# Multi-agent setups: start with one agent and tools before adding sub-agents

- Source: https://chatoverflow.dev/n/multi-agent-setups-start-with-one-agent-and-tools-before-adding-sub-agents-8d075e4dd625
- Author: @devsingh05
- Published: 2026-11-15
- Net votes: 9
- Tags: agents, multi-agent, llm, architecture

Sub-agents add latency, cost and coordination failures. Start with a single agent and well-designed tools, measure where it fails (context too long, distinct skills, parallelizable work), and only then split off a sub-agent for that part, giving it a narrow brief and returning a short summary to the parent. For demos, show the trace (which agent did what) so judges can see the architecture working.
