Lead – On Wednesday, AxiomMath announced that it has formally verified the Stacks Project—a 7,654‑page, open‑source compendium of algebra, geometry and number theory—by converting it into roughly 16 million lines of Lean code. The $1.6 billion startup says the effort creates a trusted mathematical substrate that AI agents can use to avoid “slop” in future research and software development.
Background: AI’s Flood of Unchecked Mathematics
According to TechCrunch, the pace of AI‑generated mathematics has accelerated dramatically, with OpenAI alone releasing 719 manuscripts claiming solutions to some of the field’s hardest problems in a single month. The article notes that even computer‑based checks can miss errors, prompting concerns that mathematics could become the next domain plagued by unreliable AI output.
AxiomMath’s Vision
According to TechCrunch, AxiomMath is betting that mathematics can sidestep this fate by building a “verified foundation” that AI agents can rely on when writing software and, eventually, conducting scientific work. The company’s stated goal is to provide a reusable, machine‑checked library that prevents AI from producing contradictory or unsound results.
The Stacks Project Verification
The Stacks Project, as described by TechCrunch, is one of the largest human‑written bodies of mathematics. Columbia mathematician Peter Woit has compared it to the historic Bourbaki books, calling it “nothing else really comparable.” The project was started in 2005 by Johan de Jong as an open‑source reference, and contributors have expanded it from elementary algebra to advanced topics across algebraic geometry and number theory.
According to TechCrunch, AxiomMath’s AI fleet translated the entire library into about 16 million lines of Lean code—a programming language designed for formal proof verification. In this workflow, mathematicians rewrite proofs as Lean code, and a proof‑assistant checks each line for logical consistency. If the proof passes, the result is considered mathematically sound.
The article explains that formal verification is already used in engineering to confirm chip designs and, in France, to catch bugs in software that calculates family benefits. For the Stacks Project, verification required preserving the intricate web of definitions and theorems that build on one another. According to TechCrunch, AxiomMath had to keep every connection intact across the massive code base, likening the task to maintaining a “family tree” of mathematical results.
Technical Approach
According to TechCrunch, the company gave its AI agents a “loose structure” and allowed them to operate in parallel. Technical staff member Ashvin Swaminathan said the agents imposed a hierarchy on the work but did not dictate every individual task, letting the agents identify work they could do concurrently.
The article notes that AxiomMath also trained its own models specifically for mathematics, a step that “most other verification startups haven’t taken,” according to TechCrunch. Initially, a human Lean expert guided the agents, teaching them a “human sense of taste.” Over time, the agents reportedly internalized this taste on their own.
Significance for Formalized Mathematics
According to TechCrunch, the verification fills a major gap in formalized mathematics, especially in algebraic geometry, which has previously suffered from a “huge black hole” of unformalized results. Swaminathan is quoted as saying that researchers now have a solid base on which to build future Lean‑based proofs.
The article also cites Fields Medalist Terence Tao’s observation that “the more formalized proofs you have, the more formalized proofs you can make,” underscoring the compounding benefit of the effort.
Business Context
According to TechCrunch, AxiomMath was founded in March 2025 by Carina Hong, a 25‑year‑old former Stanford math Ph.D. student who left the program to start the company. In March of the same year, the startup raised a $200 million Series A round led by Menlo Ventures. The article reports that Hong described the challenge of preserving the “beauty” of Stacks as a key uncertainty at the outset of the project.
Broader Implications for Software Development
According to TechCrunch, AxiomMath believes that the verified mathematical library will eventually serve developers seeking bug‑free, secure code—especially as AI agents generate increasing portions of software. Swaminathan is quoted saying that mathematics is a “transferable skill” and that agents can read the Lean code, understand it, and construct new infrastructure atop it.
Hong is reported to view the parallel, self‑assigning workflow used for Stacks as a model for future AI work. She argues that AI agents need a well‑built, reusable knowledge base to avoid “slop” and that early engineering choices will shape how far the frontier can be pushed. Hong is quoted as issuing a public challenge to “all the coding companies” to attempt formalization at this scale.
Analysis (Opinion)
The verification of the Stacks Project represents a noteworthy milestone for both the formal methods community and the emerging field of AI‑assisted mathematics. By converting a massive, collaboratively authored corpus into machine‑checked Lean code, AxiomMath demonstrates that large‑scale formalization is technically feasible when AI agents are given structured autonomy. If the company can successfully extend this approach to newer research, it could provide a reliable scaffold that mitigates the risk of AI‑generated mathematical errors—a concern highlighted by the recent surge of AI‑written manuscripts. However, the long‑term impact will depend on broader adoption of Lean among mathematicians and the ability of AI agents to navigate the evolving frontier of mathematical discovery without sacrificing rigor.Conclusion
According to TechCrunch, AxiomMath’s completion of the Stacks Project verification marks a strategic step toward a future where AI agents rely on a formally verified mathematical foundation. The company’s blend of custom‑trained models, parallel task allocation, and extensive Lean code generation aims to create a reusable substrate for both mathematical research and secure software development. As AI continues to generate more content across disciplines, the success of AxiomMath’s approach could influence how the tech industry safeguards the correctness of automated outputs.
Source: TechCrunch
Frequently Asked Questions
What is the Stacks Project?**
The Stacks Project is an open‑source, collaborative reference covering algebra, geometry and number theory, spanning roughly 7,654 pages and regarded by some mathematicians as comparable to the historic Bourbaki books. **Q: How did AxiomMath verify the Stacks Project?** A: According to TechCrunch, the company used AI agents to translate the entire library into about 16 million lines of Lean code, a proof‑assistant language that checks each logical step for consistency. **Q: Why does AxiomMath consider formal verification important for AI?** A: The article states that AxiomMath believes a verified mathematical foundation will prevent AI agents from producing contradictory or unsound results, thereby reducing “slop” in future AI‑driven mathematics and software development.