Are there any little annoyances like that that you could use as a reason/excuse to write an app?

Yes, waiting on hold.

But I can't think of how to write a piece of software to wait on hold for me. There is an app that does this, LucyPhone, but I can't tell from descriptions what they are doing. Are they transferring your call (somehow) to somewhere else, freeing up your phone, calling you back when they get a person, and transferring the call (somehow) over to you? That would be beyond my resources obviously -- I want to do something that requires no backend/server-side crap.

I guess in theory you could listen to the audio stream coming from the on-hold phone and try to detect when it stops looping and someone picks up but that seems like the sort of thing that wouldn't work.

see it for what it is. A very small laptop. Lousy keyboard, sure, but lots of other advantages (like being with you at more times)
Think "how could I have solved this everyday problem, if I had the information power of the internet, coupled with quite some calculation performance in my hand.?

You would think you could but I am not really coming up with anything that doesn't require a massive back-end like a maps app. I'm trying to come up with something that would be useful that's implementation is only the executable running on the device.

I think I do have an idea now ... although it doesn't really solve a problem, but I think it is something that people would like and it isn't in the App store: there is not a good raster-to-vector converter targeting people's normal photographs in the app store. There are a couple apps that are targeting the sort of raster-to-vector that is like scanning documents without OCR, usually referred to as tracing or auto-tracing i.e. I have this pen-and-ink concept art that I want to get into Illustrator, let me take a picture of it on my phone and and run it through this app ... This kind of raster-to-vector is covered.

The other kind seems not to be; e.g., do you want to see what that picture I just took of you would look like as a cartoon?

I have been thinking about a raster-to-vector algorithm for years basically:

1. First pass of Perona-Malik anisotropic diffusion filtering
2. Image segmentation from here (minus the Gaussian blur pass that they are doing because 1. above will be better)
3. Then for the raster blob to vector blob converter, I'll roll my own or use potrace.

and this is going to be my excuse to finally write it, but I don't know 1., and 2. are quite computationally intensive, especially at the resolutions that phone cameras have these days, would people mind waiting 30 seconds? Actually I guess I could just down sample the picture first.