Google Antigravity's /teamwork Targets Open Science Problems
Original: Advancing science and engineering with Antigravity's /teamwork
Why This Matters
Multiagent AI tackling real open scientific problems marks a shift from demos to research utility.
Google Research's Antigravity platform, using its /teamwork multiagent system, has attacked open problems in mathematics, chemistry, systems software, and ML engineering alongside human expert collaborators, with external access now available.
Google Fellow and VP Vahab Mirrokni announced results from Antigravity's /teamwork feature, a multiagent AI system built to enhance human-AI collaboration on unsolved problems across scientific and engineering domains. The work spans mathematics, chemistry, systems software, and ML engineering, all conducted in partnership with human experts at Google and academic institutions over recent weeks. The platform pairs AI agents with researchers rather than replacing them — a deliberate design choice the team emphasizes throughout. Google is making the tools externally accessible through a researcher access program. The announcement ties to Gemini 4 Argon, suggesting the underlying models powering these multiagent workflows are from Google's latest Gemini generation. No specific benchmark numbers or solved problem names were included in the available text, but the scope — open problems across four distinct technical disciplines — signals an ambitious, broad deployment rather than a narrow showcase.