Advancing science & engineering with Antigravity’s /teamwork
Summary
- Google Research announced results from Antigravity's /teamwork multiagent systems, utilizing Gemini 4 models like Argon, Carbon, and Hydrogen to collaborate with human experts on open problems in mathematics, science, and systems engineering.
- The /teamwork system addresses language model limits by pairing parallel exploration with dedicated falsifiers like the Lean 4 proof checker and partition-by-construction context handling to maintain focus over long-horizon tasks.
- Researchers and academic partners used these agentic frameworks to prove the 2013 Courtade–Kumar conjecture, resolve open questions in optimization and quantum computing, and produce formal verifications spanning millions of lines of Lean code.
- The multi-agent LEAP framework autonomously produced machine-checked formalizations for landmark mathematical problems, including the Weil conjectures, Hilbert's tenth problem, and aspects of the Navier-Stokes equations.
- Google also introduced Deep Review via /teamwork's document review pattern, providing automated manuscript feedback for conference submissions to help catch technical gaps and proof bugs.
Why it matters
These advancements demonstrate how structured multi-agent systems and long-horizon reasoning frameworks can scale human-AI collaboration to solve complex, long-standing scientific and mathematical challenges while formally verifying results.