AI in Design, AI in Manufacturing, and the Bridge
Walk any engineering conference in 2026 and the same picture assembles itself from separate talks. One session shows AI on the factory floor: inspection, process optimisation, quality prediction. Another explores AI in the design studio: generative concepts, feasibility checks, faster modelling. And the CAD vendor on stage announces AI agents built directly into the tool. Each story is real. Each is also told in isolation.
The question that connects them is the one that gets the least airtime: how does design intelligence actually reach manufacturing? A senior engineer's understanding of how a specific part should be tooled, drawn, and built is not a document you can hand off. It is judgement, applied to geometry that is often unique. AI is now helping on both banks of that river; the crossing itself, translating a finished design into the manufacturing tooling and documentation it implies, is still mostly manual, and it is still gated by a few experienced people. That crossing is where FDES works, and understanding why the new SOLIDWORKS agents do not close it is the fastest way to see the shape of the gap.
What Aura, Leo, and Marie Actually Do
First, credit where it is due: the SOLIDWORKS AI companions announced at 3DEXPERIENCE World 2026 are a genuine step forward, and for a large number of everyday tasks they are exactly what engineers have wanted.
- Aura is a knowledge companion. It draws on SOLIDWORKS' documentation and your 3DEXPERIENCE resources (including 3DSwym discussions) to answer "how do I do this in SOLIDWORKS" questions and summarise community posts, wikis, and ideas. It runs inside your 3DEXPERIENCE workspace, so your data and conversations stay private to your company. Think of it as expert product knowledge, on tap.
- Leo, the engineer's companion, is available now in SOLIDWORKS 2026 through the SOLIDWORKS Labs beta, with capabilities expanding through 2026 and beyond. Driven by natural language, it can generate assembly structures and sketches, drive prompt-based drawing creation, run a simulation study for validation, and report back on mass properties and feasibility as you design. It brings a conversational workflow into the design environment.
- Marie, arriving later, is positioned to bring materials and scientific rationale into the design process.
SOLIDWORKS 2026 also brings AI-assisted capabilities into everyday CAD tasks: accelerating drawing creation, automatically recognising fastener-like components, and reconstructing a 2D PDF or a neutral STEP file into a usable parametric model (in beta). These represent a meaningful shift in how AI is becoming embedded into everyday CAD workflows. If your engineers spend time on standard modelling, documentation, and looking things up, these tools will give that time back.
Why They Stop Where They Do (By Design)
Here is the part worth understanding clearly, because it is easy to misread as a criticism and it is not. The SOLIDWORKS companions are horizontal: one assistant serving millions of users across every industry, from consumer products to heavy fabrication. That breadth is the whole value proposition, and it sets two boundaries that follow directly from it.
First, they are trained on general knowledge, not on your company's history. Aura learns SOLIDWORKS; Leo learns engineering and mechanics. Neither is trained on your hundreds of completed tooling projects, or the specific way your senior engineers decide which faces on a novel part become the die area, the punch contact, or the electrode surface. A general model cannot also be a model of your proprietary judgement; those are different things, and a companion for everyone is deliberately the former.
Second, they are assistants, not unattended automation. SOLIDWORKS positions these companions to keep an engineer in the loop: they suggest, generate, and check, and a person reviews and decides. That is the right and honest posture for a general tool. It is also the opposite of what a production pipeline needs, where the goal is to produce a release-ready, repeatable, auditable result without a senior engineer hand-holding every job.
So the boundary is structural, not a shortcoming to be patched in the next release. A copilot for the world and a trained model of one company's engineering knowledge are two different products with two different jobs.
The Gap: Company-Specific Intelligence
That structural boundary leaves a specific, valuable space open. It is easiest to see as a short list of things a general copilot is not built to do, and a custom AI layer is:
- Learn your engineering judgement from your own data. A model trained on your archive of finished projects captures how your company reads a part, the recognition step that classic automation could never encode because every incoming job is a different shape.
- Automate end to end, with validation. Not "help an engineer draw the tooling," but produce the die, punch, electrode, and manufacturing drawings for a never-before-seen part, deterministically, with confidence thresholds deciding when a result ships and when it needs an engineer's review.
