Study: Multiple AI Agents Often Worse Than One
Using multiple AI agents may not improve results and could waste money.
An analysis synthesizing nine primary empirical evaluations examined 516 unique multi-agent language-model configurations—and the comparison often confounds coordination with sample multiplicity and inference compute. Against single-pass baselines, multi-agent accuracy differences are negative overall and near zero when omitting the largest contributing study.
Against baselines matched on sample count or resource budget, multi-agent systems average -0.042 in accuracy, and multilevel random effects modeling on the log odds scale remains negative across sensitivity checks. Multi-agent accuracy equals or falls below the baseline in 74% of comparable contrasts, and study means are nonpositive in almost all evaluated studies. Because communication adds no information when agents share identical evidence, positive returns require partitioned evidence, external verification, or error diversity across models.
- Multi-agent AI systems often perform no better than a single AI pass, and sometimes worse.
- Communication between agents only helps if they have different information or can verify each other.
- Companies could save money by using simpler AI setups instead of complex multi-agent systems.
Why It Matters
Businesses may waste money on complex AI; simpler, cheaper AI can be just as effective.