WhoBuilt

How This 5x Founder Runs His Startup Solo With AI Agents

May 24, 2026· 39 min· 11K views·Watch on YouTube

Speaker

Ryan Carson, a five-time founder building Untangle, joins Peter Yang to explain how he runs the startup solo with OpenClaw, Codex, Devin, and scheduled workflows.

Build an executive assistant with OpenClaw

Carson built Claw Chief from Markdown files and scheduled jobs to help his OpenClaw assistant, R2, manage meetings, email follow-ups, priorities, and prospecting. A priority map gives R2 context about quarterly goals and important people, while a sweep every 15 minutes checks his inbox, calendar, and tasks. Carson says the setup takes work and can be fiddly.

Give agents clear rules and access

Carson describes agents as scheduled jobs paired with skill files, with AGENTS.md pointing R2 to the priority map. He uses Slack threads to organize conversations, Todoist for tasks, and separate email and GitHub accounts for R2. For meetings, R2 checks Carson's calendars and sends available times, then books from its own account.

Automate prospecting and keep the system maintainable

R2 backs up its setup nightly and runs a daily prospecting workflow that finds Connecticut family law attorneys and mediators, adds them to a Google Sheet, and sends outreach. Carson uses Firecrawl for web research and says the workflow has generated 10 to 20 meetings in a few weeks. He recommends checking for prior outreach and carefully reviewing instructions, since small details in a Markdown file can produce unintended email behavior.

Use cloud agents for engineering and review

Carson uses Devin's cloud development environment to avoid local dependency and configuration issues, and schedules playbooks for recurring browser tests and smoke tests. He puts every change through a pull request and uses a “Land PR” playbook to review, resolve issues, and merge; he says he ships at least 10 PRs a day. For features, he talks through an idea, has an agent clarify it into a PRD, then asks it to build and test the feature.

Build repeatable marketing workflows

Carson uses Devin to manage Google Ads through a custom CLI and API, including campaign changes. For social content, he combines Descript clips, Gemini video descriptions and copy, generated cover images, and Publer publishing. He says a designer created the initial brand assets and reference images, which he uses with a design.md file to guide later image generation.

Takeaways

  • Pair scheduled runs with concise skill files so agents repeat work consistently.
  • Give agents a clear priority map and specific rules for handling tasks and communication.
  • Use pull requests and repeatable review playbooks to make agent-written code easier to ship.
  • Automate recurring tests in a stable cloud environment and refine the playbooks when they expose problems.
  • Invest in the initial documentation, tools, and reference material before expecting agents to run a workflow reliably.

Mentioned in this talk

Search inside this talk

What was claimed 18

Show all 18 facts

Related talks

How This 5x Founder Runs His Startup Solo With AI Agents, Ryan Carson · WhoBuilt