The Agentic Sprawl: Why More AI Bots Could Hinder Productivity
AI-generated image · US National Wire
As OpenAI and Anthropic push multi-agent architectures, new research suggests that 'agent armies' create operational friction that hinders productivity.
Q: What is the current trend regarding AI agent deployment in the enterprise?
A: According to reporting from The Register, both OpenAI and Anthropic are promoting the use of large numbers of AI agents to maximize the benefits of artificial intelligence. OpenAI previously used the term "swarm" to describe multi-agent architecture, though it has since transitioned to the "OpenAI Agents SDK." Anthropic, meanwhile, is focusing its efforts on multi-agent orchestration.
Q: Is there a direct correlation between the number of agents deployed and increased productivity?
A: Not necessarily. The Register reports that while the business models of OpenAI and Anthropic imply that productivity rises as more agents are run, research suggests otherwise. Hidenori Tanaka, head of NTT's Physics of Artificial Intelligence (PAI) Lab and associate professor at Harvard Center for Brain Science (CBS), stated that adding more agents does not automatically improve performance, comparing it to how hiring more people does not inherently make a company more effective.
Q: What happens when an enterprise deploys too many AI agents?
A: According to The Register, excessive agent populations can actually hinder productivity. Research conducted by Tanaka and Elizabeth Pavlova, a senior data scientist at Harvard CBS, indicates that as the agent population grows, communication becomes more difficult. This leads to the emergence of polarized groups that split into competing camps, each validating its own collective position through internal consensus rather than reaching a broader agreement.
Q: How was this "tipping point" of agent productivity measured?
A: Pavlova and Tanaka used a "Flag Game" to assess multi-agent interaction, a study scheduled for presentation at the ICML 2026 Workshop AI4GOOD. In this simulation, agents are given fractional views of a national flag and must communicate—via broadcasting, passing guesses to listeners, or interacting with a manager system—to reach a consensus. The game concludes when 85 percent or more of the agents agree on the flag for three consecutive rounds or after a set number of steps.
Q: What is the "ideal" number of agents for optimal performance?
A: Based on the Flag Game results cited by The Register, the ideal number of agents is 16. The researchers found that with fewer than 16 agents, the system lacks sufficient evidence to reach a consensus decision. However, exceeding that number causes the agents to become polarized.
Q: What is the primary takeaway for enterprise leaders managing these systems?
A: The core message from Pavlova and Tanaka, as reported by The Register, is that enterprise leaders should prioritize quality over quantity. Simply increasing the volume of AI agents can create operational debt through fragmented communication and competing internal camps, rather than driving higher ROI.