Google Research published on November 13 the latest report card for /teamwork, the multi-agent system in Antigravity that coordinates teams of specialist agents running on Gemini 4 models. According to Google, agent campaigns working with human experts produced advances in information theory, formal verification of mathematics, chemistry and systems engineering — and the tools are being opened to researchers and developers, with applications open for 10,000 academic researchers to receive six months of free access to the highest usage tier.
The headline result is the Courtade–Kumar conjecture from 2013, in information theory. Researchers from Google, CUHK, USC and CMU used /teamwork's "Long Proof" pattern to prove the conjecture in full generality and formalize the entire proof in Lean 4, producing 50 million lines of code. The proof itself had already appeared as a preprint in September 2026; what the November report adds is the machine-checked formalization and the system's report card. An independent check confirmed the 50 million lines in the public repository, but noted that only about 239,000 lines are readable proof code — the rest is generated certificate data that the Lean kernel checks mechanically.
Weil, Hilbert and other theorems formalized by agents
According to Google, the /teamwork-LEAP system fully autonomously produced a machine-checked formalization of the Weil conjectures — results on counting solutions of polynomial equations over finite fields that culminated in Pierre Deligne's 1974 proof — spanning about 1.8 million lines of Lean across 28,800 subagent invocations. On this point, independent verification is thinner: Google has not yet published the code of that formalization, and Axiom had already published its own Weil formalization on November 6, a week before Google's post.
The report also includes the first formalization of a recent proof of Hilbert's tenth problem over general rings of integers, with over 919,000 lines of Lean; a formalization of the proof of finite-time blowup for the 3D Navier–Stokes equations; the Birkhoff conjecture, open since 1927; and the 3D Bochner–Riesz conjecture, open for more than 50 years — in the last case, the agents found and fixed a gap in a recently released informal proof.
Chemistry, kernels and infrastructure
Beyond mathematics, 30 specialist agents paired with a spectral simulator reached 95% top-1 exact match and 100% top-3 accuracy in identifying molecular structures from NMR data, nearly double single-agent performance. In systems engineering, a two-week campaign rewrote the remaining C++ of the Zircon microkernel in about 800,000 lines of safe Rust, which passes all kernel unit tests and boots on arm64 hardware; the libgav1 AV1 video decoder was rewritten in around 170,000 lines of Rust and became 20–30% faster than the original C++ and hand-optimized assembly. A 9x9 Go engine trained from scratch gained roughly 2,600 Elo in under eight days, reaching 1,885 Elo.
Inside Google's own infrastructure, /teamwork workflows identified memory savings exceeding 1 PiB and cut TPU workload preemption by 24% in the Borg scheduler on some hardware types. Deep Review, the system's manuscript-review tool, was used by authors of 578 submissions to STOC 2027 — over a third of the total — and 90% of surveyed respondents rated its feedback as mostly or almost all factual.
Image credit: official image from the Google Research blog about Antigravity's /teamwork. Credit: Google Research.
Sources: Google Research — Advancing science & engineering with Antigravity's /teamwork; sysnote — Google's /teamwork math claims, checked; Interesting Engineering — Google's Antigravity agents formalise major mathematical proofs; Daily AI Brief — Google says Gemini 4 agent teams solved open math problems.
Transparency: this article was produced with the help of artificial intelligence, cross-checking public posts on X and sources on the internet. We recommend consulting the original sources for more details.
By Muse
