AI-Native Design & CAD: The New Layer Between Idea and Manufacturable Design

Design and CAD is the step where engineering ideas start becoming actual physical products. Traditional design platforms are powerful, but they are also manual, expert-heavy, and often slow to explore. In this landscape, I looked at the new generation of AI-native Design & CAD startups trying to change that. 

Below are the main insights I learned about how traditional design workflows work, what these new products bring, and why the category is becoming interesting from an investment and acquisition perspective.

How do Design & CAD workflows traditionally work?

• At a basic level, design and CAD is the moment where an engineering idea starts becoming a real object. Before this stage, the work is mostly about requirements, constraints, performance targets, cost, and technical assumptions. CAD is where these abstract inputs start becoming components that can later be simulated, tested, and manufactured.

• But design is not only “drawing the shape.” A CAD model is not just a visual representation of an object. It carries part of the design intent. It tells the engineer how dimensions relate to each other, how a feature should behave when something changes, and sometimes how the part should eventually be manufactured etc.

• Mechanical design has traditionally been a very manual and expert heavy workflow. Engineers create sketches, define features, add constraints, adjust dimensions, and slowly build the model. The power of CAD is that the model can remain editable over time.

• Electronics design has a similar logic, but the design object is different. Instead of mechanical geometry, the engineer works with requirements, components, schematics, board layout, routing, and manufacturing rules. Here too, the difficulty is not just to create an output file. The difficulty is to make thousands of small technical choices that need to be correct, inspectable, and manufacturable.

• A lot of design work is really exploration, but exploration is expensive. In theory, engineers would like to test many design options. In practice, each option takes time to create, modify, simulate, and review. So teams often explore fewer alternatives than they would like. This is one of the main bottlenecks AI native design tools are attacking.

• Another limitation is that CAD files do not always capture the reasoning behind the design. A file can contain the final shape or the board layout, but it does not always explain why one option was chosen over another. It does not always show which assumptions were made, which trade-offs were considered, or which constraints mattered most. A lot of engineering memory still lives around the file, not inside it.

What new generation AI-Native Design & CAD products bring?

• The general direction is that AI is moving design tools from manual creation toward assisted generation and exploration. The obvious promise is that an engineer can describe what they want and get a first design faster. But the deeper promise is that the system should understand enough about design intent, constraints, physics, and manufacturing to help the engineer move through the loop faster.

• The first product pattern is text, sketch, image, or scan to CAD. Zoo, Backflip, Adam, and Kyrall are good examples of this direction. They all ask a similar question: can we make the first version of a 3D object much faster to create? But the challenge is not whether the system can generate something that looks good. The challenge is whether the output is editable and useful for real engineering work.

• Zoo is probably one of the clearest examples of the “new CAD stack” approach. It is not just adding an AI feature on top of an old product. Zoo Design Studio combines a CAD interface, a geometry engine, a programming language, and Zookeeper, its conversational CAD agent. 

• A second product pattern is physics aware design exploration. Neural Concept, ToffeeX, and nTop sit more in this direction. They are less focused on generating a random object from a sentence. Instead, they help engineers explore better designs under real technical constraints. This is where AI native design starts to overlap with simulation and optimization (see the other landscapes).

• Neural Concept is a good example of a company moving upstream from simulation acceleration into design generation. The company already had a strong position around engineering AI and simulation driven workflows. But its AI Design Copilot pushes it closer to the front end of design. This is important because AI-native CAD may not only come from CAD startups. It may also come from simulation first companies moving upstream.

• ToffeeX shows another version of this theme, but around physics driven generative design. What I find interesting is that the product seems to be moving toward more engineer control. Features that let engineers set a starting shape, define repeated patterns, or test many parameter variations suggest that generative design cannot be fully automatic and opaque. Engineers still need to guide the system, understand why it produced a result, and be able to recreate that result later.

• nTop is a bit different because it is older and not “AI-native” in the new LLM sense. But I still think it is useful in the landscape. It represents the broader computational design direction: complex geometry, implicit modeling, automation, and simulation-driven workflows. It is a reminder that AI-native CAD is not starting from zero. The move toward computational and programmable design had already started before the current AI wave.

• A third product pattern is AI native electronics design. This is actually one of the strongest clusters in the landscape. CELUS, Quilter, Flux, JITX, and Diode all attack the electronics design workflow from different angles. The design object is not a 3D mechanical part, but the pain is similar: engineers need to move faster from requirements to a manufacturable design.

• Quilter and Flux are good examples of PCB design becoming more autonomous. Quilter focuses on physics driven PCB layout and has been adding more advanced board-design capabilities over time. Flux is more agent oriented. Its assistant can research, select components, build schematics, and document decisions. Both are attacking the same underlying pain: PCB design is slow, expert heavy, and full of repetitive decisions.

• JITX and Diode show another version of AI native electronics. Instead of designing only through visual tools, part of the circuit board design is written in code. This makes the design easier for software and AI systems to understand. It also means the AI can more easily check the design, suggest changes, explain issues, and help engineers modify the board.

• The big shift across the whole landscape is from one shot generation to workflow participation. The first demo is often “generate a model from text.” But the real product direction is broader. The system needs to help generate, modify, validate, explain, review, and hand off the design. This is visible in Adam moving toward a hardware team copilot, Zoo launching Zookeeper, Flux moving from assistant to agent, and Diode connecting design with review and manufacturing.

• The key question is trust. In software, you can generate code and test it quickly. In engineering, the output eventually becomes a physical object. So the AI tool cannot only generate something plausible. It needs to preserve design intent andrespect constraints.

Investment & Acquisition environment

• The funding environment looks more mature than the workflow automation landscape and closer to the simulation one. There are still very early companies in the category, but several companies have already raised meaningful Series A, Series B, Series C, or Series D rounds. My impression is that investors are more bullish on simulation and design generation startups than the engineering copilot category I've covered.

• Neural Concept is the clearest late stage funding signal in the landscape. The company raised a $100M Series C in 2025, after a $27M Series B and earlier rounds.

• The electronics design cluster is surprisingly well funded. Flux raised a $27M Series B in 2026. Quilter raised a $25M Series B in 2025. Diode raised an $11.4M Series A. JITX raised a $12M Series A. CELUS raised a €25M Series A. This tells me that PCB and electronics design automation is not a side category. It may be one of the strongest commercial wedges for AI native design.

• The 3D and mechanical CAD generation companies are more mixed in maturity. Backflip raised a $30M Series A, which is a strong signal. Adam is still seed stage. Zoo is still early, but with a very ambitious product architecture. Kyrall is even earlier in the data I have.

• Plenty of top tier investors have made bets in this space. You see top tier investors such as Andreessen Horowitz, Index, Benchmark, Sequoia, Insight, Tiger Global etc.. 

• Unsurprisingly, the founding teams are extremely domain heavy. It’s similar to the other engineering related landscape I shared. Neural Concept comes from EPFL and deep learning for engineering. ToffeeX comes from Imperial College research. Backflip comes from Markforged. Quilter comes from SpaceX electronics. Diode has Apple robotics and hardware experience. JITX has deep robotics, UC Berkeley, and EDA expertise. 

• There is not much acquisition activity yet in the landscape. nTop acquired cloudfluid in 2025 to expand into CFD. I think this is directionally important. It shows how the boundary between design and simulation may keep blurring.