Supervise a fleet of coding agents, not a single window
One agent in one window is easy. A fleet of them is where the real work is, and where the tools stop helping. Arbiter launches, names, and drives many live sessions at once, with status you can trust about which ones are actually alive.
A single coding agent in a single window is a solved problem. The clients are good at it. The trouble starts at the second agent, and gets worse at the fifth: now you are the scheduler, the status board, and the memory of what was running where. That is not supervision. It is manual labor dressed up as control, and it falls apart the moment you look away or the machine reboots.
Supervising a fleet is a different job than running a session, and Arbiter is built for the fleet.
From one window to many, on purpose
Arbiter can bring up a set of live agent sessions at once and hold them as a group rather than a scatter of windows. You launch them together, and they come up already tracked: each one a real, accounted session Arbiter knows about, not a terminal you have to remember you opened. Going wide becomes a deliberate act instead of a slow accretion of tabs you are afraid to close.
Named, so you are steering and not guessing
A fleet you cannot address is a fleet you cannot steer. Every session carries an identity you can point at, so an instruction lands on the exact worker you meant and not on whichever window happened to be in front. When you want to drive one, you drive that one. Nothing sprays commands across the group by accident, and nothing important stays hidden behind a window you have not clicked into for an hour.
Status that tells the truth about what is alive
A fleet dashboard is worthless the moment it lies. The failure that erodes trust is not a crash. It is a panel that still shows a worker as busy long after it died. Arbiter computes liveness from the real state of each session rather than from a stale label, and it speaks one consistent vocabulary for that state across the whole fleet. The counts add up because they are derived, not guessed. When Arbiter tells you five sessions are working and two are waiting on you, that is the actual picture, which is the only kind of status worth acting on from a distance.
One operator, a whole fleet
Put those together and one person can supervise a fleet without becoming its bottleneck. You see everything at once, you address any one of them precisely, and the state you are reading is real. That is the difference between running agents and merely having several of them open. It is also the ground that remote supervision stands on, the subject of What Is Remote Agent Supervision. You cannot supervise from your phone what your own desk cannot show you honestly.
Above the stack
The fleet is agnostic by construction. Arbiter drives the clients you already use, dispatches Warden across providers, routes code questions to TheAuditor, and leans on Curator for memory. Bring the agents you have. Arbiter is the commander over them.
Arbiter is a proprietary working alpha in public binary release preparation. Follow the public release channel or RSS feed for updates.