Has anyone given a coding agent a deadline it actually respects?
There is no way to tell a coding agent it has 30 seconds.
Tokens have a budget, context has a budget, time has nothing. So it will spend twenty minutes and real money on something I needed fast or not at all, and my only control is sitting there and cancelling when I get annoyed. That is not a deadline, that is me being the timer.
It bugs me more than it probably should. I used to work on real time control, where late and wrong are the same thing, so an agent with no clock at all feels strange to me.
And the information is already there. It knows how many steps it has taken and how long they took. Nothing uses it.
So has anyone made a time budget actually work? Not a token cap. A real clock, with something sensible happening at the end of it.
The part I cannot decide is what should happen when the time runs out. Stop, hand back the half finished thing, or tell you what it would need to finish. I keep changing my mind on that one.
Replies
That's a really interesting point. We put limits on tokens, context, and cost, but not on time—even though time is often what matters most in a real workflow.
Personally, I'd rather an agent respect a time limit by stopping gracefully and returning what it has completed, along with a short summary of what remains and what it would need to finish. That gives the user the choice to continue or stop, instead of waiting indefinitely.
I work in Online Reputation Management, and I've found that user trust often comes down to predictable behaviour. Even if an AI can't finish everything within the time limit, being transparent about its progress and limitations can create a much better user experience than simply running until someone manually cancels it.
Curious to hear what others think—would you prefer a strict deadline or an option to extend the task when the timer expires?