I’ve always had trouble with task management, but I bought into the to-do list craze of the late 2000s and early 2010s. “These apps can save me; they’ll be my accountability partner,” I thought to myself. I championed Evernote for the better part of a decade, but early on it seemed clear to me that whatever to-do list I created there was destined to get buried underneath all the notes I was creating. Because yes, I’d gone fully digital — personal, work, and everything else lived on Evernote. I had to find an alternative, and lucky for me, it was the App Economy after all, so I had a ton of options. Wunderlist, Any.do, Google’s… thing? I’d switch between all these to-do list connoisseurs like that guy from Icy Tower jumping off one ledge onto the next. The experience always left me wanting; it felt like a burden to be the one deciding what I needed to get done during the week and setting the deadlines.My earliest experience building with LLMs was creating chatbots—because of course it was; that’s what everyone was doing. But it did pique my interest in their classification abilities. LLMs can read any piece of text, understand the sentence structure, and even recognize names and objects, all without any sort of special training. By then, I had fully migrated to Notion for all my task management. Notion is excellent because it’s fully elastic and can be whatever you make of it. I had built my task management system with related databases, labels, views, and all the bells and whistles. But it still never solved my predicament: Who takes on the burden of “managing” my tasks? Because I sure can’t. Notion fits well in an enterprise setting, where the burden of management falls on the team members and the machinations are in place for the product to work at its full potential—the standups, the notetakers, and so on…So I pursued a gimmicky project with a simple concept: a journaling app that takes over managing Notion for you. A user would yap about their day, what they did, and what they want to do. On submission, the app would save the entry in Notion and send that text to an LLM. The LLM would then infer tasks for the user for the rest of the week and check off items already on their task list. The entire premise rested on the user providing unstructured notes and the app taking action on their behalf. This was in late 2024, and I had learned that pattern from coding agents — you give them context and they translate that to code. I never pursued the project further; instead, over the following months, I experimented with connecting agents to my Notion, which practically works, though not without the typical crutches.Then I got to experience Meta’s Muse, which —to me— is a closer embodiment of that vision of delegating task management. It’s a cloud agent, so I no longer have to start it from my laptop. It is persistent, so it holds and continues to grow its memory. It is connected to my other data wells, and importantly, yes… Notion. All these elements have existed in fragments. You can cook up a seafood medley of cloud agents, push memory files to them, deploy them on a private server, and hook up your telegram to start chatting with your assistant. But that’s not a product, they’re just a stack of functions in a trench coat.The experience is not perfect. It feels like an intermediary standing between me and my task list. But it feels like the early innings of consumer apps taking advantage of the underterministic, non rule-based nature of LLMs. Muse is proactive in unpredictable ways, but still largely reliable when it should be. As long as I’m engaging, it will always text me about tasks I got pending and ask about their status.
I think Muse can be a great assistant, but even though it is proactive, it’s still trapped in the chat box experience. I don’t need an assistant asking me what’s on my plate, I need a manager to take over. And if I’ve learned anything about Meta, it’s that they have perfected passive engagement and I’m confident that’s what they’ll bring to Muse.