AI
Study measures how coding-agent teams actually coordinate

Across 1,902 runs plus 244 sealed-environment runs, the authors model messages and file operations as temporal networks. Shared files cut output tokens by about 42% at eight agents on message-heavy work; naming a coordinator did not reliably improve success.
Why it matters and impact: The research suggests coordination channels and shared artefacts matter more than simply appointing a supervisor. Results are an unreproduced preprint and may depend on tasks, agents and prompts.




