A new arXiv paper examines multi‑developer agentic systems (MDAs), where many programmers each create a small AI agent. When these agents are combined, they reach 70% accuracy on the GAIA benchmark, while the best single‑developer agents stay below 35%. The MDA outperforms a typical single‑developer system by 12 percentage points, even when the latter receives several times more development budget. The authors suggest MDAs could become a new way to scale AI by pooling diverse talent.
Why it matters
Collaborative AI development could let smaller teams create higher‑performing assistants without massive budgets.
