Billion-Dollar Valuation: AI Math Company Begins Verifying Code

Brian Burns posted on X: How can companies doing AI math get billion-dollar valuations from top venture capitalists? Math is cool, but how do you make money?

His answer is automated formalization. This technology was originally designed to let AI translate mathematical theorems described in natural language into a formal language that proof assistants like Lean can check. Verifying code follows the same logic: translate program behavior and specifications into a formal language and have a machine prove that they are consistent.

The speed at which AI writes code has already exceeded the speed at which humans can review it. With the cybersecurity situation tightening, all production software in the future may be required to pass formal verification. How big is this market? The entire software industry.

Harmonic has already entered the game. Today it announced that it is using AI to verify code, starting with C and Verilog. The exact quote: “We're bringing Aristotle to code.” Bringing Aristotle to code.

Harmonic paired this line with a video.

C is a general-purpose programming language, while Verilog is a hardware description language. One is the foundation of software, the other the foundation of chips. Harmonic is covering both right out of the gate—a big step.

Formal verification is different from testing: testing looks for bugs across a finite set of inputs, while formal verification uses mathematical proof to show that a program conforms to its specification in all cases. It used to be mainly used in places where mistakes cannot be tolerated, such as chip design and aerospace. Harmonic wants to turn it into an automated quality check in the AI code-writing pipeline.

Reactions in the tweet's comments were mixed. One person said, “Shipping Verilog on day one is a flex.” Another was willing to pay for it: “Half of what my agent writes runs, but something always feels off. I’d gladly pay for a tool to catch it before I do.”

Others poured cold water. One user posted a detection screenshot showing that 79% of the content was AI-generated, with the caption: “You can write a script to detect it yourself. AI isn’t that good yet.”

The proof capabilities in the hands of math companies are moving from the ivory tower into the software supply chain. The next thing to watch is whether it can keep up with the speed at which AI writes code.