Ryan Carson

Ryan Carson is the founder and CEO of Untangle, which builds an AI paralegal for divorce attorneys. Seeing the divorce process up close in his family led Ryan to build the first version for people navigating it themselves. He then shifted the product to the family-law firms representing them. A firm selects the documents it needs, and Untangle explains each request to the client and follows up on missing records. One of its attorney partners, Renée C. Bauer, uses it to compare financial affidavits with bank statements and flag discrepancies for her review.

From web development to Treehouse

After earning a computer science degree in Colorado, Ryan worked as a web developer in the UK. In 2004 he founded DropSend, a web app for sending large files, and Carsonified, which ran technology conferences and workshops. His team produced developer conferences for Twitter and Stack Overflow. Income from Carsonified and a small loan gave Ryan and Gillian Carson the means to start Think Vitamin Membership in 2010. Subscribers learned web design and development through video lessons. In 2011 the pair relaunched the school as Treehouse. Alan Johnson joined as lead developer and later became a co-founder. The school helped beginners prepare for their first job in tech and taught more than a million students during Ryan’s decade as CEO.

At Treehouse, Ryan tested several of his convictions about how to run a company. In 2013 the school moved managers into individual contributor roles. It restored management roles as the team grew past 100. Ryan also championed a four-day workweek on the grounds that people could do good work and lead more balanced lives. In 2016, as revenue growth slowed, he and the board ended the policy in hopes of producing more courses. Revenue continued to fall. Treehouse tried selling to businesses and reduced its staff before Xenon Partners acquired it in December 2021. Looking back, Ryan took responsibility for changing the company’s priorities too often and giving too little attention to profitability.

Returning to programming with AI

After the sale, Ryan used ChatGPT as a tutor to learn Next.js. In 2023 he worked on a GPT-4 system at Masterschool to assess students’ progress. He also built Maple, a GPT-4 sales coach, writing both the application and its 17-part curriculum. Maple closed in 2024, the year he founded Untangle. In 2025–2026 he also worked on AI development tools as a builder in residence at Amp.

Ryan’s own tools address two practical questions about handing a feature to a coding agent. How small should the assignment be, and what will the next agent need to know? His AI Dev Tasks prompts turn a feature request into requirements and break those requirements into tasks a developer can review individually. In his implementation of Ralph, adapted from Geoffrey Huntley’s pattern, a fresh agent takes on a change small enough to finish in one session, such as adding a filter to a list. The next agent starts with the committed code, a checklist of remaining work and notes from earlier sessions.

Managing and hiring at Untangle

At Untangle, Ryan usually speaks by voice to a manager agent in Devin that coordinates roughly 20 coding agents in the cloud. People set priorities. The manager chooses tickets whose dependencies are ready and assigns each builder enough work for one pull request. Ryan schedules the manager to check on every builder at 15-minute intervals because a stalled builder can otherwise sit waiting for help. Before approving an interface change, Ryan reads what the feature is supposed to do and watches a recording of it in use, even when every automated test passes.

Ryan wants to see prospective engineers at work before a shared interest or an easy conversation can sway him. Candidates record themselves using an agent to ship a feature, usually in an application they already maintain. An agent ranks the submissions, and the top five candidates receive a paid trial to prepare a real pull request for merge. Ryan saves the first call for the finalists. The first time he tried this approach, he distrusted the ranking. In hindsight, he thought he should have hired the candidate the agent placed first.