AI Transforms Hardware Design: Software Engineers Architect Systems, Hardware Engineers "Vibe Code"

This conversation, "Vibe Coding Hardware," from Naval, featuring Guillermo Rauch, Blake Scholl, and Max Hodak, reveals a profound shift in how we build physical products, driven by the rapid advancement of AI and software engineering principles. The core thesis is that the traditional, siloed, and often manual workflows of hardware development are becoming obsolete. Instead, software engineers are increasingly architecting the systems, enabling hardware specialists to "vibe code" their components with unprecedented speed and iteration. The non-obvious implication is that this democratization of complex design, while accelerating innovation, also concentrates power in those who can effectively wield AI for synthesis and verification, potentially creating new forms of competitive advantage and market consolidation. Anyone involved in building physical products, from startups to established enterprises, will gain an advantage by understanding how AI is fundamentally altering the design, iteration, and manufacturing processes, moving beyond mere automation to true generative design and system integration.

The Software-Defined Hardware Revolution: Beyond Spreadsheets to Systems

The traditional image of hardware engineering often conjures up complex spreadsheets, manual data transfers, and siloed expertise. Blake Scholl of Boom Supersonic highlights this stark reality: "There's a lot of engineering, hardware engineering, that happens in Excel spreadsheets on engineers' laptops in a silo. It's very complex spreadsheets, sometimes like VB script code. All of this is actually software, but it's treated as if it's not software. There's no source control, there's no automated testing." This manual, disconnected approach drastically limits iteration speed and introduces errors. The core insight here is that these "software-like" processes, when not treated as actual software with proper engineering practices, become bottlenecks.

The shift, as articulated by Scholl, is towards software engineers architecting the systems that hardware engineers then "vibe code." This isn't just about faster design; it's about a fundamental change in productivity. Consider the example of designing a turbine blade. Classically, this involves a single engineer spending a day on one blade for one analysis, accounting for thermal expansion and structural integrity across cold and hot states. With software frameworks automating these conversions and analyses, two engineers can now design an entire jet engine. This illustrates a layered consequence: immediate efficiency gains (reduced design time) lead to downstream effects like drastically lower development costs, enabling smaller teams to tackle massive projects and accelerating the pace of innovation in the aerospace sector.

"The idea was we could reduce the cost of iteration, but it was slow going because we could never get enough, we could never afford enough software engineers. What we've gotten into is this mind-blowingly different model where the software engineers actually create the architectures because they understand systems, they understand algorithms, they understand division of concerns. Then the hardware engineers can vibe code their pieces because they know about hardware engineering."

-- Blake Scholl

This dynamic has profound implications for enterprise software. Scholl notes that the need for specialized hardware collaboration tools is diminishing because companies can now "internally... just coding the right things that you need at any given time." Spreadsheets, once a workaround for custom software, are now being supplanted by more robust, scriptable environments like Python, allowing for "believable simulations." The consequence here is a move away from off-the-shelf, generic tools towards bespoke, internally developed solutions that are perfectly tuned to a company's specific hardware engineering needs. This creates a competitive moat for companies that invest in building these internal capabilities, as their unique workflows become harder for competitors to replicate.

The Intelligence Imperative: Why You Always Want the Smartest Model

A recurring theme is the relentless pursuit of intelligence, particularly in AI models. The conversation touches on the strategic implications of open-source versus proprietary models, with a particular focus on China's approach. The argument is made that China's investment in open-source AI is partly a strategy to leverage its hardware superiority. By generating software "on demand," they aim to overcome any disadvantages against Silicon Valley. However, the underlying intelligence of these models is crucial.

The critical insight is that, when it comes to complex problem-solving, especially in areas like coding, there's no substitute for the most intelligent model available. The speaker posits, "I think intelligence is an unalloyed good. You always want more intelligence, and when these models make a mistake, you don't know it. It's always cheaper than a real person and real time, so you'll just use the most intelligent model available." This leads to a potential oligopoly in AI, where a few frontier models dominate because their superior intelligence, even at a higher cost, is more efficient than relying on less capable models that might require more iteration or produce incorrect outputs.

"I always want the most intelligent programmer. I always want the most correct answer. I always want the best judgment. Given the amount of leverage that I'm going to pour into it through capital and code and people and marketing, I want to make the right decision every time."

