A16z general partner Aaron Price-Wright speaks with Alex Bowden, co-founder and CEO at Unlimited Industries, and Davide Asnaghi, CEO at Diode Computers. Alex's company uses AI to vertically integrate design, engineering, procurement, and construction for large infrastructure projects, while Davide's team applies AI to design and manufacture custom circuit boards faster than ever achieved in the United States.

The conversation explores physical world AI applications at two different scales - from micro-level circuit board design to macro-level construction automation. Both companies are working to transform traditionally slow, manual industries by treating everything as code and leveraging AI models' existing programming capabilities rather than requiring massive domain-specific training datasets.

The Vision for Fully Automated Construction and Electronics

Alex envisions construction automation happening in two phases: first, AI generates optimized design packages from site specifications and requirements, then autonomous robotics handles physical construction including earth movers and humanoids.

"You will literally feed in a site, a bunch of different requirements about what you're trying to build and AI is going to explore tens of thousands of different permutations about how optimally design that facility at a button click" - Alex

Davide targets 2-year timeline for circuit board design automation, focusing on manufacturability constraints: "If the design is constrained, you can manufacture it at 100% automation today. The robots are already here."

Solving the Data Problem Through Code-First Approaches

Both companies bypass traditional AI training data limitations by reframing physical design problems as code generation tasks that leverage models' existing programming knowledge.

"We basically built a compiler that gives the model enough hints that it feels like it's writing a Python program instead of designing a circuit board" - Davide

Alex's approach creates parametric relationships where "everything's just an updated variable" rather than starting over when design requirements change mid-project.

Diode's strategy generates validated design blocks that become training data for future models, creating "compounding interest" in AI capabilities over time.

Overcoming Industry Resistance and Incentive Misalignment

Construction industry incentives discourage technology adoption because project financing prioritizes stable returns over innovation: "There's no upside. No one actually wins from that environment" - Alex

Diode sells end products rather than software tools to avoid high switching costs: "You don't want to convince people to buy your software. You want to convince people to buy the end product."

Both companies found it easier to teach domain experts AI tools rather than teaching software engineers the physical domain expertise required.

Vertical integration allows clean interfaces with traditional industry players rather than forcing adoption of new tools and workflows.

The Manufacturing Automation Gap and Skilled Labor Crisis

Electronics manufacturing achieves 80% automation through surface mount technology, but the remaining 20% manual work prevents full US competitiveness with Asian manufacturing.

Skilled trades face severe shortages with electrician salaries now exceeding Silicon Valley software engineer compensation in many markets.

"Microsoft employed a third of the electricians in the state of Georgia" when building a single data center, illustrating infrastructure workforce constraints - Aaron

Companies increasingly turn to modular manufacturing to concentrate scarce skilled labor, even at cost premiums over traditional construction methods.

Simulation, Physics, and the Path to Full Autonomy

Both companies use simulation primarily as training tools rather than inference-time verification, with Davide noting "simulation needs to be a training tool and then you need to get physics to tell you you're right or wrong."

Alex emphasizes designing systems for full autonomy from the start: "Making sure that you design a system to actually be fully autonomous and to not be human in the loop... it's driven a very different architecture."

The data scarcity problem may resolve through either collaborative open sourcing or companies generating sufficient proprietary datasets through end-to-end operations.

Physical constraints in both domains are highly structured and rule-based, making them more tractable than open-ended AI problems despite limited training data.

Humanoids, Robotics, and the Future of Physical Work

Alex sees humanoids as essential for construction sites due to manufacturing efficiency benefits: "Centralizing around a design and mass manufacturing that design... outweighs the nuanced custom efficiency."

Davide focuses on existing robotic arms with improved computer vision rather than humanoid form factors for electronics manufacturing applications.

Both anticipate robotics handling dangerous work like mining while humans focus on higher-level design and oversight roles.

The transition requires preserving tacit knowledge from aging skilled workers while building AI systems that can eventually encode manufacturing intuition and best practices.

Resources Mentioned

Mastering Compound Interest From Basics to Advanced Strategies

's strategy generates validated design blocks that become training data for future models, creating "compounding interest" in AI capabilities over time. Overcoming Industry Resistance and Incentive M