
What happened
A paper by University of Macau and The Hong Kong Polytechnic University researchers tested web-application development tasks and found agent gains split into three stages, with efficiency peaking at roughly 8 agents before coordination overhead makes new agents fade fast.
Why it matters
This suggests that past a certain headcount, extra agents spend tokens on coordinating with each other rather than on the actual work, so simply adding more agents does not scale productivity.
Summaries like this, every morning.
The paper's starting point is a classic software-engineering idea: The Mythical Man-Month, which warned that adding people to a late project makes it later. The researchers asked whether the same coordination cost appears when AI agents are added to a system. They ran experiments on web-application development tasks and measured token efficiency as the number of agents grew. What they found was not a straight line but three stages: first parallel gains, then a peak around 8 agents, then a stage where coordination dominates and new agents barely help. The paper argues this mirrors human teams, where a 10-person group and a 100-person group have very different management costs. To address this, the authors adapt the Surgical Team structure from the same book: instead of all agents talking freely, a Coordinator aggregates and distributes information, and each agent gets only what it needs. The paper reports this model keeps competitive quality versus existing multi-agent methods while running more efficiently. For companies now experimenting with multi-agent setups, the implication is that role, communication, and control design matter more than the number of agents deployed.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Anthropic sold 980 million shares at $83 and opened at $111, raising $81.3 billion; a $26 billion private plac…

IBM published commentary arguing that AI value comes from a company-wide "AI Operating Model" — connecting int…

With Google Chrome's Modern Web Guidance plugin loaded, Claude Opus 5.5 wrote 2,740 JavaScript lines across 16…

Google detailed the Mirror Dimension, an isolated replica internet with social media, shopping and bank dashbo…

Meta employees said Muse mixed up chats across accounts and surfaced messages and bank balances they never sha…

Anthropic's shares began trading on the Nasdaq at $111.50, 33% above the $83 IPO price set Wednesday, which wa…
