AI agents move from sidekick chatbots to real design collaborators
AI virtual companions in design are specialized software agents embedded directly inside engineering platforms that understand models, simulations and production data, then perform or automate domain-specific tasks such as program management, design validation, simulation setup and manufacturing planning while staying within the tools engineers already use every day.
Dassault Systèmes has switched this idea on for real: its three AI virtual companions—AURA for program management, LEO for complex engineering and MARIE for deep science—are now live on the 3DEXPERIENCE platform, supporting workflows across design, simulation, manufacturing and operations with 19 competencies. This is not just another chat window sitting next to CAD; it is Dassault 3DEXPERIENCE AI wired into the same environment that already runs product development for many enterprises. In effect, the platform now comes with AI workers that can interpret user intent, reason in an industrial context, and generate outcomes grounded in industry-accurate reality. The key takeaway: AI agents are no longer an optional add-on but are becoming an expected part of enterprise design automation.
AURA, LEO, MARIE: 19 competencies and one clear direction
Dassault’s trio is opinionated by design. AURA is positioned as the agreeable companion for program management, LEO focuses on complex engineering tasks, and MARIE pushes into assertive deep science work. Together they are “enriched with 19 competencies” to co-create products, assets and services and solve complex industrial challenges faster and more efficiently. Each competency is an area of knowledge, such as mechanical design, composed of skills like conceptual sketching and detailed design. In practice, these AI virtual companions design and execute units of work: they guide users, reveal invisible insights, explore untapped spaces and execute relevant tasks across business, product design, engineering, simulation, manufacturing and operations. Because they interpret intent and reason using structured industry knowledge, they are built for manufacturing AI agents scenarios—checking feasibility, proposing alternatives, and closing the loop between program management, engineering and operations rather than acting as generic text bots.
This is the beginning of a division of labor where engineers keep ownership of decisions, while AI agents do the legwork across 19 clearly scoped domains. It is a pragmatic, constrained approach that matches how real enterprises adopt automation: start with well-bounded competencies, then expand.
Embedded in 3DEXPERIENCE: automation without leaving your CAD
The most important design choice here is not the personalities; it is the CAD AI integration. AURA, LEO and MARIE run directly inside the 3DEXPERIENCE platform, so engineers do not need to switch tools or export data to benefit from AI. Each companion can act on units of work using structured industry knowledge and know-how, interpreting user intent in context and taking decisions that affect models, simulations and manufacturing plans. This is where enterprise design automation stops being a buzzword: repetitive design validation, simulation setup and manufacturing planning tasks can be executed as AI-assisted workflows spanning the same design, simulation, manufacturing and operations data that teams already trust. Because the virtual companions operate on AI factories from OUTSCALE—Dassault’s cloud and AI operator deployed across three continents—customers get industrial-grade AI at scale while data and intellectual property remain protected on sovereign infrastructure.
For ordinary users, the impact is tangible: fewer manual clicks to prepare simulations, fewer nights spent checking drawings against standards, and faster loops between design and shop-floor planning. The work does not vanish, but its most repetitive layers can be offloaded to agents that understand both the product and the process.
Virtual twin factories and the new division of labor
Dassault is already projecting a bigger picture: what it calls virtual twin factories, where virtual companions act as workers alongside humans, generative experiences run highly automated processes, and virtual twin as a service provides predefined industry world models. This vision makes a bold claim: manufacturing AI agents will be treated as peers in digital factories, with defined roles in program management, engineering, simulation and operations. It is an ambitious framing, but the underlying logic is sound. If AI virtual companions design and execute well-bounded units of work with industry-accurate outcomes, they can become reliable workers for digital validation and planning tasks before anything hits the physical line. The risk is that enterprises overestimate what current competencies can handle and underinvest in governance and human oversight. The smart move is to treat virtual twin factories as a direction of travel, not a switch you flip in one release.
Used wisely, this concept lets teams stress-test product and factory changes with agents that never tire of parameter sweeps and scenario comparisons. Used carelessly, it can tempt managers to offload accountability to algorithms that still require expert supervision.
An industry shift: Onshape’s AI push and what teams should do now
Dassault is not alone. PTC’s cloud CAD platform is rolling out an early access program called Onshape Labs to give users a preview of upcoming AI features and a way to give feedback. Initially this includes AI Quick Render and workflows that link Onshape CAD data to Nvidia’s robotics simulation framework, with plans for AI agents, automation capabilities, and an AI-powered drawing checker. PTC expects Onshape Labs to be generally available later this summer. The signal is clear: embedded AI agents inside CAD and design tools are becoming table stakes, not differentiators. For engineering leaders, the right response is to start small but deliberate. Pick one or two repetitive workflows—such as design validation checks or simulation template setup—and pilot AI automation there. Measure cycle time and error-rate changes, then expand. Waiting for the perfect roadmap is now the bigger risk than experimenting with today’s tools.
Enterprise teams that treat AURA, LEO, MARIE and forthcoming Onshape agents as colleagues to train—rather than magic buttons—will get the real advantage: more time for engineers to solve new problems, and less time lost to clerical CAD work.
