Looking at learning more javascript and/or python during the summer. I am using vs code. What AI would you recommend? Is the free tier enough? Or do I need to pay to play?
1 Like
If I was learning my first programming language, I would probably start with free (donation supported) Khan Academy
Then, to make programs that use AI like LLMs I would explore taking free course(s) on Best Large Language Models Courses & Certificates [2026] | Coursera
Depending on your computer, you can run free AI models locally (ollama), that may not be great, but will probably be good enough to get the gist of how things work.
This is what I’d do if learning was the goal.
If my goal was to turn a specific idea into something real for people to use asap, then, I’d do things differently.
Personally started off using free LLMs, ollama/lm-studio, free chatgpt, then ChatGPT+ $20 per month, and now I’m hooked, and spending more… Personally using vs code github pro+ which gives you access to many models, but their billing model changes tomorrow that will 10x my bill, so am looking at other options (claude max, openai pro or ???)
2 Likes
Thank you for your reply. I have a specific idea that would be nice to have running in 3 weeks time. What do you think about chatGPT+? I have seen a lot of hype about claude. I haven’t really had time to look at their billing models but I guess I could spend $20-50 for one month.
With claude code, you can also pay as you go with API pricing. At some point the subscriptions are a better deal, but given there’s months where I don’t use it at all API makes sense for me.
If you use it to learn and not write your code for you (in which case you won’t learn anything) you probably won’t use much.
1 Like
Could you give an example of how “much” something costs?
If you want to actually learn those languages, no AI…
This is a very different objective
If you don’t look after learning the languages, but having a project “done” by a certain time, yeah, Claude or Codex can do that for you… or at least they’ll make “something”…
Just be aware that by the time the project is “done”, you’ll probably still don’t know anything (or more) about the language it’s programmed in.
1 Like
Not really, depends on which model you choose and I haven’t used it for just learning (where you presumably have rather little output tokens used)
I think chatgpt+ will be what I try first!
Has anyone tried boot.dev? It seems like a different approach to learning something new.
Do yourself a favour and install Codex with an IDE plugin, it’s much easier and fluid to use
1 Like
I’ve been playing with the free version of Claude. The people I know doing it professionally are using Claude Code. I think it’s ~US$20/mo or something for the paid version that includes code. I don’t know how the tools in Code are unless you’re starting from a relatively hardcore integration point.
As above, if the goal is to learn the language, start without AI and see how far you get, you can always bring it in later. I’d then start with having it give you examples of how to do things and use those as a template to write stuff yourself.
If you’re trying to just make a specific thing, by all means dive in and give it a shot but, fair warning, it’s 90% good, 10% laying traps for you. I’ve had it remake 3 small projects that I did myself and in each case it has done well for the majority of it but made boneheaded mistakes that may have taken me a long time to identify if I hadn’t spotted them or wasn’t already proficient in coding.
4 Likes
I use Claude code both professionally and personally. I’ve used it to help write a few personal apps over the last few months for projects I’ve always wanted to do, but never had the time to hand code. I don’t come from a real programming background, but have learned a few languages over time just due to the nature of my work.
As Jono mentioned, Claude does well at getting an app spun up fast, but I’ve read a few horror stories about the code it’s produced (and have seen a few in my personal stuff too). Claude is great for POC or personal apps that will live behind a firewall. Work is using it more and more for helping to update/manage/write code, but we also have teams of developers that PR everything before it goes to production.
AI has been interesting for coding. In some cases, it has produced bugs that a human has caught, and in other cases it has caught bugs that humans have written.
I’d be hard pressed to put something into production in 3 weeks with it if starting from scratch for both the project and coding knowledge.
If I was starting from scratch, I’d probably start with the $20 claude plan and claude code. VS Code has a plugin you can use to run claude code from inside the app, but I still tend to run it in a separate terminal (i’m weird).
The hard part you’ll have is not knowing if the code it’s producing is valid, or full of bugs. You’ll have to do a lot of QA/QC type testing of the entire app including using it in ways you’re not supposed to to try to break it on purpose.
1 Like
I developed for FreeCAD and the last weeks that I was doing that, I used a lot of Copilot, it was very cheap like 20 euro or 10 euro a month. Almost never hit limits. But as soon as you try to contribute vibe code stuff people gets annoyed so might want to not do that.
