Gas Town, created by Steve Yegge, was an early project focused on agent orchestration but was never meant to function as a reliable professional tool. Yegge acknowledged that it was more of an experimental endeavor designed to push the boundaries of what agent orchestration could achieve, without regard for practicality. In contrast, Anthropic has introduced “Agent Teams,” an experimental alternative aimed at managing multiple agents more effectively. In the Agent Teams setup, one primary agent coordinates a group of worker agents, each maintaining its own context and capable of communicating with one another. This design suits scenarios where independent tasks can be worked on simultaneously without overlap. Anthropic’s earlier system, “subagents,” operates differently by working in sequence and sharing context. Subagents are specialized for specific tasks, making them effective in situations requiring collaborative effort on a single problem. Historically, companies that dominate early stages, like Docker in container orchestration, can struggle if they fail to meet user needs, allowing competitors like Kubernetes to take over. Anthropic appears aware of this dynamic, pushing to create their own orchestration solutions instead of risking irrelevance. The emergence of Agent Teams reflects ongoing experimentation in the field, as both users and AI developers explore the challenges of coordinating multiple agents. The fewer distinct agent roles in Agent Teams compared to Gas Town may influence how well these agents maintain focus and collaborate over time.