Every platform makes one bet early that shapes everything after it. Ours was to build agents on the edge: durable, addressable compute that lives close to the user and survives between sessions. This is a short note on why.
Latency and presence
The edge gives us two things at once. Low latency, because the agent runs near the person it works for. And presence, because an edge-resident object can stay alive, hold state, and wake itself without a server we have to keep warm. Those two together are what make a proactive agent feel less like an app and more like a colleague.
What it unlocks
The same foundation makes the harder features tractable: co-edited canvases, self-scheduling work, per-user agents at scale, and on-prem deployments that keep data inside a customer’s boundary. We did not pick the edge to be clever. We picked it because it is the shortest path to agents that keep working.