Cognition, the company behind Devin, recently raised $400 million, but the author has declared Devin officially dead, suggesting there's a better option available. The rise of async coding agents has made it easier to automate coding tasks without constant supervision. While early versions of these agents struggled, advancements in AI models have improved their reliability. Startups and labs are now focused on these background agents, but many developers still prefer to manage their agents closely, limiting broader adoption. The author decided to create a DIY solution for async coding agents, combining different technologies for a more flexible setup. This involved using Slack for triggering tasks, a Github app for access management, and serverless computing through Modal. The initial implementation faced challenges, such as permissions issues and the need for isolated storage for each agent. After several iterations, the author refined the system, allowing agents to run with the necessary permissions and providing better observability through structured logs. Building this system required understanding the limitations of existing cloud agents and customizing the setup for personal needs. The author highlighted the frustrations with existing models and interfaces, emphasizing the benefits of a tailored approach. This project reflects a shift in how developers can leverage AI, making async coding agents more practical and effective for complex tasks.