I spent $20,000 on Devin in a month. Here’s what I learned | Ryan Carson (solo founder)

The episode opens with Ryan Carson discussing his approach to managing a large number of AI agents, specifically through a folder-based system and a piece of paper for prioritization. He emphasizes that managing the throng of agents effectively is now a core skill for solo founders and leaders. Ryan highlights the importance of categorizing tasks by priority (P0, P1, P2) and using tools like folders to organize multiple concurrent threads, which allows for managing thousands of agents efficiently. This section underscores the shift from traditional management to managing multiple AI agents as a new frontier in productivity and organizational structure.

Ryan shares his evolving stack of AI tools, notably his transition from local to cloud-based agents like Devon, which handles various tasks including bug fixing, customer operations, and code reviews. He discusses how he is leveraging cloud services to scale his AI management capability, increasing his spending significantly to access more credits. Additionally, Ryan talks about integrating tools like Claude Design and Codex into his workflow for designing systems and building product features by converting design markdown into actual code. This section explores how managing AI-driven workflows effectively is critical for solo founders and technical teams, emphasizing automation and strategic tooling.

Ryan emphasizes the importance of talking to real customers over solely relying on AI output. A key lesson he recounts is how direct customer engagement, such as in-person interviews and shadowing clients, helped redefine his product direction for Untangle—an AI-powered platform for family law firms. This highlights the need to validate AI products through real-world feedback and human interaction, rather than assuming that more AI output directly correlates to better products. Ryan advocates for strong customer connection as a foundation for successful AI-enabled products.

In discussing hiring, Ryan explains that he is experimenting with assessing potential engineers through video recordings of tasks, essentially hiring AI managers in the form of humans. He compares managing AI agents to hiring human team members, emphasizing that the key is observing how well they perform tasks. Ryan also touches on the challenge of hiring in the AI era, where traditional interviews are supplemented or replaced by practical assessments involving AI tools. The conversation underscores the shift towards operationalizing AI performance in talent acquisition and management.

The episode covers the technical workflow for design and development, where Ryan and Claire discuss using Claude Design to create design tokens, then leveraging Codex to generate implementation-ready components and sync back into Figma. Ryan mentions practical tools like Merge Mommy for automating pull request reviews and risk assessments, streamlining the code review process. The emphasis is on building repeatable, automated systems for design and development tasks, illustrating how AI can elevate efficiency in technical workflows.

The conversation touches on the limitations of Slack for managing AI collaborators, with Ryan advocating for more public, thread-based communication to foster transparency and collaboration. He discusses strategies like moving away from DMs and private channels to more open, AI-involved workflows, mimicking the effectiveness of public threads for real-time, context-rich work. The episode suggests that AI may challenge traditional work cultures and communication practices, urging a transition toward more visible, thread-based interactions.