I now got Claude hooked to my home assistant, and I am full confident that coding by hand is soon a thing of the past. Only devs who are really experienced won’t have issues as someone needs to check the AI. But low level coders are completely jobless. Learning to code is like learning math without a calculator, it’s only useful to understand what you are doing. But as soon as you graduate you never do manual math again. Unless you are into that stuff.
Just go into woodworking no AI there to ruin your party.
We’ve already seen forum members here use AI to transform a photograph into a carving and cut that. One is a picture of Ryan sitting in the 2025 RMRRF chair that Dave made, turned into a carve.
Even after the AI back end data/compute cost bubble pops, there will remain uses for machine inference, machine learning, and the variety of modeled tasks that are emarging today that we sloppily call AI.
3 Likes
I also use AI for Instagram posts and writing mails, a quote on a cutting board.
Not too design my products, but I can do a brainstorm session with AI. It’s not that’s better then doing it with a co-worker. It’s that I don’t have a co-worker 
.
But coding is all vibe-coding which is just asking a machine to make something, which won’t be more then asking a coffee machine to make a coffee in the future.
1 Like
It’s a lot more nuanced than that, and we are a long way from not needing coders.
“Vibe coding”, asking AI to code something you can’t code yourself, is completely different than increasing your efficiency and supplementing the knowledge you already have.
There’s a reason people reject vibe coding. AI with no human safeguard tends to produce a lot of unmaintainable garbage.
You want to vibe code a small personal app for a few people? Yeah, it’s great at that… but in my line of work where we maintain enterprise-class applications with millions of lines of code in total, it’s not so great, and for a good part of it, has made people less efficient, not more.
There’s good and bad to it, but as someone who has used it a lot for more complicated things, we are a long way from replacing software engineers. Quality and cost are major factors still that make it prohibitive
5 Likes
I have been using VS code with githuib co-pilot which has a range of different models it uses. With most AI models you will get a certain allowance to get you hooked and then have to pay.
Different models are better at different things, and certainly claude code is a good option.
Also I would say the models are getting better all the time, and the agentic approaches to plan the coding are certainly streamlining workflows - although there are times I find they are counter productive as well…. so use with care as others have said
^^ this is the correct answer hands down.
I have been using Google Antigravity. I think it really got a bump with the new Gemini Flash models. I have the subscription which is 225 SEK (ca 24 USD), it gets me along nicely. They quite recently split the app into one “agentic” window which is more like a chat you might be used to from ChatGPT, but is aimed towards coding. You also have the Antigravity IDE which is a fork of VS Code, but with Gemini in it.
I really like how Antigravity wants to do an implementation plan, before starting actual work. I usually comment on the implementation plan, since models tend to assume stuff when they don’t know what you want.
My two tips for agentic coding is:
- Be clear about what you want. Long descriptions are good!
- Don’t waste your back pressure. If you are copy pasting errors, you are doing something wrong. See: Don't waste your back pressure ·
i dont have them finished yet the finished clock is the picture I used to get the depthmap for vetric to cut the replacement for my clock
3 Likes
I use Cursor both personally and professionally since 2 years ago now. It’s a fork of VS Code so all your extensions work, $20/mo for a few hundred requests per month and then you can pay as you go, and you can use whatever models you want (having tried them all, Claude Sonnet and especially Opus are consistently the best at coding in my use cases).
For my uses however, I’m not doing enterprise level maintenance etc, but doing data science experiments. Having the LLM to bounce ideas off of and to have it search for things is a huge efficiency boost. For example I might have it search for latest and greatest object detection models, and then have it set up a testing pipeline where I run several models on the same dataset, and while it offers up its statistical analysis, I look at the actual results with my eyes. More often than not, I catch something it missed or I come to a different conclusion than it… which is fine for me, I’m already an “expert” in my field, a life long computer nerd, and I completed the basic and intermediate Python courses in DataCamp before diving into the AI coding assistant world. So I’m basically harnessing all the “good” things (efficiency, brain storming, multi tasking) and rejecting all the “bad” things (keeping tasks focused, scrutinizing the results, etc).
It’s also fantastic for helping me learn. Not only can I ask it questions like “why is this step maxing out my RAM and taking forever to process?” But when I have it make a Python script to run multiple tests (like many different models or many different parameters of the same model) on the same dataset, the trends become obvious and you can know exactly what those different models/different settings do. But again the caveat is that I’m already an expert and am putting in a lot of effort to understand what the LLM is doing at every step of the way. If you just let it cook and assume it did a great job, you will crash and burn at some point and be stuck in a world of confusion…
1 Like