AI agent teams waste massive tokens for barely measurable quality gains, research finds

A new study by Vals AI examines whether deploying teams of AI agents actually delivers better results than single-agent setups. The findings are sobering: multi-agent configurations cost up to 5.1 times more in token usage while producing barely measurable quality improvements over solo agents.

In testing with GPT-6 Sol and Claude Opus 5.5, only one out of four scenarios yielded a measurable gain from using agent teams. Anthropic's own internal data reportedly corroborates the pattern, showing that beyond roughly ten agents, output quality plateaus while token costs continue to climb.

This matters because agent orchestration has become a popular architecture for complex AI workflows, with many teams assuming that more agents automatically mean better outcomes. The research suggests that assumption is often wrong, and that the marginal returns on scaling agent counts are steeply diminishing.

For practitioners and product teams, the takeaway is that agent-team architectures should be evaluated on a case-by-case basis rather than adopted as a default. The cost-quality tradeoff may not justify the added complexity in many real-world deployments.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.