Hi Erèbe, Thanks for the suggestion. I've only had a very quick look at the auto-update package, so please correct me if I'm wrong. But I think there is an important difference between auto-update and the suggested pollT (or the async-refresh package): auto-update does not completely decouple the execution of the IO action from the calling thread. I'm referring to this comment: mkAutoUpdate :: UpdateSettings a -> IO (IO a) Generate an action which will either read from an automatically updated value, **or run the update action in the current thread**. pollT and async-refresh are implemented such that the caller really only retrieves the result of a previously executed IO action. Thus even if the IO action might block, the caller does basically not block (more than required for reading a TVar). For the use case it was created for (refreshing of authentication tokens in a micro service) it might be very important that the required tokens are *always* guaranteed to be available already at the time a request needs to be made. But maybe I'm overlooking something and auto-update does indeed also provide this functionality. Best, Moritz