Macadamia Is Sending AI Agents Into CAD Files, and Engineers Should Pay Attention

The Macro: Engineering Is the Last AI Holdout

If you look at where AI agents are being deployed right now, the pattern is obvious: text-heavy, screen-based work. Customer support. Sales emails. Legal document review. Code generation. Marketing copy. These are all jobs where the inputs and outputs are words on a screen, and that’s where AI has the easiest time showing value.

Engineering is different. Mechanical engineers work with CAD files, P&ID diagrams, simulation outputs, and technical specifications that have to be physically correct, not just linguistically plausible. A chatbot that writes a slightly wrong email is an inconvenience. An AI agent that approves a slightly wrong structural calculation is a liability. The stakes are higher, the tolerance for error is lower, and the domain knowledge required to be useful is deeper.

This hasn’t stopped companies from trying. Autodesk has been adding AI features to Fusion 360. PTC has AI capabilities in Creo. Siemens is building AI into its Xcelerator platform. But these are all feature additions to existing products, not purpose-built AI agents that can independently review and modify engineering work. The difference matters. An AI feature that suggests a bolt size is useful. An AI agent that can review an entire assembly for design errors and fix them is a different category of product.

The engineering services market is massive. EPC (Engineering, Procurement, Construction) firms collectively generate hundreds of billions in revenue. These are companies with thousands of engineers doing repetitive technical work: checking drawings, verifying specifications, cross-referencing standards. It’s high-value work done by expensive people, and a significant portion of it is verification and compliance rather than creative design. That’s exactly the kind of work AI agents should be good at.

The Micro: An AI Mechanical Engineer Named Cooper

Macadamia’s product is called Cooper, and the positioning is bold: “an AI Mechanical Engineer that detects and fixes design errors.” Not “assists with” or “helps find.” Detects and fixes. That’s a strong claim for a domain where errors can have physical consequences.

The company integrates with 20+ engineering tools including AutoCAD, Ansys, Revit, MATLAB, Jira, and Slack. The breadth of integrations tells you something about their target customer. This isn’t for the solo mechanical engineer working in one tool. This is for the EPC firm that has dozens of engineers using different software across different project phases, where coordination errors between tools are as common as design errors within them.

Security positioning is prominent. SOC 2 Type 2 certification and GDPR compliance are listed on the website, which matters for enterprise engineering firms that deal with proprietary designs and classified projects. Getting SOC 2 Type 2 as an early-stage company is a real investment of time and money, and it signals that enterprise sales are the primary go-to-market, not self-serve developer adoption.

Abel Van Steenweghen is the CEO. He has a master’s in Computer Science from TU Delft and has built AI startups before. Brecht Pierreux is the other founder. His background is in mechanical engineering from ETH Zurich and Caltech, and he’s worked at SpaceX and NASA’s Jet Propulsion Laboratory. That combination is specific and credible. You have someone who understands AI systems and someone who understands what mechanical engineers actually do. The JPL and SpaceX credentials carry weight in a domain where “does this person understand my work” is the first question every potential customer asks.

They’re part of YC’s Winter 2025 batch. The company is hiring for a Founding GTM Lead and a Founding Full-Stack Engineer, which suggests they’re in the phase of going from “product that works” to “product that sells.” Pricing is enterprise-only, contact for details, which is standard for this market.

The Verdict

I think Macadamia is one of the more interesting vertical AI plays I’ve seen recently. Going after engineering rather than knowledge work is contrarian in a good way. The market is large, the incumbents are slow, and the founders have exactly the right combination of AI and domain expertise.

The risk is the classic vertical AI trap: the product has to be right. Not “pretty good.” Not “better than nothing.” Right. An AI agent that catches 95% of design errors sounds impressive until you realize the 5% it misses could be the ones that cause a building to fail inspection or a part to break under load. The accuracy bar in engineering is not like the accuracy bar in marketing copy. Engineers will test this product aggressively, and they will find the failure modes.

In 30 days, I’d want to see case studies from actual EPC firms using Cooper on real projects. Not demos, not sandboxed examples. Real drawings, real errors caught, real time saved. In 60 days, I’d want to understand the error rate. What percentage of Cooper’s suggestions are accepted by engineers, and what percentage are overridden? That ratio tells you everything about whether the product is trusted or merely tolerated. In 90 days, the question is expansion. Can they move beyond mechanical engineering into electrical, structural, or civil? Each discipline has its own standards, its own tools, and its own failure modes. The broader they can go, the bigger the company. But going too broad too fast in a domain where accuracy is everything could be fatal. This is a company worth watching closely.