Anthropic advises folding subagents into the main agent loop, with a migration kit

In an article dated October 22, 2026, Anthropic's Applied AI team says that in several enterprise deployments on Claude Platform, a single agent using skills and tool search frequently beat both a one-prompt design and a subagent design on quality, often at lower cost and latency. It says one customer saw roughly 2x faster runtime moving from a multi-agent harness to a single loop. These are Anthropic's own observations, not published measurements.

The article still recommends subagents when a single loop degrades from context bloat, when tasks need parallel work with fresh context windows, or when one agent struggles to pick tools or faces conflicting instructions. It suggests folding subagents in one at a time, turning their prompts into skills and their API calls into MCP tools, and comparing evals with and without each subagent.

Anthropic also offers a migration kit that chains its Claude API skills (build-eval, prompt-audit, hillclimb, cost-optimize) into a guided workflow, with a demo to try on a mock system.