Anthropic: one agent with on-demand context can beat many subagents

Anthropic says companies should test consolidating subagents into a single agent that loads context on demand. In one architecture comparison the single-agent setup ran about twice as fast. It published a migration kit on GitHub.

Why it matters: This reverses the multi-agent orchestration pattern many teams built for weaker models, and says capable models make that scaffolding overhead. It matches recent findings that fancier coding-agent harnesses cost more without scoring better.