ai agents are finally useful. not generating text, actually doing work. writing code, collate tax documents from bank statements, book restaurants, manage social media. they do work while you do other things
fifteen billion hours a year lost to personal administration in the eu alone (ft. numerous policies). bills, paperwork, medical letters, office communications.
agents could handle this. some already do.
but now you have five agents running, each in its own window, each asking for permission, each halfway through something, each waiting on you for the next step.
plate spinning. it is!
the companies building agents know this. they’re all inventing their own coordination interfaces. kanban boards for code. task managers for knowledge work. custom dashboards for every new agent.
we’ve been here before. every app used to have its own notification system, its own way of showing progress, its own paradigm for interaction.
then operating systems standardized it.
agents aren’t that different from apps. they need the same primitives. lists of tasks. progress tracking. permissions. ways to show they’re blocked and need you.
the difference is they’re slow. after decades of instant responses, suddenly we’re waiting again. progress bars instead of loading spinners.
and they need trust boundaries. approve this email before sending. confirm this payment. allow access to this folder once, this session, or forever.
claude code solved this. permissions scoped exactly right. plans stored as text files you can read and edit. clarity about what’s happening and what needs approval.
turns out a plan is just a structured to-do list.
so if every agent needs plans, and every plan is a to-do list, why are we building new interfaces?
your reminders app already does this.shared lists. subtasks. you look at it every day. your family uses it. it already has your context, your calendar, your location.
apple could extend it. let agents read and write to lists. let them pick up tasks and show when they’re blocked. let you delegate work to installed agents the same way you’d assign to a person.
linear did this for work. agents appear as teammates with avatars. you @mention them in threads. they get assigned tickets.
weft is open source, self-hosted. you write “create a cleaned up google doc with notes from yesterday’s standup and send me an email with the link” and an agent writes code to make the doc, summarize notes, connect to gmail.
that’s the pattern. multiplayer to-do lists where ai is just another collaborator.
apple has access to all your personal context. they have a coordination surface you already check every day.the coordination layer shouldn’t be a new app; it should live where coordination already happens, as the future of ai interfaces isn’t new apps… it’s ai embedded inside the apps you already use.
extend reminders to work with agents. let agents read and write to lists. let them pick up tasks, add subtasks, show when something’s blocked.
add a marketplace to discover agents. wedding planning, savings optimization, meal prep. chip away at those billions of hours.
the schelling point for you, your family, and future ai agents is a task manager with shared lists that you already look at.
the reminders app is already a powerful runtime.
if apple chooses to see it.