At a glance
WHAT IT’S REALLY ABOUT
A solo founder’s system for managing 15 AI agents nonstop
- Ryan Carson explains how he runs 10–15 concurrent cloud-agent threads in Devin using folder-based priorities (Bugs, P0–P2) plus a weekly paper list to stay focused.
- He recounts Untangle’s pivot from consumer AI divorce help to B2B software for family law firms after direct customer conversations revealed lawyers would pay for discovery and paralegal-shortage relief.
- The episode argues that AI-driven output is not product strategy: founders must still talk to customers because frontier models can’t reliably decide what to ship.
- Carson shares operational playbooks like “Watchdog,” which audits every customer account, summarizes issues, and checks whether high-frequency PRs have already resolved them.
- Vo and Carson compare tool roles—Devin for background cloud execution and ops, Codex for low-latency pair programming and verification, and Claude Design for generating reusable design systems—plus PR automation patterns like Merge Mommy and LAN PR.
IDEAS WORTH REMEMBERING
5 ideasAgent management is becoming the core job; organization beats raw prompting.
Carson argues the scarce skill is no longer writing code but directing, prioritizing, and verifying many parallel agent workstreams—similar to managing a large org without the pyramid. His solution is lightweight structure (priority folders + weekly paper priorities) so he can keep 10–15 active threads coherent.
Use a dual system: in-tool task triage plus an external focus anchor.
He organizes Devin work by folders (Bugs, P0, P1, P2, Investors) and keeps a separate weekly priority list on paper to prevent attention from being hijacked by endless agent outputs. The combined system creates an “anchor” so high-leverage work doesn’t get lost among fast-moving PRs and fixes.
Turn operational chaos into a repeatable monitoring playbook (e.g., Watchdog).
Watchdog is a repeated playbook that logs into each customer account, summarizes recent activity, surfaces errors (e.g., Sentry), and distills the “top three” problems—then checks whether they’re already fixed/in-progress given the high PR volume. This turns overwhelming operational visibility into a predictable, repeatable monitoring loop.
AI increases output, but customers—not models—determine what to build.
Both hosts stress that shipping more code does not automatically create a better product because models can’t reliably choose what the market wants. PMF came from outbound emails, calls, and on-site visits with law firms—not from autonomous “self-improvement” product loops.
Treat coding agents as business operators, not just programmers.
They describe using agents beyond software engineering: deal desk, customer triage, docs, and internal ops—especially valuable because the agent can both understand the codebase and execute changes. The implication is that “coding agents” are actually general business operators when given the right context and permissions.
WORDS WORTH SAVING
5 quotesIf you're out there listening and you are doing engineering, you're doing your work locally, you really need to open your eyes. Like, I, I think the future is pretty much 100%, you know, cloud agents.
— Ryan Carson
I think one of my big messages for today is that all of us have to uplevel our ability to manage agents. Like, that is our job, right?
— Ryan Carson
I really try to res- like, constrain my output capacity. Not on quality, not on bugs, but, like, because I don't- I don't think I get multiples of quality off of multiples of output.
— Claire Vo
I think what, what, what is happening is people are not getting out of their chair enough and actually talking to real people.
— Ryan Carson
I don't wanna talk to you. I don't wanna have a meeting with you. I don't wanna get to know if I like you. I just wanna see how good of an agent manager you are.
— Ryan Carson
High quality AI-generated summary created from speaker-labeled transcript.
