Google Research reports Antigravity /teamwork agent results, opens tools to researchers

Google Research's Vahab Mirrokni described outcomes from Antigravity's /teamwork multi-agent workflows, where human experts set constraints and agents write solvers, search large candidate spaces and check long chains of reasoning. The post cites work with researchers at CUHK, USC and CMU, including a Lean 4 formalization of the 2013 Courtade–Kumar conjecture in information theory, plus results touching optimization, quantum query complexity, matrix multiplication, streaming algorithms and data structures. /teamwork splits context across specialized agents and relies on verifiers such as the Lean 4 proof checker, unmodified test suites and hardware counters. Google says /teamwork, /boost and other agents are available in Antigravity now, with more features coming for external users.