I kept losing ideas in the gap between having them and doing anything with them. Chat is a bad container for that: a linear thread flattens non-linear thinking, and the second an idea has to leave the thread to become a task, it dies.

So we built Synapse on an infinite canvas instead. You dump raw thoughts, they float with soft physics, and the AI finds connections between them as you go. Related ideas drift into clusters, and labels fade in above each group, naming themes you never consciously picked.

I wanted the AI to feel like another person at the board, not a sidebar. It brainstorms alongside you, dropping its own nodes that push back on or extend your thinking. When a cluster feels ready you click it, pick an action (research this, draft a doc, generate a mockup), and an agent runs it in the background and drops the result back as new nodes. You never leave the canvas.

Halfway through the 24 hours I cut features the team had gotten attached to. My job was less writing code than keeping design, product, and engineering pointed at the same thing, catching bad assumptions before they turned into bugs.

Building it moved where I think the real frontier in AI is: not model capability, interaction design.