Digitally Closing the Loop Between Engineering & Production
Gartner says half of PLM platforms will include generative AI by 2026. Here's what's real, including a correction on what GE's AI Wingmate actually does.
PLM, ERP, and MES systems have traditionally functioned as passive systems of record: places where CAD drawings, bills of materials, and change orders get filed after the fact. What's changing is the nature of that record. Rather than asking engineers to leave their core tools to "use AI" somewhere else, AI is increasingly being embedded directly inside the systems they already trust, turning a static repository into something closer to an active collaborator.
Gartner has forecast that by 2026, 50 percent of PLM platforms will incorporate generative AI, up from about 5 percent in 2023. That's a fast shift for enterprise software, and it's happening because the rationale is straightforward: PLM and ERP systems already hold the most authoritative product data in the business, CAD files, BOMs, engineering change orders, compliance records, supplier data. Embedding AI where that data already lives amplifies its value more than extracting it into a separate tool ever could.
AI That Fits Into the Engineer's Existing Workflow
Major PLM vendors are already rolling out AI assistants built directly into their existing interfaces rather than as a bolt-on. An engineer can ask a system in plain language to summarize every engineering change order for a specific assembly over the last two years and get a synthesized answer pulled from actual revision histories, instead of manually cross-referencing a change log by hand. Aras has introduced AI agents powered by Azure OpenAI Service for exactly this kind of conversational search across PLM data.
Bluestar PLM, built on Microsoft Dynamics 365 ERP, offers a concrete example of what this looks like in practice. With Microsoft Copilot embedded across engineering and operations, the system auto-generates part summaries by pulling design history, open issues, and revision status from both PLM and ERP simultaneously, produces item descriptions in multiple languages using live product data for global operations, and flags documentation that needs review when something changes upstream, if a material becomes obsolete in the ERP, the system alerts engineering to check the drawings and specifications that referenced it. PTC's Arena PLM AI engine follows a similar pattern, monitoring component lifecycle changes directly inside the workflow rather than requiring someone to notice the change manually.
AI-Driven Quality Checks Inside PLM
A parallel pattern is emerging in design review itself. Feedback gets captured directly on the drawing as structured markups and comments, with every change tracked under version control, so engineers encounter review feedback in the same environment where they created the design rather than managing it across separate emails and spreadsheets. By catching common drafting errors and standards violations early, AI acts as a first-pass reviewer that reduces the cognitive load on human reviewers and shortens the overall review cycle, the same first-pass pattern this magazine has described for CAD review generally, applied here inside the PLM environment itself rather than as a separate tool.
A Real Example: GE Aerospace's AI Wingmate
GE Aerospace: AI Wingmate
GE Aerospace, working with Microsoft, launched an internal generative AI platform called AI Wingmate in September 2024, rolled out to roughly 52,000 employees. Built on Azure AI and the Azure OpenAI Service, it gives employees a secure internal system for quickly retrieving task-relevant information instead of searching through documents or waiting on a colleague's reply, and for drafting communications or working through a complex problem with AI support.
By early reporting, the system had handled around 500,000 employee interactions and processed more than 200,000 pages of internal content since launch.
✓ Real platform, real usage numbers, independently reportedTraining the Workforce, Not Just the System
None of this technology works if the people using it don't trust it, and the manufacturers that get this right tend to invest in AI literacy before they scale deployment. Bosch has trained more than 65,000 employees through its internal AI Academy since 2019, covering how the technology works, its limitations, and how to validate AI-generated output rather than accepting it uncritically. When employees understand that AI assists rather than replaces them, and that a human retains final authority on anything that matters, resistance tends to drop measurably.
The pattern that holds across every credible deployment in this category is the same one this magazine keeps returning to: AI drafts, a person reviews and approves, and accountability never moves off the human side of that line. AI compiling a compliance document doesn't remove the requirement that a qualified professional verify and sign off on it. AI proposing a design optimization doesn't remove the engineer's judgment on feasibility and risk. That division of labor is what keeps trust intact in regulated environments like aerospace, automotive, and medical devices, where the cost of getting it wrong is measured in more than rework.
Leadership Takeaway
Closing the loop between engineering and production isn't primarily a technology problem. It's a data and trust problem: AI embedded in a PLM system full of inconsistent, poorly structured data will amplify that inconsistency rather than fix it, and AI introduced without genuine workforce buy-in gets quietly worked around regardless of how capable the model is. The manufacturers making real progress here started by cleaning up the underlying data and investing in literacy before scaling the AI itself, not after.
Not sure if your PLM data is clean enough for AI to actually help?
A short operations review usually surfaces whether the real blocker is data quality, workforce readiness, or the PLM integration itself.
See the GrayCyan Operations AI TeardownFAQ
It means AI assistants live inside the PLM or ERP interface engineers already use, answering questions about revision history or flagging documentation that needs review when upstream data changes, rather than requiring engineers to export data into a separate AI tool.
Not specifically. It's a company-wide internal generative AI platform for general knowledge access and productivity, rolled out to about 52,000 employees. GE Aerospace has separate, distinct AI investments in engine monitoring and predictive maintenance that shouldn't be conflated with Wingmate's actual documented scope.
It's a genuine Gartner forecast reflecting a real, fast-moving trend, up from an estimated 5 percent in 2023. As with any single analyst forecast, treat the specific percentage as directional rather than a guarantee for any particular platform.
Because resistance, not capability, is usually what stalls adoption. Bosch's AI Academy, which has trained more than 65,000 employees since 2019, reflects a broader pattern: when employees understand AI's limitations and know a human retains final authority, adoption accelerates measurably.
Poor underlying data. AI embedded in a system full of inconsistent naming conventions, duplicate records, or outdated documentation will amplify that mess rather than resolve it, which is why serious deployments typically start with a data cleanup before scaling AI on top of it.
Looking for AI advice at your company? Talk to our Editor-in-Chief
Nishkam Batta
