AI formalized a graduate math textbook in one week for less than it costs to hire one expert

What happened

An AI system automatically converted a 500-page graduate-level math textbook into formal code. This process took one week and cost about the same as hiring a human expert team.

500 pages of graduate-level textbook formalized

130,000 lines of formal code generated

5,900 Lean declarations in formalized code

30,000 parallel AI agents used

1 week to complete formalization

Why this matters

Formalizing textbooks means translating every definition, theorem, and proof into a computer-readable language. This makes the math verifiable by machine, eliminating human error and ambiguity. It also means that AI systems can now learn directly from highly structured, verified knowledge, rather than just raw text.

30,000 workers finished a job in one week that would cost less than the salary of the one person who could have done it alone.

Who wins, who loses

who wins Any organization that needs mathematical proofs verified and previously could not afford the labor, who now gets it quietly reclassified from luxury to infrastructure.

who loses The small community of human experts whose specific value was that formal mathematics verification was too difficult and too expensive to outsource, counting on that scarcity to remain.

also Chip designers, cryptographers, and anyone whose product ships with a proof attached.

Jargon decoder

Lean a computer programming language designed for writing and verifying mathematical proofs

formalization converting mathematical proofs into code a computer can check for errors

multi-agent software engineering many AI systems working together on shared code at the same time

The signal

Why this hasn't landed yet

The result requires understanding what Lean is and why formalization is hard, which is two prerequisites most science journalists do not have, so the story gets filed under 'AI does math thing' rather than 'the economics of mathematical certainty just changed.'

What happens next

Mathematics publishers and university presses are next. Once one graduate textbook is formalized open-source in a week, the question of why the rest of the catalog has not been formalized becomes difficult to answer politely. Expect the first institutional formalization mandate within two to three years.

The catch

The benchmarking community will spend the next six months locating the specific theorems where the 30,000-agent system produced formally valid but mathematically trivial proofs, which is how every large-scale automated verification claim has been walked back since the early interactive theorem prover era.

The longer story

The longer arc

Automated theorem proving has been a formal research program since the 1950s, with milestones arriving roughly once per decade. The four-color theorem machine proof in 1976 and the Flyspeck project's complete formalization of the Kepler conjecture in 2014 both drew the same reaction: impressive but narrow. This is the first result at textbook scale, which is a different category of claim.

Part of a pattern

This is part of a pattern of multi-agent AI systems crossing labor-cost thresholds in high-precision domains previously considered safe from automation, following similar cost crossovers in legal document review and clinical coding over the past eighteen months. The pattern is: the domain held until the cost comparison became impossible to ignore.