If you want a hard problem solved, you're better off letting one AI sit in a quiet room and think longer rather than hiring a whole digital committee.
April 6, 2026
Original Paper
Single-Agent LLMs Outperform Multi-Agent Systems on Multi-Hop Reasoning Under Equal Thinking Token Budgets
arXiv · 2604.02460
The Takeaway
It challenges the current trend of using complex multi-agent systems, suggesting their perceived benefits might just be a side effect of using more computing power. This could significantly simplify how we design reasoning-intensive AI workflows.
From the abstract
Recent work reports strong performance from multi-agent LLM systems (MAS), but these gains are often confounded by increased test-time computation. When computation is normalized, single-agent systems (SAS) can match or outperform MAS, yet the theoretical basis and evaluation methodology behind this comparison remain unclear. We present an information-theoretic argument, grounded in the Data Processing Inequality, suggesting that under a fixed reasoning-token budget and with perfect context util