Brain Dump

Inspiration

I'm a long time explorer of to-do apps. Meaning, I'm still in a search for the one I would actually use. I even made my own minimalist app, three.do, at some point. And about a year ago I speculated about my perfect kanban-based GTD-like to-do app, which I never got around to building.

The #1 problem with all the to-do apps is the immense friction when it comes to actually capturing tasks and then grooming them into hierarchies. You either have to type them out neatly right away, or you'll have to spend a lot of time grooming them later. This ultimately made me give up on every task tracking app I ever tried. All those apps might have advanced tagging and scheduling and collaboration but ultimately suck at the first cruicial step: the brain dump.

Now, I decided to try and see if I could use Coda AI to build a frictionless brain dump for myself.

What it does

The doc is set up to parse and process brain dumps, either structured-looking or natural languages, and properly add them to the database of tasks and subtasks.

How we built it

Just like any other of my Coda docs, with tender love & care. 🙂 Best Practices aboard, and I also tried to make the interface nice and intuitive.

Unfortunately there was no time to finish building the template — I had a vision for a complete task tracker but ultimately decided to pursue building more assorted ideas rather than polish just one. Ultimately this template could integrate with this mind-blowing GTD-like doc of mine and replace its rather lackluster first step. But again, that would take a few extra days, which sadly I didn't have.

Challenges we ran into

It is almost impossible to get Coda AI to reliably match new incoming tasks agains the tasks that already exist. E.g., sometimes it would understand that if I tell it to "Under Cleaning, add xxx", it should put the tasks under the existing "Clean the room" supertask. And sometimes it wouldn't, and would basically create duplicates.

For this case I created the Merge interface that allows to quickly move all child tasks from one parent task to another, and delete the duplicate.

Tasks Inbox UI was also quite challenging to build: I strived for simplicity and had to use quite some tricks.

Another problem that wasted my time (in vain) was that the direct Telegram -> Webhooks integration stopped working for some reason. Previously it worked just fine, and this time it decided to not deliver the webhooks. So I wasted some time debugging the integration without any luck, and then had to resort to a webhook forwarding service.

Accomplishments that we're proud of

I built a task parser that I could actually see myself using. Also made a very minimalist yet powerful task inbox grooming workflow.

What we learned

That prompts should be debugged: sometimes trying to combine a prompt out of pieces kills the formatting somewhere, and it affects how Coda AI understands my input. E.g., sometimes it would not detect tasks because of that

What's next for Brain Dump

Build the rest of the doc: the Eisenhower Matrix, and all sorts of kanbans for planning. And trying to use it myself on a daily basis.

Built With

• coda
• coda-ai
• gpt