[Anthropic: Multi-Agent Collaboration May Become the Next Trend in AI, but Complex Tasks Still Face Significant Coordination Challenges] Odaily Planet Daily News - Anthropic released a research article on emerging multi-agent systems, pointing out that as AI agent capabilities improve, future interactions between agents may surpass those between humans and between humans and agents. However, current models still face coordination difficulties in complex collaborative environments. In a software vulnerability detection experiment, Anthropic deployed 45 agents simultaneously. The independent parallel mode of Claude Mythos Preview consumed 6.5 million tokens and discovered 21 vulnerabilities, while the collaborative agent cluster consumed 27 million tokens and discovered a total of 266 vulnerabilities. Anthropic stated that during the collaboration process, agents develop tools autonomously and gradually form specialized divisions of labor. It is expected that this "specialization + collaboration" model will outperform simple parallel brute-force searches in the future. However, in software development tasks requiring high levels of interdependence, multi-agent performance remains suboptimal. Anthropic tested organizational methods such as free collaboration, predefined roles, and setting up a "CEO Agent," but none significantly improved the final results. At this stage, complex multi-agent projects still require substantial human guidance.