Anthropic: multi-agent Claude workflows found 66 of 70 planted bugs vs 27 for a single agent
新智元 · wechat · 2026-10-11
Anthropic published test results: with 70 bugs planted in 116k lines of code, a single agent found 14, 15, and 27 bugs across three runs, while Claude's new dynamic workflows found 66 in all three runs. On October 10, Managed Agents dynamic workflows entered public beta, and Claude Code Projects opened to all remaining waitlisted Pro and Max users.
Managed Agents lets the lead agent write its own orchestration program—assigning sub-agents in stages and merging results—with up to 1,000 agents per workflow (64 concurrent). Anthropic's contract-review example uses a second full re-check pass and explicitly flags unreadable contracts as 'uncovered,' separating 'no issue found' from 'not checked.' Projects adds persistent project memory and coordinates multiple cloud threads working in parallel.
Pricing: Projects uses existing plan quotas; Managed Agents bills per token plus $0.08/session-hour runtime, and exceeding a budget pauses work while in-flight requests still complete. Anthropic didn't disclose model choice, time, token spend, or false-positive rates, so whether the multi-agent payoff is cost-effective remains an open question.
More from coding & agent
- Dev builds Claude-coded 'daily wearable' firmware packing 27 games and 16 utilities — ezshine · 2026-10-12
- MCP memory app keeps persistent context across Claude, ChatGPT and Codex — Jolly-Scientist-2484 · 2026-10-12
- HANDBOOK.md benchmark: top agents pass just 36.2% of 65 policy-heavy agentic tasks — bibryam · 2026-10-12
- LangChain's Harrison Chase backs the bet that AI feeds will be the form factor of 2027 — hwchase17 · 2026-10-12
- AWS open-sources Strands Decider 2B, a laptop-runnable decision model with no text generation — dl_weekly · 2026-10-12
- aiprox-mcp lets you discover, hire and pay autonomous AI agents across Lightning, Solana and x402 rails — modelcontextprotocol · 2026-10-12