The hardest problems I've worked on, low latency HFT, performance at
scale at Atlassian, all had one thing in common: they exceeded my
cognitive capacity. Where was my team actually solving these? On the
whiteboard. Where we could manage complexity visually, breaking it into
small chunks and their connections.

I've always thought that the ultimate problem to solve is making humans
better at problem solving itself, as that's a multiplier on everything
else. With AI there's now the chance to re-think from first principles
how we interact to maximize each others strengths and minimize
our weaknesses.

Our working memory is extremely limited? Externalise it onto a graph.
We're slow at processing? AI extends your thoughts in the background.
We're single-threaded? See only one task neighborhood at a time in this massively parallel system.

I started by bringing the whiteboard with me everywhere - building a
tool to do voice → mindmap so I could keep track of complex
investigations without having to remember the trace of where I was up
to. That graph then contained the perfect context to feed into agents,
so they could know exactly what I know.

This evolved into how I use Voicetree today: I brainstorm out loud,
creating a mindmap of my thoughts. A coding agent decomposes the task
into a dependency graph. An orchestration agent spawns subagents into
the graph to work til completion - surfacing themselves to me only when
their work doesn't pass the bar.

Voicetree

The spatial IDE for recursive multi-agent orchestration

The spatial IDE for recursive multi-agent orchestration