- Carry your intent across the design-to-manufacturing line. Translate a completed design into the specific tooling and documentation your shop floor actually needs, using your rules, not generic ones.
- Retain knowledge you are about to lose. Capture the judgement of the few senior people who can do this work today, so it stops walking out the door at retirement.
We covered how such a system is built in AI in CAD Automation: What Actually Works in Production, and where the pattern applies across industries in When Every Part Is Unique. The short version: a model trained on your history answers "what is this geometry?" on a part it has never seen, and conventional CAD automation then builds every feature natively. The AI is the brain; SOLIDWORKS is the hands.
The Design-to-Manufacturing Bridge
This is the bridge from the SYNC stage, drawn as a picture. AI now assists on both sides. The crossing in the middle, where a specific design becomes specific manufacturing tooling, is the part that needs a model of your intelligence, and it is the part general companions leave open.
Complementary, Not Competing
None of this is a reason to skip Aura or Leo. It is a reason to know which job each tool is for. The two layers stack: the SOLIDWORKS companions make every engineer faster inside the design environment, and a custom layer automates the company-specific, non-repetitive work that no general assistant is built to own. Here is the division, side by side:
| SOLIDWORKS AI (Aura / Leo) | Custom AI layer (what FDES builds) | |
|---|---|---|
| Knowledge base | SOLIDWORKS documentation, general engineering knowledge, and the 3DEXPERIENCE resources it can draw on | Your own proprietary past projects and the engineering judgement encoded in them |
| Data & hosting | Runs in the 3DEXPERIENCE cloud platform; your data stays private to your workspace | Runs on your own hardware, on premises; your data never leaves your network |
| Geometry | General CAD workflows across every industry | Domain-specific recognition of your part geometry, learned from your history |
| How you use it | Conversational co-pilot: an engineer asks, it assists, the engineer stays in the loop | Automates the workflow, with confidence thresholds and validation deciding when it proceeds on its own and when an engineer reviews |
| Scope | General and horizontal: the same assistant for every SOLIDWORKS user worldwide | Narrow and vertical: specific to your parts, your tooling, and your rules |
| Best at | Accelerating a person while they design inside SOLIDWORKS | Scaling expert knowledge across repetitive workflows with variable geometry |
| Design to manufacturing | Assists within the design stage; recognises standard components; helps prepare drawings | Translates your design intent into the manufacturing tooling and documentation it implies |
Read the two columns and the relationship is obvious: they are not substitutes. One is a horizontal assistant that makes the tool easier to drive; the other is a vertical automation of your specific process. A company that adopts Aura and Leo and then builds a custom layer on top is not hedging; it is using each for exactly what it is good at.
Stacked, the picture is three layers, each doing the job it is best at, with a feedback loop that improves the model over time:
When You Need the Custom Layer
The general companions are enough for most teams most of the time, and that is fine. You are looking at the custom layer when the following are true together:
- Your parts are genuinely non-repetitive. Every incoming job is a new shape, so there is no template for rules to anchor to and no standard prompt for a general copilot to satisfy.
- Expert judgement is the bottleneck. A handful of senior engineers are the only people who can interpret a novel part before manufacturing can proceed, and that judgement is not written down anywhere.
- The work is repetitive as a workflow. The same tooling set and the same drawing package, produced again and again, just never for the same geometry twice.
- You have history to learn from. Hundreds of completed projects in your archive whose outcomes encode the decisions you want to capture and keep.
If that is your shop, the SOLIDWORKS companions will make your engineers faster, and they will still hit the same wall your automation attempts always have: the moment a design has to become your specific manufacturing tooling, on a part no one has seen before. That is the crossing. It is the one part of the picture a general assistant is not built to own, and it is exactly the part we do. If you saw our talk at SYNC 2026 and want to know what this looks like on your own workflow, that is the conversation to have.
SOLIDWORKS, AURA, LEO, and 3DEXPERIENCE are trademarks of Dassault Systèmes. This article is independent commentary by FDES Technologies and reflects our own views.
