Have you started integrating AI tools like GPT-4 into your development process? Share your experiences and insights into how AI is changing the way you work.
Follow the DEVteam for more discussions and online camaraderie!
For further actions, you may consider blocking this person and/or reporting abuse
Top comments (12)
I guess as probably most of people, I use GitHub Copilot, sometimes gpt but usually not related to code.
Unless coding within a very restricted stack perfectly mastered, there is just too much libs/methods to remember, and this kind of tool used as a searching shortcut seems just an evidence to me.
I don't use it at all. I understand that I'm rejecting new and possibly helpful technology, but I prefer my code to be human written. Stack Overflow and documentation solve 99% of my problems anyways.
I don’t use it much in my day-to-day work, but for things where I’m not as comfortable, it’s much better than trying to figure out the right string of keywords to get what I need from Google.
I’m also not usually looking for a complete solution…a partial solution I can build off of is my preferred method. That way, I get the push I need, but I also build my knowledge base and don’t have to ask next time I’m in a similar situation.
I would use it if I'm stuck where to start with a project AI would give me a decent brief on how i go about it. i also always state to AI don't ever give me code let me figure it out myself or google, Stackoverflow or if you dont understand a concept AI will literally explain it to you like your a 10 year old lol
Not really. I tried Copilot for a while, and I was really frustrated and disappointed. I used it for a bit over a year, and I found that more often than not it just made different busywork for me. Instead of it saving me time, I ended up having to tweak the result. The most common problem was that I couldn't get it to produce decent code, and I burned so much time trying to get it to that I lost track of what I was doing and I fell rapidly out of flowstate.
Removing distractions is essential to maintaining a healthy flowstate, and all of the AI tools I've tried so far are first and foremost a distraction.
I take full advantage of the extra productivity. However, just don't fall down the ChatGPT rabbit hole where you entirely depend on it for answers. Some answers are more easily obtained through google and stack overflow. Knowing when to use the right one is key.
What I'm really digging is that GitHub Copilot will auto-complete mis-aligned brackets and missing syntax which is a common issue that slows down developers. It's very good at writing comments, which makes the task of writing them a bit easier. And then if you're truly stuck, it can come up with an approach that you may not have thought of.
Can't wait until it is more integrated into IDEs, potentially reading your codebase to gain context and provide more complete answers, without DEVs having to be very specific with stating the question or problem to the AI.
No.
Not that I know of anyways. Our tooling, as provided by MS and other vendors likely has it ingrained in some way that is unclear or underdeclaired or obfuscated.
The coding challenges we encounter most are usually with unusually structured business data, unorthodox UI/interface demands from business, and often nearly impossible responsiveness demands (such as deliver live, aggregated, unfiltered, unpaged, data from a multi-terabyte data source from a data centre far from the client in a browser at a remote location in under 3s from request start to completion.)
((Also can you make it offline editable and looks good on an iPhone 4))
It does make the work easier. I mainly use it to refine the structure of any project that I'm going to work on.
I can define the problem in a very detailed way, and it provides an architecture on how things can flow. I further refine it until it meets the standards.
This isn't true for every project, but I certainly did it with designlyfe.tech/ (my buildspace product). It has 18+ APIs and a long list of features, so the architecture was huge.
Ultimately, I found out that the architecture could be improved.
I would say experience + AI makes a deadly combination.
I have started using AI to either help me solve a problem I'm struggling with or suggest ways to improve my code.
I'm using it as a pair programming buddy in many ways rather than using it and trusting it to produce code that I do not understand.
No, but I am finding some interesting things. The work I do, apparently, ChatGPT and Gemini (the ones I tried) are not helpful. Broken code and never fits the architecture.
However, a few times I used for some small single use scripts and it was pretty good.
I was super hyped about ChatGPT at the start but now it somehow became so bad (my theory is that they’ve put too many guardrails on it) that’s it’s often just a waste of time. It’s much faster to read the docs.
I find tebnine quite handy as it often guesses exactly what I'm about to do and makes a suggestion that is usually very close. Usually something needs altering but it is much quicker than having to think. I also like having it explain the code as it provides a rich and detailed explanation that goes way beyond what most humans I've discussed code with are prepared to explain leading to me getting a better understanding of how it all works.