What do an insurance company, a cybersecurity firm, and a manufacturing giant have in common? All three have built multi-agent AI systems - networks of AI models that coordinate to complete complex tasks by handing off work to each other - and someone has reverse-engineered how they work.
A developer broke down the agent architectures from Lemonade, CrowdStrike, and Siemens into five templates you can open and run directly in a browser. Agent swarms have become a dominant pattern in enterprise AI over the past year, but the implementation details almost never make it outside internal engineering teams. Lemonade's agents handle insurance claims processing, CrowdStrike's coordinate security incident response, and Siemens' tackle industrial workflow automation.
This fills a real gap. Most enterprise AI case studies describe outcomes without showing the plumbing - you get the headline metric but not the architecture that produced it. Practitioners trying to build their own multi-agent systems for smaller-scale projects end up guessing at structure from sparse conference talks and marketing documentation. Runnable templates give you something you can actually break open, modify, and learn from.
The in-browser execution is a practical choice too. No local environment to configure, no dependencies to install - you open it, run it, and start poking at the agent logic immediately.
The five templates aren't production-ready drop-ins, but they're closer to real-world enterprise patterns than most publicly available agent examples, which tend toward toy demos. For anyone building multi-agent workflows for business operations, seeing how a company like CrowdStrike structures incident triage across multiple coordinating models is genuinely useful reference material.