Two Trillion Lines of Legacy Code, a Vanishing Workforce, and a New Hardware Frontier
In a world where AI can effectively write software code, software organizations addressing real-world challenges face two distinct but equally urgent code challenges.
The first challenge is the need for modernization of the code itself. Currently, vast inventories of legacy code are written in heritage programming languages that organizations need to migrate to current-generation programming languages like Java or Python. These projects are slow, expensive, and risky, typically requiring specialists who are becoming harder to find and even harder to retain.
The second challenge is language translation for modern workloads. An engineer building a computer vision model typically writes in Python or PyTorch, which are languages optimized for that work. But when the destination is a drone, an FPGA, or an edge device, that code must be translated into the low-level machine instructions the chip understands. This translation, verification and validation work is a bottleneck that can take weeks or months depending on complexity of the project and require a class of embedded systems expertise that is genuinely scarce today. The developers who understand these heritage languages are retiring or have already left the workforce, and no one is coming behind them — creating a widening gap between the volume of code that needs to be translated and the human expertise available to do it.
Speed matters in both cases, but trust is what makes speed valuable. An automated translation that produces output you cannot verify only introduces a new source of risk. This is why Code Metal's combination of AI-driven translation and formal verification is foundational rather than incremental. If the output cannot be mathematically proven correct, the system tells you so. It does not hallucinate an answer. In mission-critical environments, that distinction is the difference between a tool and infrastructure. The tools that transformed software development for cloud environments were not designed for the harder problem downstream: translating verified code across heterogeneous hardware, proving semantic equivalence, and deploying to edge environments where there is zero margin for error. A new infrastructure layer is required. That is what Code Metal is building.
Proven Traction in Mission-Critical Environments
Code Metal is the leading AI platform automating the translation and deployment of code for mission-critical industries including defense, semiconductor, automotive, and aerospace. The company's proprietary approach enables customers to port and optimize code across any chip or edge environment while ensuring it is verified, validated, and production-ready, enabling engineering teams to accomplish in hours what had previously taken months or longer.
Code Metal has earned the trust of the most demanding customers in the world precisely because it was built for this problem from the ground up. When the U.S. Air Force needed a modern approach to its most complex software deployment challenges, no existing platform offered a viable solution, so Code Metal built the trusted solution. The Air Force is not an outlier. At defense primes like L3Harris and RTX, translating code for a new hardware target has historically been a weeks-long effort, performed by specialists whose ranks thin every year. These organizations need mathematical certainty that the code they deploy will behave exactly as intended when it matters most – and that is precisely what Code Metal delivers.
A World-Class Team
Peter Morales, an AI research scientist with a decade of experience across BAE Systems and MIT Lincoln Laboratory, founded the company with a team that understands this problem at a level few others can match. At MIT, Peter was a foundational member of the AI Technology group, leading automation and robotics projects for the Air Force including algorithms protecting the F-35 and the U.S. Capitol from drone threats. His co-founders bring equally rare credentials: a founding author of MATLAB, formal verification engineers from NASA, and veterans of the MATLAB compiler team. These are not generalist AI engineers retrofitting large language models onto an old problem. They are the people who built the tools the industry has relied on for decades and who understood better than anyone why those tools were no longer enough.
The underlying capabilities — static analysis, code verification, software emulation, schematic generation — are complex, multi-step processes built over years and not easily replicated. Code Metal’s platform is not a diagnostic tool or a copilot. It deploys on-premise, configures to each customer's proprietary IP and hardware environment, and embeds directly into CI/CD workflows. Once integrated, it becomes foundational infrastructure — a permanent layer in how organizations deploy software to hardware.
Smith Point’s Conviction
Our conviction that Code Metal is grounded in evidence from the field. The platform dramatically compresses engineering timelines, embeds deeply into customer workflows and becomes indispensable infrastructure for mission-critical software deployment. Across defense and commercial customers, engagements expand organically across divisions once the platform is in place, the high ROI quickly becomes apparent, and adoption follows.
The forces driving demand are structural, not cyclical: the urgent need to modernize vast inventories of legacy code that organizations can no longer afford to maintain; edge AI moving from cloud into physical systems; regulatory mandates around memory-safe programming and DevSecOps across the defense establishment; and a workforce crisis that makes automated code translation existential for every customer Code Metal serves. These tailwinds do not reverse.
We are proud to deepen our partnership with Peter and the entire Code Metal team and to help scale this exciting business.
Learn more: Read Code Metal's Series B press release here.
