When several coding agents run the same app, Docker looks like the obvious answer: put every frontend in a container, put every API in a container, and put every worker in a container.
It works, but it adds friction. Stale images can serve old code, rebuilds slow down feedback, caches hide what actually changed, images and layers clutter storage, and debugging gains another runtime layer.
With one workspace, that may be fine. With several agents, it gets noisy fast.
Local isolation does not require containerizing every process
Most collisions come from ports, processes, Docker services, browser profiles, logs, and environment state. Agent-Up isolates those directly.
