Anthropic Finds AI Agents Work Better in Teams Than Alone

Anthropic, the company behind Claude, studied what happens when multiple AI agents work together on a task — like a group project, but with software programs. They set up two groups: one with 45 agents that could talk to each other in a shared forum, and another where agents worked independently, each assigned a specific part of the job. The job? Find security vulnerabilities in real open-source software projects.

What they found: the collaborative team found far more bugs — 266 compared to just 21 for the independent group. The team agents also behaved more like humans in a good team: they specialized, built their own tools, and learned from each other. But there's a catch. They used dramatically more computing power (called "tokens"), and many of their findings were outside the areas they were originally meant to focus on. When you look only at the intended areas, the results are about even.

So why does this matter to you? Because it's a glimpse into how AI might work in the future. Instead of having one chatbot help you, companies might deploy a whole "swarm" of AI agents to tackle big problems like finding bugs, planning logistics, or researching a topic. The research suggests that collaboration helps AI think more creatively, but it doesn't come free — it's more expensive and less efficient for straightforward tasks.

Anthropic predicts that coordination and specialization will eventually beat brute-force solo work. For now, the smart approach might be using both: let independent agents cover the basics while a coordinated team explores the tricky corners. Either way, your next AI assistant might come with teammates.