Google DeepMind's 2025 study on multi-agent scaling found what effect — Agentic Systems Engineer Practice — Claude Cert Academy

Google DeepMind's 2025 study on multi-agent scaling found what effect when additional agents were added to sequential reasoning pipelines?

• Performance scaled linearly with agent count up to 10 agents
• Performance degraded by 39–70% compared to the single-agent baseline
• Performance improved modestly (5–10%) when more than three agents were used
• No statistically significant effect was observed across task types

Answer: Performance degraded by 39–70% compared to the single-agent baseline

The Google DeepMind 'Towards a Science of Scaling Agent Systems' (2025) study quantified what practitioners had observed: adding agents to sequential reasoning pipelines degrades performance by 39–70%. This 'Parallelism Cliff' occurs because coordination overhead, context fragmentation, and inter-agent communication costs overwhelm any benefit when each step must build on the prior one. Swarms accelerate parallel tasks but damage sequential ones.

Continue to Claude Cert Academy