This academic paper by Google Research, Google DeepMind, and the Massachusetts Institute of Technology, systematically evaluates the principles for scaling language model-based agent systems, moving beyond anecdotal evidence that "more agents is all you need." The authors present a controlled evaluation across four diverse agentic benchmarks, testing five canonical architectures—Single-Agent, Independent, Centralized, Decentralized, and Hybrid Multi-Agent Systems—to isolate the effect of coordination structure and model capability. Key findings establish that multi-agent benefits are highly task-contingent, ranging from a significant performance increase (+81%) on parallelizable tasks like financial analysis to substantial degradation (-70%) on sequential planning tasks, primarily due to measurable factors such as the tool-coordination trade-off and architecture-dependent error amplification. Ultimately, they derive a predictive quantitative scaling principle that explains over 51% of performance variance and can predict the optimal architecture for unseen task configurations.