MIT Lincoln Lab. Microsoft HoloLens. BAE Systems and the F-35. Then one problem that wouldn't let go: AI-generated code that nobody could trust. So he built Code Metal to fix it.
There is a moment in every engineer's career when a problem becomes personal. For Peter Morales, it happened at BAE Systems, somewhere in the gap between Python code written by AI and the F-35 jet that was supposed to run it. The math didn't add up. The radar-countering ML algorithm worked on the laptop. On the aircraft, you couldn't be sure — and in defense, being unsure is not an option. Morales spent his days optimizing that code by hand, coaxing high-level software into hardware-grade correctness. He noticed: this is everywhere, and nobody is fixing it.
That recognition didn't lead directly to a startup. First came MIT Lincoln Laboratory, where Morales built AI systems for counter-drone defense — including technology designed to protect the U.S. Capitol. Then Microsoft, where he helped build computer vision for the HoloLens augmented reality headset. Same underlying problem in a different costume: making AI work reliably on constrained hardware, in real-world conditions, where failure isn't abstract. He collected the problem like others collect receipts.
He took one detour. Mint State Labs, a web3 creator-tools company, was his pivot into entrepreneurship. TechStars. Protocol Labs money. An acquisition by Orange Comet in December 2022, after which Morales served as CTO. The web3 wave crested and receded. But the founder toolkit — how to raise, how to build a team, how to sell something that doesn't exist yet — came back to Boston with him.
In 2023, Morales co-founded Code Metal with Alex Showalter-Bucher (another MIT Lincoln Lab veteran) and Octavian Udrea, a chief scientist with 20-plus years in AI research and 70-plus peer-reviewed papers. The founding insight: AI can write code faster than any human, but in regulated, mission-critical industries — defense, automotive, semiconductors, factory robotics — speed is worthless without proof. Code Metal combines large language models with neuro-symbolic AI to mathematically verify that translated code is correct before it ships. Write once in Python or MATLAB. Deploy anywhere: CPU, GPU, FPGA, drone, autonomous vehicle. And prove it works.
AI can generate code at unprecedented speed, but in mission-critical environments, speed without proof is not enough.- Peter Morales, CEO, Code Metal
The market validated the idea fast. A $3.45M pre-seed from J2 Ventures in 2023. A $13M seed led by Shield Capital in 2024, when Code Metal launched publicly. By November 2025, Accel — the firm that backed Facebook, Dropbox, and Atlassian — led a $36.5M Series A at a $250M valuation, with RTX Ventures and Bosch Ventures co-investing. The customers weren't just names on a press release: the U.S. Air Force, L3Harris, Raytheon, Toshiba, Bosch, and Boeing were writing real contracts worth tens of millions.
Then February 2026: $125M Series B led by Salesforce Ventures. Valuation: $1.25 billion. Total funding: $177.65M in under three years from founding. Morales went from engineer debugging F-35 code by hand to CEO of an AI unicorn that automates exactly that process at scale. The former CEO of Tableau, Ryan Aytay, joined as President and COO.
Morales is deliberately uninterested in vibe coding — the AI-assisted coding that generates a starting point quickly. "Vibe coding is all about explaining an initial idea in text, and generating code that will get you started developing your minimum viable product," he said. "This is not where most companies spend their time." Code Metal's target is the 90 percent of software work that happens after the prototype: bringing code to production, across hardware platforms, with correctness guarantees that satisfy military contracts and automotive safety standards. That's not a consumer feature. It's infrastructure.
The company's name says it all. Code Metal: taking high-level logic all the way to bare metal hardware. Morales didn't invent the problem. He just spent enough years inside it to understand exactly what the solution required — and built a team that could build it.
Code Metal's customer list is a shortlist of industries where software errors don't fail gracefully. The U.S. Air Force does not ship code that "probably works." Neither do Raytheon, L3Harris, or Toshiba. These are the contracts Morales landed before his Series A — and they're the reason Salesforce Ventures wrote a $125M check.
"AI can generate code at unprecedented speed, but in mission-critical environments, speed without proof is not enough."On verifiable AI code — Code Metal Series B announcement
"We are building the infrastructure that ensures software can be trusted before deployment."On Code Metal's mission — Series B press release, 2026
"Vibe coding is all about explaining an initial idea in text, and generating code that will get you started developing your minimum viable product. This is not where most companies spend their time."On distinguishing Code Metal from vibe-coding tools — CNBC interview, 2025
"That requires strong guarantees the code we're converting is accurate, compliant and working as expected."On production-grade AI code — TechInvestorNews interview, 2025
His LinkedIn tagline: "AI, RL, Defense | x-MSFT, x-MIT, TechStars." Seven words. Four entire careers. Maximum signal, zero fluff.
Code Metal uses neuro-symbolic AI — combining neural networks with formal logic — so the system can mathematically prove its own output is correct. It's not just AI writing code; it's AI proving the code it wrote.
Before defense AI unicorns, Morales built creator tools for NFT economies at Mint State Labs. The arc — defense AI to web3 to defense AI — is rare enough to be a conversation piece at any VC dinner.
Bosch and the U.S. Air Force are on the same customer list. Code Metal translates code for industrial robots on factory floors and for systems that help jets evade enemy radar — on the same platform.
"Code Metal" reflects the core mission: taking high-level code all the way down to bare metal hardware. It's not a metaphor. It's a specification.
Code Metal claims to cut development time by 50-75% for edge deployments. In defense and automotive, where a single software project can take years, that's not a productivity gain — it's a competitive moat.
Bachelor of Science in Physics — a hard-science foundation that shows up in Code Metal's formal verification approach.
Electrical & Computer Engineering — WPI's hands-on project-based curriculum built the hardware instincts Code Metal runs on.
Continued graduate-level study at MIT, deepening ties to the Lincoln Laboratory ecosystem where some of Code Metal's co-founders would later originate.
Translates high-level code (Python, Julia, MATLAB) into production-ready, hardware-optimized code — then mathematically proves the translation is correct using formal methods.
Targets CPUs, GPUs, FPGAs, drones, robots, autonomous vehicles. Write once, verify once, deploy anywhere. Reduces edge deployment time by 50-75%.
Combines large language models with traditional compiler technology and formal methods. The result: AI-generated code with mathematical correctness guarantees — not just plausible output.
