Anthropic has discovered that a single AI agent with on-demand context often outperforms multi-agent setups in quality, cost, and latency, releasing a migration kit to facilitate the switch. The company's findings suggest significant efficiency gains for enterprises using AI agents.
TL;DR
- Anthropic's research shows single agents with on-demand context can double speed and reduce costs compared to multi-agent setups.
- The company has released a migration kit to help teams evaluate and transition to single-agent architectures.
- Anthropic's findings align with recent research indicating that agent swarms may not deliver quality gains but significantly increase costs.
What happened
Anthropic's Applied AI team found that a single agent with on-demand context often delivers better results than a fleet of specialized sub-agents. This approach can run about twice as fast and reduce costs and wait times, according to the company.
The recommendation comes from hands-on experience across multiple enterprise deployments. In one case, a company with a dozen specialized sub-agents in its customer-facing agent struggled with consistency and high latency. The single-agent workflow resolved these issues by maintaining context throughout the process.
Anthropic has released a migration kit on GitHub to help teams identify existing patterns, evaluate alternatives, and verify the best approach for transitioning to single-agent architectures. The kit includes a step-by-step workflow and warns against blindly merging sub-agents, as it can sometimes break previously working processes.
Why it matters
For developers and startups, this finding suggests that consolidating sub-agents into a single agent can significantly improve efficiency and reduce costs. It also highlights the importance of maintaining a clean system prompt and lean data tools to avoid conflicts and inconsistencies.
For investors, this shift towards single-agent architectures could indicate a trend in the AI industry towards more efficient and cost-effective solutions. It may also influence the valuation and competitive positioning of companies that adopt this approach.
However, there are limitations to this approach. Sub-agents may still be necessary when a single agent's context window shows measurable performance degradation, or when multiple tasks need to run in parallel with separate context windows. Additionally, the model's tool selection tends to degrade beyond 30 to 50 tool definitions, so designated owners on each team should maintain the skill libraries to keep them from ballooning.
Key facts
- Anthropic's single-agent architecture ran about twice as fast as the multi-agent setup in one comparison.
- The company's migration kit is available on GitHub to help teams transition to single-agent architectures.
- Anthropic's findings are consistent with recent research by Vals, which found that agent teams can be faster but don't deliver higher quality.
- OpenAI developer Eric Provencher has also questioned the value of agent swarms, calling them 'almost certainly' wasted money.
- Anthropic recommends packaging workflows into skills and loading tools on demand through a tool search to keep skill libraries from sprawling.
- The company advises against fetching context through code and injecting it directly into prompts.
- Anthropic's Opus 4.5, Opus 5.5, and Fable 5.1 models can manage complex plans directly in a single loop.
- The Claude Code system prompt was cut by over 80 percent for newer models, partly because skills and tools now load only when needed.
Context
This shift towards single-agent architectures is part of a broader trend in the AI industry towards more efficient and cost-effective solutions. As models become more capable, they can handle more complex tasks directly, reducing the need for specialized sub-agents.
The findings also highlight the importance of good software engineering practices in AI development. Maintaining a clean system prompt, lean data tools, and fixed output guardrails can significantly improve the performance and consistency of AI agents.
As the AI industry continues to evolve, it will be important for companies to stay up-to-date with the latest research and best practices. Anthropic's migration kit and recommendations provide a valuable resource for teams looking to improve their AI workflows.
