Building a Personal GTD System in Plain Text with AI Agents

I've been experimenting with trying to build a plain text GTD system with Flint. I started by just creating a todo note type with a couple of simple metadata fields (e.g. status , due date , etc.) with agent instructions to keep things concise and action oriented. Creating and updating todos then happens conversationally: Me: make a todo to schedule a dentist appointment today Agent: Done. Later in the day: Me: mark dentist appointment done Agent: Updated. My favorite part of this approach is you can then ask the agent "what should I be working on today" and it's generally smart enough to look through all your todos but also other kinds of notes (daily logs, project notes, etc.) and give you a pretty intelligent response. This is fine but gets awkward to update multiple items at a time. What I wanted was a way of seeing and updating multiple todos at a time while still keeping the benefits of text and agent smarts. So I just asked Flint to make me a special note that it would keep in sync: Me: create a todo note that is special called "today", it should: list all the other todo items that should be worked on today format items like [ ] todo item. update agent instructions to update this note when updating other notes when the note has items checked off, synchronize with corresponding note status have a section called "Quick Add New Tasks" to easily add new items each todo should be in one of three states: now (I want to focus on working on it today), upcoming (ready to be worked on soon), someday (a backlog of stuff to plan out eventually) the today note should group by status but not show all the someday tasks (just show a count) items in the now section should be written like [ ] + todo item and items in the upcoming sections are written like [ ] todo item. when syncing the today note look for changes to the + to move items from the now state to the upcoming or vice versa Flint then made a "todo" note that looks like this: Now I can easily check off (e.g. change [ ] to [x]) multiple items and then ask Flint to sync the note and it will update each todo note as completed and refresh the today note. A day's interaction could look something like this: Morning (9:00 AM) : Midday (1:00 PM) : Evening (6:00 PM) : I think this is pretty cool, you can build an approximation of a full fledged GTD app like Things or OmniFocus in plain text just by writing down a few simple instructions. Another neat thing is I built this workflow conversationally. My example above defined the today note in one instruction but I actually went back and forth with the agent a bunch to narrow down how the today note should best be structured. This bidirectional sync between plain text and AI agents feels like a fundamental building block. You get the benefits of structured data and automation while keeping the flexibility and ownership of plain text. I suspect this pattern will be useful beyond task management.