Knowing When an AI Agent Is Actually Finished
The honest answer for most setups today is: you check. Unless you have wired something up yourself, nothing tells you when a long agent run finishes - you either watch the terminal the whole time, or you switch away to other work and interrupt yourself repeatedly to look back.
Why that becomes a bad habit
Long agent runs create exactly this pattern. You switch to another task, then break your concentration every so often to check whether the run finished. A light in your peripheral vision, or a small screen you glance at without switching windows, removes the reason to look at all.
What a status light actually does
The Status Light is a chunky, free-standing desk unit about the size of a chess piece, with three domed lenses stacked vertically - red, amber, green - powered over USB-C and working on both Windows and macOS. The common convention is red for running, amber for waiting on you, and green for a clean finish, though nothing forces that mapping; they are three independently addressable lamps.
It is worth being clear about what it is not: it does not integrate with your agent automatically. It ships as hardware plus the manufacturer's utility, and mapping it to your specific agent's state is a few lines of glue you write once, typically in a start and stop hook. If you want something that recognises your setup with zero configuration, this is not that yet.
A pixel-faced alternative
The Agent Pet does the same job differently: a small orange pixel-robot about the height of an espresso cup, with a square LCD face whose customisable expression shows your agent's state - typically thinking, waiting, or finished. It runs on Type-C power with no battery to manage. The trade-off is the same as the light: the hardware is the easy half, and you supply the hook that tells it what your agent is doing. People buy the light to be informed and the pet because it makes the desk nicer to sit at; in practice, the pet is the one that tends to get photographed.
Is wiring this up worth it
If you already keep a terminal tab open and glance at it a few times, or your runs finish in under a minute, writing a start and stop hook and buying a light is more setup than the problem deserves. It earns its keep once you catch yourself alt-tabbing every minute or two during genuinely long, multi-minute runs. Count how many of those checks come back 'still running' rather than 'done' over a day - that number is what an ambient indicator is trying to shrink, not eliminate, since you still have to write the hook and you still have to decide what red, amber, and green mean to you.
For more on wiring status hardware into a specific agent's start and stop events, see our Status Light guide for Codex CLI, or browse the rest of our status indicator notes.