I’m currently working on something related to AI in software development teams, and I want to hear how developers in this forum are using AI for coding these days.
I thought the general sentiment was more negative than it seems based on:
My company pays for Cursor, and every time I switch off of it, I feel like something’s missing. It definitely has some issues, but it makes life much easier.
Wade said:
My company pays for Cursor, and every time I switch off of it, I feel like something’s missing. It definitely has some issues, but it makes life much easier.
I feel the same way; it’s hard to go back once you’ve been using it. Do you also use ChatGPT or Claude on the side, or is it just Cursor now?
I stopped using Copilot and just use ChatGPT now. I can’t afford $20 a month for both. Whenever I can’t be bothered to think and know it’s something AI can handle with minimal input, like formatting data or creating basic functions, I use ChatGPT. Copilot doesn’t remember previous messages, so you have to retype your original question if you need small changes to the first output.
@Fin
Exactly! That’s also a pain point for me with Cursor. The real problem with ChatGPT and Claude is that working on the prompt and finding stuff from different files just takes up too much time, which makes certain use cases less efficient.
I’ve been getting used to Cursor lately, but sometimes I feel like I’m spending more time trying to make it work the way I want than actually gaining any time from it. I’m also working on producta.ai, which works really well for teams using Linear.
I use AI daily but not for everything. The biggest issue is that the AI doesn’t have our entire codebase as context. If it did, it would be incredibly powerful, but for now, it’s limited.
AI is super helpful for some problems but useless for others. Figuring out which problems fit into which category is a skill in itself, and learning how to craft good prompts is another skill that can make the AI more useful.
But eventually, you hit a point where you try using AI for a problem and it fails just enough that you keep trying, and before you know it, you’ve wasted time. So learning when to stop using the AI is another important skill.
Here are some ways I use AI:
Creating test boilerplate for a code module
Writing the rest of a set of tests based on examples
Writing documentation for functions and APIs
Crafting regex patterns and documenting them
Refactoring HTML into MaterialUI components
Adding small features to a website
Extracting code into helper functions
Typing complex functions without referencing TypeScript docs
My company gives us GitHub Copilot and M365 Copilot. I find both very useful, but I treat them more like a smarter Google search than anything else. You shouldn’t expect them to do everything for you, but as a fast way to get information, they’re great.