Useful GenAI use cases for a todo.txt application?

[ransome1]

Lately I was wondering if there are interesting use cases for any sort of LLM integration in sleek. Although I use LLMs quite a bit as daily drivers, I am often puzzled about the awkward ways it is being implemented into services. If there are any use cases for sleek I really want to make sure those make sense.

One idea is to use AI for assistance while writing todos, similar to how JIRA uses LLMs to create templates, fix typos, or add context from other tickets.

I experimented with Mistral AI's APIs by sending a todo: Fix sorting in drawer and create release tomorrow

I provided context like the todo.txt format, today's date, and my existing todos. The AI responded with something like: 2025-02-22 Fix sorting in +drawer and create @release tomorrow due:2025-02-23

The results differed quite a bit depending on how much instructions I added for the todo creation.

It seems the AI can understand the todo.txt format at least roughly, calculate due dates, or cross-reference existing todos.

I'd like to discuss with sleek's users if this integration is beneficial or if there are other use cases worth exploring.

[fredofredofred]

Personally I like to keep my todos very simple and to the point. A lot of people use LLMs to communicate with other people but todos are just for me.

Another option would be to feed a to-do in to a LLM to get suggestions on how to proceed?

Or perhaps to suggest how some items should be prioritised? (if that's something it could feed back)

However I might not be the best person to suggest as I don't use it - I haven't yet found a use-case which I feel adds much value.

[fredofredofred]

That's not too bad on the CDs.

What about suggesting and adding contexts? A lot of the time you only realise which contexts are most relevant once you're half way through the task. Admitedly not too hard to do manually.

Alternatively scanning a document/email and suggesting tasks in a todo.txt format? Lots of the AI suggestions seem to be along those lines. I guess you'd need to specify the format in the question and hope it did something sensible.

[fredofredofred]

I tried chatGTP with 'Generate some tasks in todo.txt format based on typical tasks someone would need to do when redecorating their home' and it wasn't great. It has no idea about priority, generating a mood board happened after the walls were painted etc.

It took a few refinements to get something sensible but it looked like this after a bit:

(A) Choose color palette for rooms +project @home @design
(A) Measure room dimensions +project @home @planning
(B) Research furniture styles and designs +project @home @shopping
(C) Create mood board for living room +project @home @design
(A) Order paint for walls +project @home @shopping
(A) Order new furniture (sofa, chairs, coffee table) +project @home @shopping
(B) Order rugs for living room and bedroom +project @home @shopping
(B) Purchase curtains and blinds for windows +project @home @shopping
(A) Hire painter for wall painting +project @home @contractor
(B) Hire contractor for wall-mounted shelving +project @home @contractor
(C) Clean and declutter all rooms +project @home @Cleaning
(C) Remove old wallpaper from bedroom +project @home @Demolition
(B) Sand and refinish wood floors in hallway +project @home @contractor
(C) Reorganize closet space +project @home @organization
(B) Arrange professional upholstery for old furniture +project @home @upholstery
(B) Install new light fixtures in the living room and kitchen +project @home @installation
(C) Add decorative accessories (pillows, throws, artwork) +project @home @decorating
(C) Plan and buy plants for interior decoration +project @home @decorating
(B) Assemble and arrange new furniture in living room +project @home @assembly
(A) Repaint bedroom walls +project @home @Painting
(C) Reorganize bookshelves and display items +project @home @organization
(C) Install new mirrors in hallway +project @home @installation
(B) Finalize room layout for optimal flow and functionality +project @home @planning

[fredofredofred]

Referencing todo.txt wasn't enough, adding "Indicate priority on each item using letters in brackets before each item (A), (B), (C) etc. Add projects tags +project and suitable context tags @context"...

It also decided we'd probably already completed the first four tasks... so I told it to omit that bit.

[ransome1]

@fredofredofred when sending these kind of requests to an LLM we can explicitly pass through the context. In theory we can literally pass through the whole todo.txt documentation. So I am quite positive, that the results will come back with todo.txt attributes properly applied :)

So I think we have already identified 3 lose use cases:

1. You click on a button in the list of existing todo. It opens some sort of window with the list of possible follow up todos. Next to each todo there can be a button Add to list.
2. You can type in something into the text field (within the open dialog), for instance I need to redecorate my home. The results could then either be selectable with help of a button, or simply put into text field, where each line represents a todo:

3. You can drop a document into a drop zone, the LLM tries to come up with todos, which derive from that file. Propably the hardest one to implement, but if this works, a really nice `show case`.

[fredofredofred]

1 & 2 make sense.

For 2 I think I'd often be inputting an existing task which was too abstract to action. I guess it could be implemented as a button when you're creating a task to add more detail/develop it - then I wouldn't end up with that in the first place.

I can see use case No. 3 being useful when people get instructions/recipes, or with meeting minutes/emails. However it probably needs more context in each case - that the items you want from the recipe are for your shopping list, or that you only want the actions assigned to 'Fred' from the minutes.

Potentially it could allow the user to write the instruction (with some prompts) and paste in a document.

I guess the whole strength of AI is it's so flexible, people would find their own use-cases.