Running my agents in a VPS (crowdhailer.me)

A developer has launched an experiment to run multiple AI agents in a Virtual Private Server (VPS) environment, enabling them to operate asynchronously with minimal oversight. By using coding harnesses like Claude Code and Codex, the goal is to explore the capabilities of these agents when allowed full autonomy, including permission settings that let them execute tasks independently. Each agent is assigned its own user account on the VPS to ensure isolation, allowing the developer to observe their interactions and performance effectively while maintaining the flexibility to discard any misbehaving setups quickly. This approach underscores a significant shift in how developers can leverage AI tools for more complex tasks without constant human interaction. The use of a shared GitHub account for all agents enables collaboration and streamlined coding practices. The experiment reveals lessons about trust in automation, the importance of rapid iterations with disposable VPS, and the potential of AI to generate actionable insights beyond simple code suggestions. As the AI/ML community continues to explore the intersection of autonomy and collaboration in development, this experiment opens new avenues for enhancing productivity and refining the deployment of AI agents in real-world applications.