Why Multi-Agent Systems Get More Expensive — Routing, Handoffs, and Quadratic Cost
Why Multi-Agent Systems Get More Expensive — Routing, Handoffs, and Quadratic Cost A five-step multi-agent workflow. Not one line of business logic changed — only the decision of who goes next was handed to a model. The overhead went from 750 tokens to 18,000 tokens (750 = five hops × 150-token summaries). Nobody wrote a bug. The extra 17,250 tokens come from one thing: everything each agent said…
Multi-agent systems incur higher costs due to routing and handoff management, leading to a quadratic cost complexity. In a workflow involving five agents for analyzing a 20-page quarterly report, each agent produces about 800 tokens. When agents pass outputs to subsequent agents, each round reads the full history of all previous outputs, resulting in a quadratic increase in input size.
For example, in a five-round task, the total input grows from 400 tokens to 10,000 tokens, with the quadratic component increasing from 2,000 to 36,000 tokens. This quadratic term is present in both routing and handoff processes, both of which read the same prior context. Routing costs are constant if the routing model only uses the goal, candidate list, and previous outputs without needing to understand the full report.
To mitigate these quadratic costs, four strategies are proposed: fixing routing to avoid re-reading the full history, limiting the number of rounds, implementing a deterministic approach for the routing model, and classifying orchestrators based on whether the next step truly depends on the previous step's output. Applying these cuts can bring the cost back to a linear complexity, making multi-agent systems more scalable and cost-effective.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.