The Problem

AI coding agents are powerful, but they run in a single context window. When you need to delegate tasks (code review, testing, documentation), you’re limited by that single thread.

The Solution

brl-subagent adds enterprise-grade task delegation to the pi coding agent. Spawn isolated AI processes with their own context windows, configurable models, and dependency tracking.

Key Features

• Isolated contexts: Each subagent runs in its own environment
• Model configuration: Choose the right model for each task
• Dependency graphs: Define task prerequisites and execution order
• Observability: Track progress, view logs, monitor resource usage
• Retry logic: Automatic retry on failure with configurable strategies

Why I Built It

I needed it. Working with AI agents on complex projects, I found myself wishing for a way to parallelize work: have one agent handle code review while another writes tests. Existing solutions didn’t fit my workflow, so I built my own.

What I Learned

• How to manage child processes in Node.js
• The importance of structured logging for debugging AI workflows
• That the best tools are the ones you build for yourself