What happened
Axiom Math, a startup valued at $1.6 billion, announced the formal verification of the Stacks Project, a comprehensive open-source library of algebraic geometry and related mathematical fields. Using a fleet of proprietary AI agents, the company translated approximately 7,654 pages of human-written mathematics into 16 million lines of code written in Lean, a programming language used for formal verification. This process ensures that every theorem and definition within the library is logically sound and consistent, as verified by a proof assistant program.
Axiom Math, founded in March 2025 by CEO Carina Hong, utilized its own AI agents to process the Stacks Project. The project, originally started by Johan de Jong in 2005, is considered a foundational text in modern mathematics, covering algebra, geometry, and number theory.
The verification process involved translating human-written proofs into Lean, a language that allows computers to check the logical validity of mathematical arguments line by line. The agents operated in parallel, identifying tasks and managing the complex, interconnected dependencies inherent in the Stacks library.
Axiom trained its models specifically for mathematical reasoning, moving beyond general-purpose coding capabilities. According to the company, the agents were initially guided by human Lean experts but eventually developed their own 'sense of taste' for formalizing proofs.
The company raised $200 million in a Series A round led by Menlo Ventures in March 2025. The current valuation of the startup is reported at $1.6 billion.
Source details: techcrunch.com
Why it matters
The formalization of the Stacks Project addresses a significant gap in mathematical research, specifically within algebraic geometry, where researchers previously lacked a verified foundation to build upon. By converting these proofs into machine-readable, verified code, Axiom Math provides a reliable 'substrate' for future AI agents. As AI increasingly generates software and scientific research, the risk of 'slop'—or logically flawed, unverified output—grows. Axiom’s approach suggests that by grounding AI agents in a verified, interconnected library of mathematical truths, developers can ensure that future AI-generated systems are built on a foundation that is logically consistent rather than prone to error or paradox. This shift from 'vibe coding' to formal verification represents a strategic attempt to standardize the reliability of AI-generated mathematical and software infrastructure.
Formal verification is a rigorous process typically used in high-stakes engineering, such as chip design or critical software development, to eliminate bugs. By applying this to the Stacks Project, Axiom is attempting to create a 'source of truth' for AI agents.
The 'black hole' in formalized mathematics, particularly in algebraic geometry, has historically hindered researchers from building on existing work with absolute certainty. Axiom’s library aims to remove this barrier, allowing future proofs to be built on top of verified results rather than starting from scratch.
The project serves as a test case for whether AI agents can handle large-scale, delicate, and highly structured data without introducing logical errors. If successful, this methodology could be applied to other scientific domains, potentially reducing the prevalence of AI-generated inaccuracies in technical fields.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Which component of an AI application is the machine-learning model itself?
What to watch next
Axiom Math CEO Carina Hong has issued a public challenge to other coding companies to pursue formalization at this scale. Observers should monitor whether this move triggers a broader industry trend toward integrating formal verification into AI development workflows. Additionally, it remains to be seen how effectively this specific library of Lean code will be adopted by the broader mathematical and software engineering communities to verify new, independent research. The company's ability to maintain the integrity of these 16 million lines of code as new mathematical discoveries are added will be a key indicator of the project's long-term utility.
The primary metric for success will be the adoption rate of the Axiom-formalized Stacks library by the academic and engineering communities. If researchers begin using this code to verify their own new work, it will validate the company's foundational approach.
Industry response to Hong's challenge will be significant. If major AI coding firms adopt similar formalization strategies, it could signal a shift in how AI-generated code is audited and validated for enterprise use.
The company has not disclosed specific pricing or access conditions for the broader developer community beyond the announcement of the project's completion. Future updates regarding the accessibility of these tools for third-party developers remain an unknown.