-- Naval

This preference for maximum intelligence has a cascading effect. Companies will gravitate towards the most capable AI for critical tasks like coding, simulation, and design. This means that while cheaper, less intelligent models might be useful for simpler tasks (like industrial production, support, or browser automation, where Gemini or Chinese models might suffice), the frontier of innovation will be powered by the most advanced, often proprietary, models. The consequence of this is that companies that can afford and access these top-tier models gain a significant advantage in speed, accuracy, and innovation, potentially out-competing those who rely on less capable AI. This creates a dynamic where falling behind in AI capability means falling behind in the ability to generate everything.

Vertical Integration and the "Hands-On" Future of AI

Max Hodak's perspective on vertical integration and extreme urgency introduces another layer of consequence: the necessity of building capabilities when they cannot be bought. His team's preference is to buy components like PCBs because they are readily available. However, when innovation requires integration so deep that it approaches "a single block of covalently bonded matter," they must build it themselves. This leads to owning specialized facilities, like a captive MEMS foundry.

The surprising impact of AI, ironically, has been in regulatory interactions. AI can now rapidly process thousands of ISO standards, trace compliance requirements, and generate documentation, tasks that previously consumed months of a regulatory team's time. This immediate payoff from AI in managing complexity is significant. However, Hodak points out a crucial boundary: "the software still needs hands." Even as AI becomes smarter, its ability to make things--to translate digital designs into physical reality--remains a limitation. This highlights a future where AI enhances human capabilities but doesn't entirely replace the need for physical execution.

"It's going to be smarter than us, but if it can't make things, then those are real, real boundaries. So we've instrumented our foundry as well as many other parts of the company in ways where, as these models get better, that should show up pretty immediately in things like the cell engineering that we're doing and the material science that we're developing."

-- Max Hodak

This leads to the idea that humans are evolving into "verifiers." Just as AI models are trained with verification data, humans are now tasked with verifying the outputs of AI, whether it's code, legal documents, or engineering designs. This elevates the role of engineers and lawyers from junior-level execution to senior-level validation and strategic thinking. The immediate consequence is that tasks previously done by junior staff are automated, while experienced professionals are freed up for higher-level work. The long-term implication is that companies that can effectively integrate AI for verification and automate junior tasks will see significant productivity gains, creating a competitive advantage for those who embrace this human-AI collaboration. The challenge lies in building confidence in these AI-generated outputs, moving from reading every line of code to trusting the evaluators and tests that confirm safety and correctness.

Key Action Items

• Adopt Software Engineering Practices for Hardware: Over the next quarter, audit existing hardware design workflows. Identify processes currently managed in spreadsheets or manual transfers and begin migrating them to version-controlled, testable software frameworks.
• Invest in Internal AI Capabilities: Within the next 6-12 months, identify critical AI tasks (e.g., code generation, simulation, regulatory analysis) and evaluate the use of frontier models. Prioritize access to the most intelligent models for high-leverage activities, even if it means higher per-token costs.
• Develop AI-Powered Verification Systems: Over the next 6-18 months, build automated testing and simulation harnesses specifically designed to validate AI-generated hardware designs or code. This moves beyond manual code review to system-level confidence.
• Explore Vertical Integration for Critical Components: Within 12-24 months, assess if key components or manufacturing processes are limiting innovation. If off-the-shelf solutions are insufficient, investigate building captive capabilities, especially where AI can enhance specialized manufacturing.
• Train Teams on AI Verification and System Architecture: Immediately begin upskilling engineers to transition from junior execution to senior verification and architectural design roles, focusing on understanding the consequences of AI-generated outputs.
• Embrace "Vibe Coding" for Hardware Specialists: Over the next quarter, empower hardware engineers with the software tools and architectures created by software engineers, allowing them to iterate rapidly on their specific domain expertise.
• Evaluate Open-Source vs. Frontier Models Strategically: Continuously assess the cost-benefit of using cheaper, open-source models for lower-stakes tasks versus investing in premium, frontier models for mission-critical innovation, recognizing that intelligence is an unalloyed good for complex problems. This pays off in 12-18 months through reduced errors and faster breakthroughs.