Dassault Systèmes brings industrial AI into virtual-twin workflows

Dassault Systèmes connects industrial AI to design, simulation and scientific work through its 3DEXPERIENCE platform. Its Virtual Companions—AURA, LEO and MARIE—are intended to work with the business context and technical tools already surrounding a virtual twin. The buying decision is therefore about a governed engineering or research workflow, including the applications and data it needs, rather than a general-purpose chatbot subscription.

01 / ProductThe platform connects business context with technical applications

The 3DEXPERIENCE platform organizes people, data and applications around shared product and process information. Access is role-based: a role provides the applications needed for a particular job. A virtual twin in this setting can connect geometric models, simulations and operational context, rather than merely presenting a visual duplicate of an object.

The Virtual Companions overview assigns business work to AURA, engineering to LEO and science to MARIE. The company describes agents that can answer, plan and use platform tools. It also says they operate within existing access rights. Those are product claims to validate in the buyer’s environment, especially when an answer turns into a change to a technical artifact.

CATIA’s industrial AI material describes LEO alongside generative experiences for models, assemblies and drawings. BIOVIA’s MARIE page focuses on scientific workflows, laboratory records and computational work. They share a company and platform context, but their outputs require different expertise: an engineer reviewing an assembly and a scientist reviewing an experiment are making different judgments.

02 / AudienceThe strongest audience already has a technical workflow to improve

A manufacturer with an established CATIA environment can examine repeated design changes or documentation work. A simulation group can explore whether an assistant helps connect a design question to the appropriate analysis. A scientific team may instead care about turning research intent into a traceable computational workflow. These are specific entry points, each with an owner and a known standard of evidence.

The Autodesk blueprint provides an adjacent view of AI in design software. Compare the existing models, engineering disciplines and downstream delivery requirements before comparing assistant interfaces. The cost of changing a technical ecosystem can be much greater than the cost of adding one AI capability to a working process.

The NVIDIA blueprint covers infrastructure and software used to build AI and simulation systems. Dassault Systèmes sits closer to the industrial applications and shared product context. A company can use both layers, but buying access to accelerated computing does not by itself create a governed design workflow or supply the engineering methods.

A small team that only needs general writing assistance is unlikely to need the breadth of this platform. The case becomes stronger when technical information has relationships that must survive the AI interaction: a material linked to a model, a result linked to a solver configuration, or a drawing linked to an approved revision. Without that context, the platform’s distinctive value is harder to realize.

03 / WorkflowA proposed design study keeps assumptions attached to every alternative

Consider a team evaluating alternative housings for a piece of industrial equipment. The following is a proposed evaluation, not a test of Dassault Systèmes products. Begin with one approved model and a bounded change request. Identify fixed interfaces, permissible materials, manufacturing limits and the performance question that the study must answer.

Use the assistant first to organize the available information. Ask for a description of the relevant constraints and locate each one in the approved project record. An engineer should resolve contradictions before any design alternatives are generated. If the source material contains an obsolete temperature requirement, a sophisticated workflow can still optimize for the wrong operating condition.

Next, request a small set of alternatives within the authorized design space using the enabled engineering skills. Preserve the starting revision and inspect how each alternative represents the required interfaces. This is where the team should distinguish between a useful geometric proposal and a design ready for release. A model that looks complete may still lack tolerances or assumptions required downstream.

SIMULIA’s AI explanation describes training surrogate models from simulation results to accelerate exploration. In a proposed study, use such a surrogate to screen alternatives only within the region supported by the training cases. Reserve detailed simulation for selected candidates and for cases near the boundaries of that region.

Write down what would make an alternative unacceptable before looking at the results. The list might include an inaccessible fastener, an unsupported material choice or a thermal limit exceeded under a defined load. Reviewers can then assess whether the workflow surfaces those constraints consistently. An assistant’s confident explanation should not substitute for the actual model, solver inputs or engineering calculation.

Compare the final candidates using the same criteria and carry the chosen result through the established design review. Retain the source revision, assumptions, generated alternatives and reason for selection. The meaningful output is a traceable decision package that another engineer can inspect, not merely a rendering or a persuasive summary of why one option appears best.

For a scientific team, use the same principle with a different artifact: the protocol, dataset and computational method should remain attached to the result. MARIE’s described laboratory and computational functions make that a relevant separate pilot. Do not use an engineering demonstration as evidence that a scientific workflow is validated for another discipline.

04 / PricingBlue Tokens are a shared consumption pool with eligibility limits

The live Blue Tokens store, consulted on 24 September 2026, displayed 25,000 tokens for a $1,250 yearly subscription excluding local tax. The page states that tokens are pooled at tenant level, expire after twelve months and auto-renew unless renewal is stopped. It does not provide a universal task-to-token conversion.

The store lists a defined group of eligible countries for direct online purchase and offers a partner or sales route for other cases. It also says the cancellation policy described in its FAQ does not apply to Blue Tokens. Confirm the complete order terms before buying capacity for an uncertain pilot, particularly where procurement expects unused capacity to remain available indefinitely.

Token consumption depends on the task. A short lookup and a complex generative operation should not be budgeted as equivalent requests. Our proposed cost measure is consumption per accepted engineering or scientific outcome, including failed attempts and reruns after a reviewer changes an input. That connects the bill to useful work without assuming that every generated output saves time.

Keep application access and AI consumption separate in the estimate. The platform’s role structure determines which technical capabilities the team can use; the token pool funds the associated AI work. Ask for an explicit list of required roles, skills and deployment options, then evaluate the workflow in that configuration. A generic token purchase is not a complete implementation plan.

05 / DistinctionsPhysics and product structure provide a meaningful basis for assistance

Dassault Systèmes’ distinctive position is the connection between generative interfaces and technical systems that already model geometry, materials and physical behavior. In principle, an agent can call a relevant engineering tool instead of trying to answer every question from text alone. That is useful only when the tool inputs and the interpretation of its output remain visible.

The company’s Virtual Companions material makes strong claims about scientific grounding and verifiable outputs. This blueprint treats those as vendor claims, not as evidence that errors are impossible. A simulation can be internally correct for its assumptions and still be unsuitable for the actual product. The review process must therefore examine the assumptions as well as the calculation.

The shared platform can also help retain the context behind a result. A decision that remains linked to its model and review is easier to revisit when a supplier changes a material or a requirement changes. That is a more durable benefit to evaluate than simply asking whether a conversational interface feels convenient on its first demonstration.

06 / QuestionsConfirm which skills work in the purchased environment

Ask the vendor to demonstrate the exact workflow on the intended cloud configuration and application roles. Broad companion names are not a precise feature inventory. AURA, LEO and MARIE can each cover several tasks, and the enabled skills determine what the team can actually attempt. Include the administrator in the demonstration so access and consumption are understood together.

For generated engineering work, inspect how reviewers see proposed changes and recover the previous state. For scientific work, examine how source records, generated text and computed results are distinguished. These are different evidence types and should not be blended into a single authoritative-looking narrative.

Also verify the hosting region and contractual data handling for the selected service. The public overview describes several cloud operating models and service-specific conditions. It is insufficient to infer that every computation stays in any region the buyer chooses. Resolve the actual processing arrangement before uploading restricted product or research material.

07 / DecisionChoose one technical outcome and require a reviewable result

Dassault Systèmes is relevant to industrial AI because it can connect assistance to detailed technical models and the tools used to evaluate them. Start with one repeatable engineering or scientific decision, establish its evidence requirements and observe the complete cost of reaching an accepted result. The platform’s breadth is a reason to scope the first workflow carefully.