The bracket took 20 minutes to model. Not a complex part—just a simple mechanical bracket, the kind hardware engineers create dozens of times a week. Twenty minutes of waiting for geometry to rebuild, for constraints to resolve, for the program to catch up with the designer's intent.
Michael Baron watched this happen at a hardware company where he worked as a mechanical engineer, and it gnawed at him. The math behind that bracket? Trivial by modern computing standards. Why was it taking longer to model a bracket than to brew a pot of coffee?
His answer: build an entirely new CAD system from the ground up.
Baron, who just emerged from Y Combinator's Winter 2026 batch with Aurorin CAD, hasn't built another plugin or AI wrapper. He wrote a new parametric CAD kernel—the computational engine doing the mathematical heavy lifting under the hood—designed specifically for contemporary processors and graphics cards. The legacy tools everyone uses, he argues, are running on technology that predates the internet. And it shows every time an engineer waits for a large assembly to open.
"Parts that take over 20 minutes in SolidWorks or NX can be created in seconds," according to Aurorin's Y Combinator launch materials. Bold claim. But Baron has the credentials to at least get a hearing.
Rocket Science as Preparation
Baron isn't your typical solo founder taking a swing at enterprise software. He interned three times at SpaceX—working on Raptor engine combustion simulation, Dragon spacecraft guidance and control, then Starshield flight software. After that, he spent time at Apple optimizing GPU driver performance. Which is to say, he knows both the hardware engineering pain points and the computer science fundamentals required to address them.
The technical bet is ambitious but straightforward: modern CPUs and GPUs are orders of magnitude more powerful than what was available when CATIA and SolidWorks were originally architected. Aurorin's custom parametric B-Rep (boundary representation) kernel was built to exploit that power. Baron positions this as fundamentally different from competitors layering AI features onto 1980s-era foundations.
Speed matters in hardware. Design iteration cycles directly determine how quickly companies can prototype, test, and ship. Lag time in CAD translates into lag time everywhere downstream. Baron understands those pressures—he's lived them.
AI That Doesn't Feel Bolted On
Aurorin integrates an AI agent directly into the design workflow, blending traditional UI elements with chat-based commands. The agent can generate constrained sketches, apply geometric dimensioning and tolerancing (GD&T) through model-based definition, and automatically produce engineering drawings. Baron describes the approach as "AI native," which in practice means the system was designed with AI capabilities baked in rather than retrofitted later.
The software is available now for Mac and Windows via free trial download at aurorincad.com, though details on pricing remain sparse. Third-party directories have floated speculation about plans starting around $49 monthly, but Baron hasn't published official pricing. Perhaps he's still figuring out what the market will bear.
Walking Into a Buzzsaw
The timing cuts both ways. On one hand, frustration with incumbent CAD tools is real and well-documented across engineering forums. On the other, every major player launched AI features between 2025 and early 2026.
PTC introduced Onshape AI Advisor, powered by Amazon Bedrock, with plans for agentic workflows on the roadmap. Siemens debuted Design Copilot NX featuring natural-language interaction. Dassault Systèmes—SolidWorks' parent company—rolled out AI assistant Aura in 2025, then expanded to multiple AI companions in 2026. Autodesk pushed updates to Fusion 360's Assistant for prompt-based operations.
Baron's counter-argument? These are retrofits, constrained by architectures conceived before GPUs were even a thing. Building AI into a system designed for 1980s hardware is fundamentally limited, he suggests. Aurorin's kernel was written with performance and AI integration as first principles.
The company has raised $500,000 via convertible note, according to CB Insights, with Y Combinator among the investors. Baron is running it solo from San Francisco—a bold choice for enterprise software—with Jared Friedman as his YC partner.
The Unglamorous Questions
Since launch, Baron has engaged actively with the CAD community on social media, posting tutorials and fielding questions. Some are predictable adoption concerns. Others cut to technical fundamentals: How does the kernel handle complex boolean operations? What about near-coincident geometry edge cases that trip up even mature systems? File format interoperability with incumbent tools?
These aren't sexy questions, but they're the ones that determine whether a new CAD platform scales beyond early adopters. Large assemblies, robust import/export capabilities, integration with existing PDM (Product Data Management) and PLM (Product Lifecycle Management) systems—these matter as much as speed benchmarks when hardware teams evaluate switching tools. Maybe more.
Baron's launch materials claim teams are already switching to Aurorin, though no customer names have been disclosed. That's typical for early-stage enterprise software, but it also means the proof remains largely anecdotal.
Long Odds, Clear Differentiation
Can a solo founder out-execute billion-dollar incumbents with massive R&D budgets and decades of customer relationships? History says it's unlikely. Then again, history also shows that incumbent advantages matter less when the underlying technology shifts—and AI plus modern hardware arguably represents exactly that kind of shift.
Baron is betting that frustration with legacy performance, combined with a genuine appetite for AI-first design workflows, creates an opening. The kernel-level rebuild at least offers technical differentiation rather than another layer of abstraction on top of someone else's engine.
Whether it's enough remains an open question. But the founder who once helped simulate rocket engines probably has some tolerance for hard problems and long odds. Twenty-minute brackets won't fix themselves.
