How One Founder Manages 20 AI Coding Agents at Once - Ryan Carson, Untangle
Summary
Ryan Carson, CEO of Untangle, details how his company manages software development using a system of AI agents, drawing parallels between managing people and managing agents. The workflow centers on a 'manager agent' that orchestrates up to 25 'builder agents' through a cloud-based system (like Devin). Key processes include using a ticket board (Linear) as the source of truth, implementing automated checks every 15 minutes, and utilizing dedicated QA and merge queue managers to ensure high-quality, coordinated PRs. Carson also outlines a novel hiring process where candidates are evaluated by having them ship a feature PR via video recording, which is then ranked by an AI agent.
Key takeaways
-
Agent Management Architecture
5:29
The system uses a 'manager agent' (a session) to oversee multiple 'builder agents' (up to 25). Humans interact only with the manager agent, which handles technical questions and unblocking tasks for the builders.
-
Automated Workflow Maintenance
8:00
A critical automation script must be implemented to force the manager agent to read all child sessions every 15 minutes, ensuring continuous progress and preventing stagnation.
-
Quality Gates and Review Loops
9:20
The process includes built-in review agents (allowing up to three review loops) and an additional QA agent that spins up a child process to run dedicated QA tests. The final human step remains watching the video of the feature before merging.
-
Scaling and Coordination
13:00
To handle multiple concurrent PRs (e.g., database migrations), a dedicated 'merge queue manager' is essential to prevent collisions and coordinate the merge process.
-
AI-Driven Hiring
Untangle has replaced traditional hiring methods (pre-screens, video chats) by asking candidates to submit a 45-minute video of themselves shipping a feature PR. These videos are then fed into a ranking agent for objective stack ranking.
Technical details
-
AI Development Environment
822s
The team operates entirely in the cloud using agents like Devin, advising against local development environments for team efficiency.
-
Workflow Orchestration
360s
The process relies on Linear as the source of truth for tickets, ensuring that every ticket corresponds to a single Pull Request (PR).
-
Agent Communication
For optimal scaling, all agent sessions must be configured to communicate with each other to unblock tasks and share necessary information.
-
Testing and Validation
560s
The system incorporates dedicated QA agents and requires builders to record a video of the feature passing tests, providing a visual confirmation for the human reviewer.
Mentioned resources
Channel & topics
Watch on YouTube · Back to latest
This independent, AI-assisted summary is provided for commentary and informational purposes. It may contain errors or omit important context. Please watch the original video for the creator's complete presentation. Video, thumbnail, and related copyrights belong to their respective owners.