First Half of 2025 is Ending, So Time for an AI Subscription Inventory

Introduction: Hello, I'm Hosaka, an XR/AI engineer.

The 15th Development JAM was held on an irregular schedule due to my day job (work to earn living expenses), but since the purpose and the deadline had strong coercive power this time, I managed to achieve my goals.

However, I can't say I'm entirely happy about that. When there's something I want to do more (XR) but it gets pushed back due to responsibilities, it causes a certain amount of stress or burden. I also made a blunder at my day job, and I'm in a situation where I can't keep going without taking a break.

That said, relying on alcohol to blank out my consciousness like I used to is no longer an option, as I've fortunately stopped being able to shake off headaches when I drink lately. I want to keep my mind clear at all times except when I'm sleepy, so this is a good thing.

Well, since it's Monday, I'll do my weekly feedback, and maybe I'll have a proper break and let my guard down late Tuesday night. (While watching Sieg Ax...)

Development: ClaudeCode, I should have done this sooner (I'm sure hundreds of millions of people agree).

Before using ClaudeCode, I dismissed it, thinking I was fine with VSCode's Cline or Cursor, but once I realized it could be installed with a single command if Node.js was in WSL, I tried it and realized it was a completely different beast.

It's like telling a taxi driver to 'go to XX' and giving directions like 'turn there' while heading to the destination. You never take the wheel yourself, and the fact that you're moving while still being a passenger is similar to a taxi in the sense that you can see the scenery (pseudo-driving). By the way, since you can't see what Devin is doing, it's more like a UserEats order.

And the advantage of a taxi—that you can sleep (do other things) if things seem generally okay—is also similar to ClaudeCode. Of course, if the driver gets lost or I know they're clearly taking a different route, I'll give detailed instructions, but as long as the direction is generally correct, saying 'Yes' to go straight until there's a fork in the road is also like a taxi.

Thanks to this, it spits out code for convenient Windows native apps that I usually wouldn't even think of because they're a pain to make, all in about an hour. I just have to build and check it in the Visual Studio I have open, which is a king-like treatment. If I have ideas, Windows seems like it could become even more convenient.

Turn: Generative AI subscriptions, who is the useless one (AI)??

ClaudeCode, which is useful even on the Pro plan, is worth considering upgrading to the $100 Max plan, as I can create 'infinite' apps as long as I have time for parallel work.

My flow is starting to take shape: generating and organizing ideas (Perplexity/NotebookLM), creating app design prompts (ChatGPT-o3), and making visual, audio, and video parts (4o images, Google AI Studio, Clipchamp, Photoshop). The rest is whether the output is for Windows, smartphone, or Web.

However, since I'm not earning income from the programs I'm making right now, generative AI is, in a sense, my own employee benefit. Therefore, I can't pay for generative AI infinitely, and I need to organize what I'm not using or what isn't useful.

The subscriptions I'm currently using, in order of importance:

1. Azure (OpenAI)

2. Google (GCE, TTS, NotebookLM)

3. OpenAI

4. Claude

5. Replit

6. Perplexity

7. Cursor

Of these, the one I definitely wouldn't miss is Cursor. I'm not really using it anyway, and it can be 100% replaced by ClaudeCode, so I'll cancel it when I get home.

Replit, the cloud development environment, is also on the bubble, but the appeal of being able to run it from various devices over the internet even while developing is huge, so I can't let that go either. Perplexity was also questionable, but its ability to pull in the latest information using an API is a unique strength, so I'll keep my subscription until a competing product comes out.

If I upgrade ClaudeCode to the Max plan, I'll end up needing over $300 in total. Well, I guess I should think about ways to make money...

Conclusion: In any case, this week I need to keep delivering ideas for ClaudeCode to work on.

Having been programming for over 40 years, there are infinite pieces of software I wanted to build but couldn't, or wanted to buy but couldn't, so the fact that ClaudeCode provides maximum output at minimum cost is almost like a secret gadget from Doraemon.

It's often said that the apps you want already exist in the world, but they always include points you have to compromise on, so there's no point in listening to other people's opinions. Honestly building the app you want, custom-made for yourself—that's the best.

The annoying thing about using ClaudeCode is that prompt input is in the terminal, which makes it hard to type, so I'd like to think of something good since I can even build an input app for this. Since it runs locally, maybe I could be happy if I build it with Node-RED.

So, I'm tired this week too, but it seems I was able to get things done and feel fulfilled. I want to quickly streamline the ClaudeCode cycle to create more of my own time and be able to pour all my resources into XR. Let's do this!