Every packaging machine on a production floor represents thousands of invisible decisions. The height of an operator interface. The clearance between a servo motor and an enclosure wall. The material grade chosen for a bracket that will endure ten years of washdowns. These decisions live in the minds of experienced engineers, passed down through mentorship, accumulated through trial and error, and too often lost when someone retires or moves on.

At 3DExperience World 2026 in Houston, Dassault Systèmes made a bold case that artificial intelligence is finally ready to capture, preserve, and actively apply that institutional knowledge. In a session titled “Design in the Age of AI,” SOLIDWORKS CEO Manish Kumar laid out a vision that should have every packaging OEM paying attention. Not because AI is coming to replace their engineers, but because it is coming to make sure their best engineering knowledge never disappears.

The invisible table

Kumar opened with a deceptively simple example: a wooden table he built for his home office. On the surface, it is just furniture. But Kumar used it to illustrate a distinction that sits at the heart of Dassault’s AI strategy: the difference between knowledge and know-how.

Knowledge is why something works: the ergonomics, the anthropometry, the physics, the embedded intelligence that determines a table’s dimensions and proportions. Know-how is how it comes to exist: grain direction, manufacturing constraints, joinery, assembly sequence.

A CAD file captures the geometry. But it rarely captures the reasoning behind the geometry. Think of the failed biscuit joint attempt, the lesson learned about cross-grain expansion, the reason a particular fastener was chosen over another.

RELATED | Designing the next-generation more sustainable package

For packaging machine builders, this distinction is everything. A cartoner’s cam profile, a filler’s valve timing sequence, the specific way a changeover mechanism was designed to accommodate three different carton sizes. These represent decades of accumulated know-how. And in an industry facing a well-documented skilled labor shortage, that know-how is walking out the door faster than it can be documented.

Kumar’s table example was deliberately low-tech, but the principle scales directly to industrial machinery. Ask a packaging OEM’s senior mechanical engineer why a particular guide rail is mounted at a specific offset, and the answer is rarely “because the CAD model said so.” It is because three years ago, on a machine running at 400 cartons per minute, a thinner rail deflected under load and caused jams every forty-five minutes. That lesson lives in the engineer’s memory. It almost never lives in the model.

From automation to composition

Dassault’s answer is not a single omniscient AI. Kumar was explicit about this: general-purpose large language models are not engineering tools. They are, as one of his slides put it, “mathematical prodigies with zero real-world experience,” brilliant at pattern matching but prone to hallucination when applied without constraints. Engineering requires compliance, not just creativity.

RESEARCH | Smarter Packaging Operations With AI

Instead, Dassault introduced what they call the “Generative Triad,” three specialized AI companions, each with a distinct role. 

• Aura is the Explorer, a connector that leverages existing enterprise and web knowledge to help with business context, sourcing, and project planning. 
• Leo is the Engineer, focused on mechanical design, validation, tolerance analysis, and manufacturing integrity. Leo is the companion that will flag when a user specifies a 2×6 plank where a 1×6 with proper bracing would deliver equal strength at half the weight. 
• Marie is the Scientist, handling deep research questions around materials, simulation, and hypothesis testing.

The key concept Kumar emphasized was the shift from automation, a single command to a single tool, to composition: orchestrating a team of distinct intelligences to navigate a design space that is, as Kumar described it, infinite. This reframes the designer’s role from operator to architect, coordinating specialized agents rather than manually executing every task.

The live demonstrations reinforced the point. Kumar showed SOLIDWORKS 2026 with the companions integrated directly into the design environment. Rather than a separate application, it appeared as a persistent sidebar that could be invoked through chat, voice, right-click context menus, or even proactive alerts. In one demo, dubbed “Vibe CADing,” Leo interpreted natural-language instructions to build and modify a 3D part in real time, including creating hook cuts at specific angles and applying fillets to selected faces. The companion showed its reasoning chain at each step: design intent, geometry engine actions, and the specific CAD operation it was executing.

What this means for machine builders

For OEMs designing packaging machinery, three capabilities demonstrated in the session stand out.

First, context-aware design assistance. The AI companions shown in SOLIDWORKS 2026 are not generic chatbots bolted onto a toolbar. They are aware of the active application, the specific part or assembly open on screen, the user’s role, and even past conversations. Kumar demonstrated Leo proactively identifying a manufacturing error in an xDesign session before the user asked. For a packaging engineer working on a complex multi-axis motion system, having an AI that understands the context of the design, not just the geometry, could meaningfully reduce the iteration cycles between concept and validated design.

Second, knowledge preservation at scale. Kumar showed how the platform can transform scattered tribal knowledge (the BOMs, the failed prototypes, the lessons learned from commissioning) into structured, searchable intelligence that informs future designs. For machine builders running lean engineering teams, this addresses one of the most persistent risks in the business: the departure of a senior engineer taking irreplaceable know-how with them.

