A workaround to [[Current (2023) language model agents are too linear]] is having the agent schedule its own followup invocations - e.g. "follow up with this user in a week", or "check daily whether there are emails I should notify the user of". This is helpful, but it's too inflexible, and doesn't adapt to incoming information. [[Amira]]'s [[Language model agents]] use this technique, among others.