LLM, Subscriptions, and Tokens

by Rian

As I have previously said, I don’t see Claude replacing a proper programmer’s job anytime soon. I am subscribing to meager Pro tier, and that’s nowhere near enough to replace my hobbyist’s job. Granted, there is the joy of not needing to fiddle with technicalities. Being able to implement new algorithm and upgrade existing core algorithm to a new one like blocks falling into places feels like a vindication. But this is from a hobbyist’s perspective.

From my experience, Claude, outside of Claude Code, was too conservative in its answers. One could argue it is avoiding false positives, AI hallucinations, as much as it can. Truth be told, I find Claude’s general search algorithm (or general googling) to be a lackluster at best. Once, I asked if Claude could identify whether the ripped media I have is a stitched version of several episodic releases. Claude gave me general direction how to search for metadata for medias. What’s the point of asking a LLM agent if I were to do it myself? At the very least, it should have given me more interesting DBs to search from. But asking generic questions and getting unhelpful answers don’t cost tokens as much.

Claude burns through “usage”, its own branding of LLM tokens, during vibe coding. I suppose I have to specify what branch of “vibe coding” I subscribe to. I prefer to know what goes in, what will be done and how, and what the output will be. For obvious reasons, it can’t produce complex codes reliably — maybe I need to fashion a better prompt for that. But when I am simply creating a pipeline from one system to another, ones that would have costed its own “subscription”, the question of LLM affordability shifts a little.

The Pro tier of Claude currently charges $200 per year, and some of the “pipeline” softwares I had just described could dwarf that bill easily. Take, for example, Paperless-ngx which I had covered before. I was coming from DEVONthink, a great software and would still recommend, but I simply cannot justify the new subscription model and the added bells and whistles I don’t necessarily need in a personal DB software. Paperless-ngx doesn’t fully replace it either; Claude merely introduced it and helped with the customization for Docker. But on my end, from the perspective of one who is paying, Claude is quickly paying itself off by eliminating competitions.

Outside of programming context would be office suites. I have stopped subscribing to Microsoft Office suite in hopes to use Apple’s free iWork instead. One of the major downside of Apple’s was the lack of sophisticated functions from Excel. Fret not, Claude can actually write a function that will mimic exactly what Excel’s one liner of a function was capable of. It’s essentially the same case as DEVONthink. I was not using all of Microsoft 365, so I found it to be extravagant, too many bells and whistles. Free software suites like Apple’s, Libre Office, or Open Office may not be enough to unwind the lock-in effect. But Claude and any other LLMs can be that filler. Knowing how to apply is up to the user, however.