Solo founder spent $20,000 on Devin agents in a single month

Ryan Carson, a five-time founder now running the family-law SaaS platform Untangle, spent $20,000 in a single month running 15 concurrent instances of Cognition's Devin coding agent, covering engineering, customer outreach, and investor updates.

Carson tracks the agents with a folder system and a handwritten paper list rather than a dashboard. He uses them to review and merge roughly 40 pull requests a day without a dedicated QA hire, run a "Watchdog" playbook for customer success across his law-firm accounts, and draft investor updates.

The lesson he draws from the month isn't that scale works: it's that raw AI output volume is the wrong goal to chase. He's moving most of his workflow from local agents to cloud-based runs, keeping only one task local by exception, and says a shift in his own customer conversations reshaped his product roadmap more than any agent output did. He's now hiring his first engineer through a process that skips the traditional interview loop.

For teams weighing whether to scale up agent usage, Carson's numbers are a concrete data point: $20,000 a month buys real throughput, but it only stays useful if the output gets checked against what customers actually need, rather than treated as an automatic multiplier.