AI agents are becoming a common format for real-world AI applications, but a new paper says their performance does not always improve as more agents are added. The paper, “Towards a Science of Scaling Agent Systems”, examines how systems capable of reasoning, planning, and acting behave in multi-step workflows.
The source says the shift from single-shot question answering to sustained interactions creates new complexity. In agent systems, a single error can cascade through a workflow, making standard accuracy measures less useful for judging performance.
According to the paper, practitioners often rely on the idea that “more agents are better”. It cites “More Agents Is All You Need”, which reported that LLM performance scales with agent count, and collaborative scaling research, which found that multi-agent collaboration “…often surpasses each individual through collective reasoning.”
The new study says it challenges that assumption through a large-scale controlled evaluation of 180 agent configurations. It says the work derives the first quantitative scaling principles for agent systems and finds that the “more agents” approach often reaches a ceiling.
The paper also says adding more agents can even degrade performance if the system is not aligned with the specific properties of the task.
Source: research.google.
Companies can share verified announcements through Newz9’s international press release submission page.
