Anthropic says single agents with on-demand context often outperform multi-agent setups

What happened

Anthropic says a single agent loading context on demand often outperforms specialized sub-agent setups in quality, cost, and latency across enterprise deployments. In one comparison, the single-agent architecture ran about twice as fast; the company also reports cutting the Claude Code system prompt by over 80 percent for newer models. Its GitHub migration kit helps teams evaluate consolidation, while retaining sub-agents where context degradation, parallel tasks, or tool selection problems justify them.

Read next

• More changes in today’s briefingMoonshot launches Kimi K3.1 Pro with weights promised November 4 and faster Highspeed API

The story so far

Anthropic has published a GitHub migration kit to help teams evaluate moving from multi-agent setups to single-agent workflows, saying a single agent that loads context on demand often delivers better quality, lower cost, and lower latency across enterprise deployments. The kit guides teams through identifying existing patterns, evaluating alternatives, and verifying which approach works best. In one comparison, Anthropic said the single-agent architecture ran about twice as fast. The company also reported reducing the Claude Code system prompt by over 80 percent for newer models. Its guidance still supports retaining sub-agents when context degradation, parallel tasks, or tool-selection problems justify them.