Using AI for Coding: My Journey with Cline and LLMs

198 点作者 me2too4 个月前

13 条评论

Prompt quality and knowing your domain is critical. One issue I had early on was experimenting with LLMs to generate a frontend application in a brand new framework I was unfamiliar with (Svelte at the time) which lead to situations where I would cruise along and get stuck in a loop. The other issue came from the increasing context size that led to more unpredictable behaviors (i would ask it to change the color of a button and it would completely change the entire page).almost all the tools i've used to date for designing frontend framework, none really replaces using cursor and being able to dive deep, however cline does seem to have gotten significantly better.the day where you can come back to a fully working web app with moderate complexity after cleaning the gutter is still some way off but thats the dream

评论 #42848860 未加载

评论 #42848241 未加载

评论 #42847376 未加载

评论 #42848550 未加载

ItsBob4 个月前

Anecdotal but here's how I described using the likes of Copilot to my sceptical colleagues (they were late to the party!):It's like having a senior software dev over your shoulder. He knows pretty much everything about coding but he often comes to work drunk...And that was the best analogy I could come up with: I think it's sped up my work enormously because I limit what it does, rather than let it loose... if that makes sense.As an example, I was working on a C# project with a repository layer with a bunch of methods that just retrieved one or two values from the database, e.g. GetUsernameFromUserGuid, GetFirstnameFromUserGuid and so on. They each had a SQL query in them (I don't use ORM's...).They weren't hard to write but there were quite a few of them.Copilot learned after the first couple what I was doing so I only needed to type in "GetEmail" and it finished it off, (GetEmailAddressFromUserGuid) did the rest, including the SQL query and the style I used etc.To me, that's where it shines!Once you figure out where it works best and its limits, it's brilliant imo.

评论 #42852226 未加载

评论 #42852049 未加载

potatoicecoffee4 个月前

I like looking at stack overflow for coding examples and seeing a couple of nerds getting angry at each other about the best way to do stuff. or some interesting other ways of doing stuff. this is also why i come to HN! so its weird when people want to drop that small joy from their workday

评论 #42848054 未加载

评论 #42848176 未加载

评论 #42848546 未加载

评论 #42849179 未加载

lopuhin4 个月前

I find it strange that the author is really happy with the quality of string comparison here <a href="https://pgaleone.eu/ai/coding/2025/01/26/using-ai-for-coding-my-experience/#backend-development-insights" rel="nofollow">https://pgaleone.eu/ai/coding/2025/01/26/using-ai-for-coding...</a> and while it would kind of work, it's a very weird piece of code from ML standpoint, e.g. it's training a TF-IDF vectorizer on just two strings being compared, which at best won't change anything (unless the same word is repeated within one product), and is a super weird thing to do as for better quality you'd probably want to train that on some corpus, or not bother at all. And also it compare two strings as bags of words, which again is not the end of the world but maybe not what the author wants here, and if they want this then it's not the easiest way of doing it. So it's taking some things which can be useful when comparing texts (tf-idf and cosine similarity) but then applying them in a weird way which does not let them show their strengths.

mzhaase4 个月前

There is an enormous difference for me between inline suggestions and using chat.If you use chat it's work, and then you have to debug and understand something you didn't write yourself.Using inline suggestions is the closest I have come to plugging my brain directly into the computer. I type 5 characters and in 80% of cases the suggestion is character by character exactly what I would have written. It speeds me up enormously.

ivoras4 个月前

I did a similar thing but with backend-heavy code, and I agree with this assessment:> In particular, I asked ChatGPT to write a function by knowing precisely how I would have implemented it. This is crucial since without knowing the expected result and what every line does, I might end up with a wrong implementation.In my eyes, it makes the whole idea of AI coding moot. If I need to explain every step in detail - and it does not "understand" what it's doing; I can virtually the statistical trial-and-error behind its action - then what's the point? I might as well write it all myself and be a bit more sure the code ends up how I like it.link: <a href="https://www.linkedin.com/feed/update/urn:li:activity:7289241001691377664/" rel="nofollow">https://www.linkedin.com/feed/update/urn:li:activity:7289241...</a>

评论 #42845788 未加载

评论 #42846951 未加载

评论 #42850230 未加载

评论 #42847189 未加载

numba8884 个月前

Looks outdated by now. Try R1.Seriously, I've been using LLMs for coding for a while and can say early experience was disappointing, but they get better and better fast. The latest o1 looks a lot better than 4o. It's reasonable to expect with proper human supervision and interface they will be able to handle big files and projects in a year or two. Interesting times...

评论 #42848865 未加载

outside12344 个月前

This is a pretty basic write up. Is there anything out there that does a survey of all of the things people have found to be successful in an engineering system for a project?(eg. Github Copilot for PR reviews, etc.)

评论 #42849367 未加载

cyanydeez4 个月前

Anyone got a line on a local first AI Tooling. Happy to pay100$. Don't want a subscription.

评论 #42847563 未加载

评论 #42846477 未加载

评论 #42845893 未加载

评论 #42846254 未加载

评论 #42848151 未加载

评论 #42847782 未加载

cube22224 个月前

Yeah this seems to be similar to my experiences.I use Zed’s AI assistant with Sonnet, and will generally give it 10-20k tokens of sample code from elsewhere in the codebase, shared libraries, database schema, etc. and more or less have a very specific expectation of exactly the code I want to get. More often than not, it will succeed I’ll get it faster than typing myself.However, it’s also pretty good at poking holes in your design, coming up with edge cases, etc. Sure, most of its observations will likely be moot somehow, but if it lists 10 points, then even if only 2 are valid and I didn’t think of, it’s already valuable to me.I’ve also used Cline a bit, it’s nice too, though most of the time a single run of Claude works just fine, and I like Zed’s AI Assistant UX (I actually don’t use it for coding other than that).

评论 #42847383 未加载

koinedad4 个月前

Was this blog also written by LLM?

8thcross4 个月前

This was my finding as well - I now use Roo Cline tho....

SunlitCat4 个月前

Well, my experience with using an LLM (ChatGPT) for coding in a nutshell:(asking Chatgpt after getting a very cut together looking example-ish example):Me: You simply read various examples from the D3D12 documentation and mixed them together without really understanding them? Admit it! :DChatGPT: Haha, I admit it, that was a bit of a ‘best of DirectX 12 documentary’! But hey, I tried to build you a solid base that covers both window handling and the basics of DirectX 12.