Human-Centered AI in Engineering: When the Copilot Is in the CAD Tool

The debate about AI in engineering spent most of the last decade in the future tense. Generative design was introduced, piloted, and admired in conference presentations. Simulation-driven design was promised as the future of product development. Knowledge-based engineering was held up as the mechanism that would finally let organizations capture and reuse what their best engineers knew. All of these trajectories were real, but none of them moved as fast as the hype. Then, between 2023 and 2025, the large language model revolution arrived in engineering tools — not as a separate application, but embedded inside the tools engineers already used every day. The copilot era in engineering is not coming. It is here, and PLM is not ready for it.

How We Got Here

Engineering software has incorporated AI-adjacent capabilities for years. Topology optimization in finite element analysis tools has automated structural material distribution since the 1990s. Siemens introduced knowledge fusion in NX in the early 2000s, allowing engineers to encode design rules that automated routine geometry modifications. PTC's Windchill has offered AI-based part search and classification for a decade. These capabilities were real but narrow — useful within a specific workflow context, invisible outside of it.

The shift that began in 2023 was qualitative, not incremental. Large language models with engineering-domain fine-tuning, combined with tool-native integration, enabled a different interaction mode: the engineer describes what they want in natural language, and the AI translates that intent into tool actions — creating geometry, running analysis, searching the PLM repository for similar historical designs, generating BOM drafts. The capabilities are imperfect and require significant human review. But they are embedded in the tools engineers already use, which means adoption friction is low and usage is organic.

The digital thread concept is deeply relevant here: AI copilots that have access to the full product thread — design history, simulation results, manufacturing data, field performance — are substantially more useful than those working with only the current design file. The quality of the PLM data infrastructure directly determines the quality of the AI copilot's context.

The Current Commercial Landscape

Autodesk AI has been integrated into Fusion 360 and Inventor as a contextual assistant — offering geometry suggestions, automating drawing annotation, and surfacing relevant design standards. Autodesk's Forma product applies AI to early-stage architectural and industrial facility design, optimizing for structural, thermal, and spatial performance constraints simultaneously.

Siemens NX AI assistant was released in commercial form in 2025. It integrates with Teamcenter, allowing engineers to query PLM data in natural language ("find all designs using this bearing type that have had field failures"), generate design alternatives from performance specifications, and automate routine detailing tasks. The Teamcenter integration is the distinguishing capability — NX AI is not just working on the current file, it has access to the full program history in PLM.

PTC Creo Copilot entered commercial availability in late 2025, offering natural language interaction with Creo Parametric for model modification, design reuse search across Windchill, and automated generation of GD&T annotations from design intent. PTC has positioned it explicitly as a Windchill-integrated capability — the copilot's design reuse suggestions are drawn from the Windchill repository, making PLM data quality a direct input to copilot output quality.

Beyond the major CAD vendors, specialized tools have emerged for specific AI-augmented engineering tasks: Ansys SimAI for rapid simulation surrogate modeling, Cognata for autonomous system virtual testing, and several startups building AI layers on top of open PLM platforms.

Use Cases and Business Impact

AI-Accelerated Design Reuse. An aerospace structures team developing a new bracket assembly used NX AI to search the Teamcenter repository for structurally similar assemblies from previous programs. The AI returned 14 candidate assemblies ranked by geometric similarity and material compatibility. The engineer reviewed the top three and adapted the highest-ranked design to the new requirements — a process that previously required a manual search of the Teamcenter archive (typically 3–5 hours for a complex geometry) and was often skipped under schedule pressure. Adapting an existing design reduced the new design's structural analysis time from 2 days to 4 hours, because the simulation model was partially inherited from the reference design. The PLM configuration management record for the new assembly includes explicit reference to the source design — audit trail for both design reuse decision and the AI's role in surfacing it.

AI-Assisted BOM Generation. A consumer electronics manufacturer piloted Creo Copilot for automated BOM draft generation from new assembly designs. The copilot, with access to the Windchill component library, proposes a complete BOM draft — including standard fasteners, connectors, and PCB references — from the assembly geometry. Engineers review and correct the draft, typically accepting 80–85% of proposed line items without modification. The time to generate an initial BOM draft fell from 4–6 hours to under 30 minutes. Critically, the organization established a data governance rule that AI-generated BOM drafts require engineer sign-off before any PLM release — the AI generates, the engineer reviews, Windchill records both.

Natural Language PLM Queries for Root Cause Investigation. A medical device manufacturer's quality team uses a Teamcenter AI assistant to query the PLM system in natural language during root cause investigations: "show me all engineering changes to part number X in the last 18 months, including the change descriptions and approvers." Previously this required a trained PLM administrator to construct the query. Now a quality engineer who understands the product but is not a PLM power user can retrieve complete change histories in under 5 minutes. The quality and compliance benefit is direct: faster root cause investigation with more complete historical context.

