DeepMind's agent-swarm system for large-scale formal proof generation, producing ~6.5M lines of Lean in this effort.
Gemini Carbon demonstrated a dual-agent workflow in which a builder agent generates models in Blender and Three.js while a reviewer agent inspects the rendered output and iterates, producing a playable 3D game of a pelican riding a bike around Google's Mountain View campus. The demo illustrates the feasibility of multi-agent self-review loops for 3D content generation.
Google DeepMind used teams of Gemini Carbon agents to produce roughly 6.5 million lines of Lean proofs showing that Signal's actual compiled code keeps messages private, verified down to the processor instruction level and checked independently by Lean so no trust in the AI is required. The effort took about four weeks versus years for expert teams, with humans setting security goals and reviewing the ~1% flagged by agents. It is claimed as the first end-to-end formal verification of a widely used protocol at the machine-code level.