Anthropic advises consolidating AI sub-agents into single systems

Anthropic has advised companies to evaluate whether they can consolidate specialized sub-agents into a single agent that loads context on demand. The company found that many specialized sub-agents can lead to higher latency, rising costs, and information loss at every handoff. In one comparison, Anthropic states that a single-agent architecture ran about twice as fast as the multi-agent setup. A single agent with on-demand context often outperforms multi-agent setups in quality, cost, and latency.

Anthropic’s Applied AI team found that a single agent using skills frequently beat both monolithic prompt designs and sub-agent setups on quality in enterprise deployments. One company had a dozen specialized sub-agents inside its customer-facing agent, which made it hard to keep a consistent brand experience and hurt usability with high latency. Anthropic calls the architecture of many overlapping agents 'Conway's Law' for agents and describes the result of overlapping agents with no shared standards as a 'Frankenprompt'.

Newer models starting with Opus 4.5, along with Opus 5.5 and Fable 5.1, can manage complex plans directly in a single loop. These models comment on their own progress and check their own work, which can make some of the extra control layers unnecessary. Anthropic cut the Claude Code system prompt for newer models by over 80 percent partly because skills and tools now load only when needed. Anthropic notes that the model's tool selection tends to degrade beyond 30 to 50 tool definitions.

Anthropic has published a migration kit on GitHub to help teams make the transition from multi-agent setups. The company recommends packaging workflows into skills and loading tools on demand through a tool search rather than giving every team its own agent. Sub-agents still make sense when a single agent's context window shows measurable performance degradation or when multiple tasks need to run in parallel with separate context windows. In one test with a support agent, folding in a sub-agent did reduce latency but broke a previously working process. The primary agent started answering on its own instead of consulting the right specialist after the sub-agent was folded in during the failed test. A study by Vals found that agent teams can be faster than single agents but don't deliver higher quality. OpenAI developer Eric Provencher has questioned the value of agent swarms, calling them 'almost certainly' wasted money.