Barriers to Adoption

PLM data quality is the ceiling. AI copilots that query PLM for relevant historical designs, similar failure modes, or applicable standards are only as good as the structured data they can access. An organization with inconsistent part classification, incomplete change descriptions, and irregular simulation result storage gets generic or wrong AI suggestions. Investing in PLM data quality is now also an investment in AI copilot effectiveness — a business case connection that did not exist three years ago.

Regulatory uncertainty around AI-generated design outputs. In aerospace (DO-178C, AS9100), medical devices (21 CFR Part 820, ISO 13485), and automotive safety systems (ISO 26262), design decisions must be traceable to verifiable technical rationale. If an engineer approves an AI-generated design modification, they are accountable for that decision — but the regulatory frameworks have not yet specified how AI's role must be documented in the design history. The FDA has provided initial guidance on AI in software medical devices; aerospace and automotive standards are still developing. Organizations in regulated industries are adopting AI copilots with explicit human-in-the-loop documentation policies to stay ahead of the regulatory clarification rather than waiting for it.

IP and training data governance. Most AI engineering tools, in their default configurations, use customer interaction data to improve their models. For manufacturers with significant IP in their design libraries, simulation results, and change histories, allowing an AI vendor's model to train on that data may represent a competitive risk. Enterprise deployment configurations that isolate customer data are available from all major vendors but add cost and reduce some collaborative learning benefits. The governance decision — what can be used as AI training data, what cannot — must be made explicitly and documented in the organization's data governance framework.

Change management workflow design. When an AI copilot suggests a structural modification and the engineer accepts it, what change management event is triggered? Who reviews? What is the approval threshold for an AI-assisted change versus an engineer-authored change? Most organizations that have deployed AI copilots have not yet addressed these questions systematically. The result is AI-assisted changes flowing through the same change management workflow as human-authored changes — which works, but does not capture the AI's role in the change history and does not differentiate review requirements based on the nature of the change.

Adoption Timeline

Phase 1 (Year 1): Pilot with explicit governance. Select one design team and one AI copilot tool. Establish explicit documentation rules for AI-generated outputs before the pilot begins — what gets recorded in PLM, how AI origin is flagged, what human review is required. Measure productivity impact and human review burden honestly; the second number matters as much as the first.

Phase 2 (Years 2–3): PLM data quality investment. Use the pilot's AI query failures as a diagnostic — every case where the AI returned irrelevant results or missed known-good historical data is a PLM data quality deficiency. Invest in the data quality improvements that will expand the AI copilot's effective context. Extend deployment to additional design teams with the governance framework validated in Phase 1.

Phase 3 (Years 3–5): Enterprise-scale AI-PLM integration. Formalize the AI engineering data layer — structured repositories of design patterns, validated simulation results, and change knowledge that AI systems can query reliably. Connect AI copilot outputs to enterprise rollout change management workflows with appropriate review thresholds. Engage with regulatory bodies on documentation standards for AI-assisted design decisions before they become a compliance gap.

Future Outlook

The 2–5 year horizon for AI copilots in engineering is defined by the convergence of three capabilities. First, multimodal AI that works simultaneously with geometry, simulation data, and textual change history — rather than querying each in isolation — will produce contextually richer suggestions and more reliable BOM drafts. Second, agentic AI workflows that can execute multi-step engineering tasks autonomously — running a simulation, interpreting the results, proposing a design modification, and preparing a change request — will shift the human role from executing engineering tasks to reviewing and approving AI-executed sequences. Third, knowledge graph integration with PLM will allow AI systems to reason over the digital thread — understanding not just what changed, but why it changed, what downstream effects were observed, and what that implies for similar decisions in new programs.

The organizations that will be best positioned for this future are those investing now in PLM data quality, change management governance for AI-assisted decisions, and the human expertise needed to review AI outputs intelligently. AI copilots make engineers more productive. They do not make engineering judgment less necessary — they make it more concentrated, applied to a higher-leverage set of decisions. That is a different skill profile than engineering in the pre-AI era, and building it deliberately is the most important investment an engineering organization can make right now.

Related Resources

• What Is an AI Copilot in PLM — a detailed breakdown of how AI copilots integrate with PLM workflows
• PLM Data Governance — the data quality foundation that determines AI copilot effectiveness
• PLM Configuration Management — how change management must adapt for AI-generated design outputs
• Digital Thread Architecture — the connected data model that gives AI engineering tools full lifecycle context
• PLM Quality and Compliance — regulatory considerations for AI-assisted engineering in regulated industries
• Enterprise PLM Rollout Guide — deploying AI-integrated PLM at scale across a large engineering organization