[DigitalToday reporter Chi-gyu Hwang] "It has been only 4 to 5 months since the official launch, but strong success cases are already emerging. Some customers have raised productivity by 50 to 60 percent."

Manish Kumar (마니쉬 쿠마), CEO of SolidWorks, Dassault Systemes’ 3D design software unit, shared results since putting the "Virtual Companion" series, positioned as engineering AI agents, into real-world use in July. He made clear that AI in engineering is an unavoidable trend.

Dassault Systemes unveiled its first Virtual Companion, Aura, in early last year. Early this year, it released preview versions of Leo and Marie, Virtual Companions designed to handle more advanced engineering tasks, and officially introduced them in July.

According to the company, Aura is specialized in discovering and exploring ideas. It connects internal data with external knowledge on the web to help users flesh out design ideas. Leo is an engineering AI that reviews manufacturability and design feasibility and suggests designs suitable for actual production. Marie focuses on science-based analysis and covers advanced areas such as drop tests, ergonomics evaluations and medical device regulatory responses.

Kumar said SolidWorks has encouraged users of the cloud-based 3DEXPERIENCE platform to try the Virtual Companions by distributing tokens. He said success cases are emerging despite the short period. "There are three Virtual Companions, but competencies and skills being applied to individual products are continuously being added," he said. "More advanced skills will be added as well."

He also cited a customer case. He said the company tested a feature that creates 2D drawings using a Virtual Companion based on its own dataset. Productivity rose 50 to 60 percent from that alone, and the goal is to lift it to 90 percent within 6 months.

Kumar said 2D drawings play a large role in manufacturing. He said that even if designs are done in 3D, factories still run on 2D drawings. Drawings must be made for parts and assemblies, and about one-third of design time is spent on that, he said. He said he has seen many people who like designing but none who like working on 2D drawings. If a 90 percent productivity gain becomes reality, the impact will be that much greater, he added.

Kumar said there are many tasks besides 2D drawings that have to be done but that people would rather avoid, and that covering such work with AI would be good for both companies and employees. He gave the example of the work of aligning holes for parts used in assembly. "No one welcomes it because it is cumbersome and boring, but mistakes in this area create big costs," he said.

Kumar also highlighted that Virtual Companion supports a range of manufacturing processes and is not AI only for design. He said Dassault Systemes’ 3DEXPERIENCE platform tracks products from the concept stage through to products deployed at customer sites and moved beyond design long ago. He said Virtual Companion also spans areas beyond design, including ideation and data management.

Competition in engineering AI appears to be intensifying in recent months. The number of startups specializing in engineering AI is also rising. Kumar said Dassault Systemes has a moat that competitors cannot easily cross in terms of breadth and depth. He said Dassault Systemes is focusing on an industrial world model that differs from large language models and general world models, based on capabilities in physics. He said LLMs learn the statistical structure of language to predict the next word, but cannot tell where a drone will be 1 second or 2 seconds later. He said world models are AI that predict the next state from a current state and mostly learn by looking at images and videos. He stressed that industrial world models must be based on actual laws of physics rather than pattern matching, and must understand areas including structures and fluid dynamics to be usable in industrial settings.

Kumar also cited the ability to cover a wide range of manufacturing processes as a strength of Dassault Systemes.

He said the company has everything from design modeling to simulation tools. He said it supports simulation regardless of scale or physical domain, from the molecular level to large aircraft such as the Boeing 777. He also said it has accumulated shape and physics data over a long period. Few companies, or only Dassault Systemes, meet those conditions, he added.

Dassault Systemes is focusing on targeting manufacturers that cannot afford to build their own GPU infrastructure, with cloud-based Virtual Companions.

Kumar said using AI in the cloud allows anyone to secure knowledge and computing power while separating and protecting data. He said startups in the software industry are already challenging big companies in that way. He said Dassault Systemes trains its AI models only with its own data and synthetic data and does not use customer data without written consent. Customer data is always separated and protected, and control remains with the customer, he added.

As AI spreads in engineering, engineers’ roles will inevitably change.

Kumar said the value of an engineer lies in finding the right intent. He said that means designing the intent, setting guardrails the model must not cross, and evaluating the output, and that the final responsibility also rests with people. He said if a bug occurs after applying an AI-provided solution as is, one cannot say it was "AI's work." Fixing bugs must also go through engineering team testing, he said. He said the more design proposals AI produces, the more important it will be to choose the one that fits best and prove it.

He named three capabilities engineers need in the AI era.

The first is finding the right problem. He said AI provides answers but cannot find the problem. People must also decide which of multiple problems to solve first, he said.

The second is summarized as working while building good relationships with colleagues and customers through soft skills. He said if AI gives anyone execution capabilities, people will look for someone who is enjoyable to work with rather than a capable "arrogant superstar."

The last is being a multi-player. He said if someone is in mechanical engineering, they must also learn electrical and electronics and software. He said robots, at the peak of physical AI, are the result of combining all those fields.