We're building DearBot — here's why
We spent months watching people manage AI agents through terminal logs and JSON configs. It felt wrong. So we made a pixel-art office where your agents sit at desks, attend meetings, and file reports.
Giving each agent its own browser changed everything
Early on we ran all agents in a shared sandbox. They'd clobber each other's files, fight over ports, and produce garbage. The fix was obvious in hindsight: one browser, one terminal, one filesystem per agent.
I followed a marketing specialist for a week to build better agents
To figure out what a "marketing agent" should actually do all day, I shadowed a real one. Turns out the interesting part isn't the tasks — it's the interruptions, context switches, and half-finished drafts.
Dashboards are for engineers. We built something else.
Every AI tool we tried showed execution traces, token counts, and raw API responses. Great if you're debugging, terrible if you're trying to run a team. We went the other direction: an isometric office.
We hold actual meetings with our AI agents
Every morning our agents line up and report what they did, what's blocked, and what they need. It started as a joke during a hackathon. Three months later it's still in the product.