By Hamza Faisal

Google says its AI coding agents have moved from writing software to doing mathematics. In a research post on Friday, Google Research described how teams of specialist Gemini 4 agents working inside Antigravity, its AI coding environment, helped prove a conjecture that had stood open since 2013 — and produced machine-checked formalisations of famous pure-mathematics results.

The system is called /teamwork. It splits a hard problem between agents with separate roles — producers that draft solutions, critics that look for flaws, and auditors that check the result from a clean context — and uses objective checkers, such as the Lean 4 proof language, to verify each step automatically. Google says the tools are now open to outside researchers, with 10,000 academic researchers offered six months of free access to its highest tier.

A 2013 conjecture, settled and machine-checked

The headline result is the Courtade–Kumar conjecture, posed in 2013. It concerns a classic question in information theory: when a string of bits passes through a noisy channel, can it be compressed to a single bit while preserving more information than the simplest possible method — just copying one input bit? The conjecture says no.

Researchers from Google, CUHK, USC and Carnegie Mellon proved it in full generality with /teamwork's Long Proof pattern, which splits a long argument across separate agents. The proof appeared as a preprint in September; the November post adds the machine-checked formalisation — the whole argument translated into Lean 4, a language that lets a computer kernel verify every logical step. Google puts that formalisation at 50 million lines of code.

That number deserves a closer look. An independent check confirmed the 50 million lines are real, but found only about 239,000 are proof a person could read; the rest is generated certificate data the Lean kernel checks mechanically. The audit trail is still strong, with a clean rebuild and acceptance by the Lean Foundation's external checker.

The Weil conjectures — and the claim that cannot be checked yet

Google also says a framework called /teamwork-LEAP, paired with Gemini 4 Argon, “fully autonomously produced an end-to-end, machine-checked formalization of the Weil conjectures” — the landmark results on counting solutions of polynomial equations over finite fields, proved by Pierre Deligne in 1974. The formalisation spans about 1.8 million lines of Lean, produced across 28,800 subagent invocations.

Here the verification story is thinner. Google's post links to Wikipedia pages for the Weil conjectures and Deligne, with no repository, no paper and no statement file, so the claim cannot yet be independently checked. A separate group, Axiom, published its own public formalisation of the Weil conjectures on November 6, a week before Google's post.

The results go beyond mathematics. In chemistry, 30 agents with a physics-informed spectral simulator identified complex molecular structures with 95 percent top-1 accuracy, nearly doubling single-agent performance. In systems work, agent teams ported the Zircon microkernel's remaining C++ code into about 800,000 lines of safe Rust, and rewrote Google's libgav1 video decoder in Rust, running 20 to 30 percent faster than the C++ original.

What it means for AI research

The pattern Google is describing — agents that propose, criticise and verify, with a machine checker as the final judge — is a notable shift from AI systems that simply generate plausible-looking answers. In mathematics, a wrong proof is not a matter of taste; the Lean kernel either accepts every step or it does not. The caveats matter: the autonomous Weil claim has no public code behind it yet. But the Courtade–Kumar formalisation is public, audited and machine-verified.