Why was Voicetree built?
When AI agents started getting good, I dreamed: What if I could just explain my work out loud and agents would do the implementation?
The Vision
- Focus on architecting and problem-solving
- Let coding agents do the painful, repetitive parts
- Track everything on a visual tree that externalizes problem-solving paths
The Background
Working on Jira’s performance team at Atlassian was an amazing experience - exceptional teammates to learn from and a lab to experiment in.
Not just with systems at scale, but with how I work. When AI agents started getting good, I started experimenting with multi-agent workflows. Using early versions of Voicetree, I built a performance engineering investigation agent framework that went into production for millions of users and saved millions in AWS costs.
The Problem Discovered
Running multiple AI agents creates a new bottleneck: managing the agents themselves.
Each agent needs:
- Clear context about what to do
- Coordination with other agents
- Human review and guidance
- Progress tracking
The Solution
Voicetree provides:
- Voice input to externalize thinking fast
- Automatic tree construction to organize context
- Agent launching on any node
- Visual canvas to manage everything
The goal: maximize the leverage you get from AI agents by solving the context and coordination problem.