RELATED | The invisible upgrade: how AI and digital twins are reshaping packaging machinery

Third, trusted outputs with traceable reasoning. Every recommendation from the AI companions is grounded in specific “knowledge universes” with traceable IP lineage. When LEO recommends dual-phase steel for a specific application, it shows which studies and data sources informed that recommendation. The companions use reasoning language models (RLMs) to check for inconsistencies, and are designed to admit uncertainty rather than hallucinate an answer. For machine builders operating in regulated environments or building to customer specifications, traceability is not a nice-to-have. It is a requirement.

The bigger picture

Kumar’s session did not exist in a vacuum. The day prior, NVIDIA CEO Jensen Huang joined Dassault CEO Pascal Daloz on stage to announce what both companies described as the largest partnership in their 25-year history, integrating NVIDIA’s CUDA X libraries, AI frameworks, and Omniverse into the 3DExperience platform. Huang described a future where AI-powered physics simulation runs 10,000 times faster than traditional methods, and where every designer will work alongside a team of AI companions trained on domain-specific expertise.

MORE COVERAGE | Beyond digital twins: NVIDIA and Dassault bet on industry world models (via engineering.com) 

Huang also addressed a question that is top of mind for many smaller manufacturers: whether the AI and robotics revolution will leave them behind. His answer was direct, and it started with a distinction that any packaging OEM should recognize. Robots, Huang noted, are already everywhere, but only where tasks are highly repetitive and volumes are massive. The automotive industry can justify the cost of explicitly programming a robot for a single task that will be performed millions of times. But the vast majority of the world’s manufacturers do not operate at that scale.

“The vast majority of the world’s industries, these are small and medium-sized factories,” Huang said. “The tasks are quite diverse, and the company is quite small. Quite small in the sense that they’re not fifty billion dollars large. Maybe they’re fifty million dollars large. Maybe they’re a hundred million dollars large. Most of these companies, in fact, drive the world’s supply chain, and all of their tasks are different.”

Anyone who has spent time with packaging OEMs will recognize this description immediately. A mid-size machine builder running changeovers across dozens of SKUs, building custom configurations for each customer, operating with an engineering team of eight or twelve people. This is exactly the kind of company Huang was talking about. And the reason those companies have not been able to fully leverage robotic automation, he argued, is straightforward: the cost of programming is too high relative to the diversity of tasks.

The answer, Huang said, is a fundamental shift from explicit programming, where every robot motion must be manually coded by a specialist, to implicit programming, where a robot learns from demonstration. Instead of a dumb robot that is explicitly programmed in a high-volume environment, the future requires AI-powered robots that are flexible enough to learn. They can learn through demonstration, through reading a manual, through watching a video, in much the same way a new technician would.

[nesletter]

For packaging OEMs who build machines for customers ranging from multinational CPG companies to regional co-packers, this trajectory has significant implications for how machines are commissioned, programmed, and maintained in the field. The vision is not a single humanoid form factor. As Huang put it, it could be two arms on a table, two arms on wheels, a full humanoid, or a single arm. The shape matters less than the intelligence driving it.

Dassault’s Daloz reinforced the democratization theme, comparing the current moment to the early days of CAD itself. Thirty years ago, only large enterprises had the resources for well-equipped design offices with supercomputers. Today, millions of engineers use these tools. Daloz argued that AI will drive a similar expansion, not by lowering the bar for engineering quality, but by making sophisticated capabilities accessible to organizations that could never afford dedicated simulation teams or materials scientists on staff. 

“I think it’s gonna lift them, actually,” Daloz said. “And it’s gonna enable more diversity of products to be created, more diversity of machinery to be created, more diversity of industries to be sustainable.”

Intelligence by design

Taken together, the message from Houston was clear: the design tools that packaging OEMs rely on are undergoing a fundamental transformation. Not from CAD to something else, but from tools that draw objects to what Kumar called “factories of knowledge and know-how.” The geometry is still there. But now it carries the reasoning, the constraints, and the institutional memory that make the difference between a machine that works on paper and one that runs reliably on a production floor.

For machine builders evaluating their technology roadmap, the practical question is not whether AI will reshape the design workflow. It is whether your organization’s accumulated know-how, the invisible intelligence embedded in every machine you build, will be captured and leveraged, or lost.

Kumar’s wooden table example was a reminder that even the simplest objects carry more embedded intelligence than meets the eye. A packaging machine carries orders of magnitude more. The tools to capture it are arriving now, and the OEMs who adopt them first will be the ones whose best engineering decisions survive the next generation.

About the author