more than an integration strategy.

For years, manufacturers have invested in connecting engineering systems so information can move more easily between teams. Those efforts remain important, but AI agents raise the expectations. They do not just retrieve information from multiple systems. They are expected to understand how that information relates to the product and use it to support decisions.

That requires more than connectivity.

A digital thread preserves the relationships between requirements, design, simulation, manufacturing, quality, software, and service throughout the product lifecycle. Those relationships provide the context that tells an AI agent how information fits together, whether it can be trusted, and what impact a decision might have downstream.

Imagine an agent evaluating whether an engineering change is ready for manufacturing.

Finding the latest CAD model is only part of the task. The agent also needs to determine whether the associated requirements have been approved, whether verification activities are complete, whether the manufacturing process plan has been updated, whether supplier changes have been incorporated, and whether quality documentation reflects the current revision.

None of those questions can be answered by looking at a single object.

The value comes from understanding the relationships between them.

This is why connected engineering knowledge is becoming just as important as connected systems. Information that has lost its relationships, lifecycle status, or traceability may still be accessible, but it is no longer complete enough to support reliable decisions.

Organizations often describe AI as requiring more data. In practice, engineering organizations need something more specific. They need connected, governed knowledge that preserves the context surrounding every engineering decision.

AI reflects the quality of the engineering environment

There is a common assumption that better AI models will eventually overcome poor information.

That is unlikely.

Large language models continue to improve, but they can only reason over the information available to them. If engineering knowledge is inconsistent, incomplete, or disconnected, the model has no reliable way to reconstruct the missing context. It can identify patterns, but it cannot determine which revision is authoritative, whether a change has been approved, or which configuration applies to a particular customer, unless that information is clearly represented.

In many ways, AI acts as a mirror.

Organizations with well-governed engineering knowledge often see agents deliver useful results more quickly because the underlying information is already structured and connected. Organizations with fragmented information typically discover problems they have lived with for years. Duplicate records, inconsistent definitions, missing traceability, disconnected change processes, and unclear ownership become visible once an AI agent begins working across multiple systems.

That is one reason governance has moved to the center of conversations about enterprise AI. As agents take on more responsibility, organizations need confidence not only in the model but also in the knowledge that supports its recommendations.

For engineering leaders, this shifts the conversation from “Can we deploy AI agents?” to “Are our engineering systems ready for AI agents?”

Those are very different questions.

Why agent governance starts with knowledge governance

As AI agents become more capable, organizations are paying increasing attention to governance. Much of that discussion focuses on controlling what agents are allowed to do. Should they retrieve information? Recommend actions? Trigger workflows? Modify data?

Those are important questions, but they will only provide part of the answer.

Before organizations define what an agent is allowed to do, they must establish what the agent is allowed to know.

An AI agent should be able to distinguish between working information and released information. It should recognize that a requirement awaiting approval does not carry the same authority as one that has been baselined. It should understand that an engineering change may affect multiple downstream processes, even if those impacts are stored in different systems.

Those capabilities do not come from the model itself. They come from the way engineering knowledge is governed.

This is one reason digital thread initiatives are becoming increasingly important in the era of AI. A connected lifecycle provides more than visibility across engineering disciplines. It gives AI agents access to the relationships that explain how information should be interpreted and how decisions ripple across the product lifecycle.

Without those relationships, an agent may retrieve the right information but apply it in the wrong context.

Preparing engineering organizations for agentic AI

Most manufacturers are still in the early stages of adopting AI agents. Many are experimenting with document summarization, engineering assistants, or intelligent search before expanding into recommendation engines or more autonomous workflows.

That progression is both practical and sensible.

As organizations increase the responsibility they give AI agents, confidence in the underlying engineering knowledge becomes increasingly important. Before asking whether an agent can automate a process, leaders should first consider whether the engineering information supporting that process is ready.

A few questions can help frame that discussion:

• Is there a clear source of authority for engineering information?
• Can requirements, designs, manufacturing plans, software, and quality records be traced throughout the product lifecycle?
• Are lifecycle states, revisions, approvals, and configurations consistently represented across systems?
• Can an AI agent distinguish between information it should observe, information it can recommend against, and actions it is authorized to perform?

These are governance questions, but they are also readiness questions.

Organizations that address them early will find it much easier to expand AI’s role over time. Those that postpone them may still achieve impressive demonstrations, but scaling AI across engineering will become far more difficult when every new use case exposes gaps in data quality, ownership, or lifecycle context.

AI agents do not replace disciplined engineering practices. They make those practices more valuable.

Engineering knowledge is the real advantage

The discussion around AI will continue to evolve. Models will become more capable. Agents will take on increasingly sophisticated work. Organizations will find new ways to automate repetitive tasks and support engineering decisions.

What is less likely to change is the importance of engineering knowledge itself.

The organizations that gain the greatest value from AI will not necessarily be those with access to the newest models. They will be the ones that have invested in trusted engineering information, connected lifecycle processes, and governance that preserves context across the digital thread.

That is ultimately what allows AI agents to move beyond demonstrations and become reliable participants in engineering work.

The next stage of AI adoption will be defined by environments in which those agents can understand the information they use, explain how they reach a recommendation, and operate within the same lifecycle context that engineers depend on every day.

For engineering leaders, that makes the priority clear. Before asking how autonomous an AI agent should become, ask whether the engineering knowledge behind it is complete, connected, and governed.

When that foundation is in place, AI becomes much more than another productivity tool. It becomes a practical extension of the engineering organization itself.