Princeton paper: multi-developer agent systems hit 70% on GAIA
The implication is that agent ecosystems beat monolithic builds: interoperability across developers' agents may matter more than one team's polished general agent. Builders should design agents to fill gaps in shared, composable systems rather than duplicate basic skills. The SealQA result also flags the trade-off — task routing wins on GAIA-style benchmarks but can cost more elsewhere.
A Princeton paper finds that combining specialized agents from many developers can outperform any single developer's general agent. On GAIA, with the same total development budget, specialists each scoring at most 35% alone reached 70% when combined, 12 points above the median general agent. The authors simulated developer populations building agents under a fixed budget, with the multi-developer system splitting each task and routing parts to different developers' agents. On web-search-only SealQA, however, the combined system performed worse at higher cost. One agent scoring 19% alone lifted a combined system from 40% to 53%.
New Princeton paper finds that combining specialized agents from many developers can beat any single developer's general agent, with GAIA accuracy reaching 70%.
Today most agentic systems are built and controlled by 1 developer, so each team rebuilds the same basic skills. The authors argue this limits scale.
They simulated populations of developers with LLM agents that build agents under a fixed budget, then compared lone systems with a multi-developer system. The multi-developer system splits each task and routes the parts to different developers' agents.
On GAIA with the same total development budget, specialists that each scored at most 35% alone reached 70% together. That is 12 points above the median general agent, while on web-search-only SealQA the combined system did worse at higher cost.
Build agents to fill gaps in shared systems, since an agent scoring 19% alone lifted a combined system from 40% to 53%.
