yesterday at 8:58 PM
While I was making a little app that converts text descriptions of events into calendar events[1], I was thinking an LLM could do this: consumes tokens and slower, but can work with even messier text descriptions.
I actually used an LLM to add coverage for the various ways different people represent dates and times in English (actually mostly Korean right now.) LLMs are even great at extracting data from text that is full of misspellings and ambiguous wording. I just give the LLM examples that were not parsed as expected. The LLM figures out the correct output itself, then adds logic so similar text will be parsed correctly in the future.
We've already seen apps that added an AI chat to their interfaces. And other apps use AI to generate "home-cooked" personal apps.
I think we will see a convergence of this: where users can chat with "malleable" apps to create features in the app itself.
So personal assistants will be creating the next generation of apps. Apps that are meant for a single user and/or even only intended to be run a single time before being thrown away.
I proposed this type of UX for Kagi; it could be applied to many types of software: https://doggo-pitch.leftium.com
[1]: https://nl-event.vercel.app, https://github.com/Leftium/nl-event
yesterday at 5:40 PM
I think a chatbot or voice is a horrible interface for most things. There is 0 chance for discoverability, so the application's function is limited by the user's imagination. Most people aren't going to think about the various things a program could solve for them. Even with a UI, most users only use a small fraction of the available options. Without some visual cue to let them know features exist, I'd bet that most users would use even less.
My day job is automation stuff within a large organization. Most people, even within a technical organization, don't seem to think about how the computer can solve problems for them. The other day I used Claude to turn a massive csv dump of data into a dashboard to see trends over time. I don't see other people doing that.. but if Excel had a button for it, I'm sure they would do it.
yesterday at 9:17 PM
Cloudflare actually ran into the opposite problem: their employees' creativity was distracting them from using AI in a productive way.
The answer to both too little discoverability and too little/much creativity is to deploy the chatbox/voice UX with "champions" that can help others use the new UX effectively.
- https://blog.cloudflare.com/how-we-use-ai-with-cloudflare-os...
- https://www.inkandswitch.com/essay/malleable-software/#commu...
> Studying spreadsheet use at companies, Nardi and Miller found "local developers" – amateur enthusiasts who could write bits of code for their coworkers and guide their team up the gentle slope of spreadsheets.
Interestingly, Cloudflare solved the excess creativity problem in the short-term by changing the UX from a magic chat to a magic email. The email could only input work you wanted to automate, and the only output was the finished work. This helped the employees' requests become more focused.
yesterday at 11:42 PM
Instead of making LLMs create scripts behind the scenes for what you want to do, it is more efficient to have LLMs use apps for you. Apps still make sense to me.
today at 12:11 AM
Do you think we will keep making new apps in the future?
Or will we stick to incumbent solutions for the most part?
yesterday at 4:51 PM
Some will, some won't. I prefer some webapps to their native ones and vice versa. I used to run some proprietary apps that I replaced with open source ones, often clients interfacing with already self hosted stuff. I rarely touch Kagi anymore because my own hosted model + searxng drops results faster. I don't expect most of these apps to go away just because of agents/assistants alone.
Proprietary apps and services will probably go into protectionism and will shed users in doing so.
Custom launchers and widgets will likely be replaced by user instruction-driven processes at some point.
yesterday at 4:19 PM
Some of them for sure! It's so much easier for me personally to just tell the currently-preferred-LLM-model or agent to do something for me instead of dealing with apps and dashboards!
yesterday at 7:35 PM
[flagged]
yesterday at 4:20 PM
[dead]