Descriptions:
Ryan Carson, a solo founder building a legal-tech product focused on improving the divorce process, returns to the “How I AI” podcast to give an updated look at his agent-driven development workflow — one that has evolved dramatically in the year since his first appearance. Carson now runs approximately 15 Devin (Devon) coding agents simultaneously around the clock, organizing work into P0/P1/P2 priority buckets and shipping roughly 40 pull requests per day, at a cost that has reached $20,000 per month at peak usage.
The conversation is candid about both the power and the limits of current autonomous coding agents. Carson explains his “Watchdog” system — an agent that audits each customer account, identifies errors (including Sentry reports), checks whether open PRs have addressed them, and surfaces the top three active problems for human review. He also shares a counterintuitive constraint he applies deliberately: he intentionally limits overnight agent output, not because of quality concerns, but because he doesn’t believe unbounded shipping volume produces proportional business value.
Host Claire also pushes back on the hype, and both agree that no frontier model yet has the business judgment to decide *what* to ship — only humans can supply that signal. The episode is a rare, numbers-grounded look at what it actually costs and feels like to run a solo AI-native company at the frontier of agent tooling, with specific lessons on priority management, decision fatigue, and when to let agents run unsupervised versus when to stay in the loop.
📺 Source: How I AI · Published August 24, 2026
🏷️ Format: Workflow Case Study
