Broad access is not the same as a useful watch

TechCrunch’s reporting on ChatGPT Work captures the tradeoff plainly: the more places an assistant can reach, the more work it may be able to take on—and the more personal context it may encounter. That is a bigger bargain than most notification settings admit.

A good watch can be narrow. ‘Tell me if this specific rule changes.’ ‘Let me know if this price crosses the number I set.’ ‘Bring this back only if the answer from the primary source is different.’ Those are understandable jobs with understandable boundaries.

The bad version is an assistant that gets access to everything because the product has not decided what it is watching. That does not feel helpful for long. It feels like a person has hired a very eager intern and forgotten to give them a desk, a brief or a quitting time.

Make the interruption earn itself

There is a simple test for any AI assistant that monitors work in the background: could the person act on the alert from the lock screen, or do they have to open the app and reconstruct the whole situation?

The best alerts carry a little of the original context forward. They say what was being watched, what changed since the last look, which source supplied it, and what remains uncertain. That is enough for a person to decide whether the moment matters without having their day reopened by a vague nudge.

AI agents will get better at waiting. The more valuable skill may be knowing how to return without taking over the room.