How We Built Our Multi-Agent Research System
- Source
- anthropic.com
- Author
- Anthropic
- Date
Context
Anthropic's research system divides broad questions among agents that investigate separate directions and return condensed findings to a coordinating . Each worker has its own working , allowing more material to be explored without putting every search result into one conversation. The company reports better results on its internal research evaluation, especially for questions with independent lines of inquiry. It also reports roughly fifteen times the use of ordinary chats, making the approach most suitable when the additional research is valuable.
This is a concrete application of the coordinator-and-workers pattern described in its earlier agent-building guide.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
- multi-agent — Using several AI agents on one problem — splitting work in parallel, checking each other, or filling different roles like planner and reviewer.
“We found that a multi-agent system with Claude Opus 4 as the lead agent and Claude Sonnet 4 subagents outperformed single-agent Claude Opus 4 by 90.2% on our internal research eval.”
“In our data, agents typically use about 4× more tokens than chat interactions, and multi-agent systems use about 15× more tokens than chats.”
“We found that token usage by itself explains 80% of the variance, with the number of tool calls and the model choice as the two other explanatory factors.”
“When building AI agents, the last mile often becomes most of the journey.”
Checking sign-in…
Loading comments…