Anthropic says one agent loop often beats a stack of subagents. Don't give every team its own agent.

Anthropic's Applied AI team said on October 22 that one agent loop with skills often beats a stack of subagents, so don't give every team its own agent. Anthropic's guide to folding subagents names a familiar failure. Each team ships its own agent, a coordinator sits on top, and details get lost in the handoffs. Anthropic describes the agent architecture becoming a copy of their org chart. One customer measured roughly 2x faster runtime with a single loop.

GTM teams are set up to repeat this. A GTM team, the sales and marketing side of a company, can easily end up with an agent per department. Anthropic's fix is plainer. One loop holds the context. Each team owns a skill, the written steps for how it does its work. Approval gates live inside the tools, and one small central team owns the system prompt.

The test is the part to copy. Start with 10 golden examples, fold in one subagent, and keep it only if the folded version scores worse. Run each task more than once: in Anthropic's sample support agent, a break showed up in one run out of two. An org chart is a fine way to run people. It's a slow way to run one agent.