@Wylder
“Write some xunit tests for this class”
Then I tweak it since AI often misses the mark, though it speeds up simple testing a lot.
Milo said:
@Wylder
“Write some xunit tests for this class”
Then I tweak it since AI often misses the mark, though it speeds up simple testing a lot.
For me, writing tests is about confirming my understanding of the system. So when tests are auto-generated, I don’t grasp the code or the tests.
Basically, I see it as useful for increasing coverage but not much else.
Also, writing tests helps me check if my API is user-friendly. Tests should be simple to write, and every line should convey clear information. If writing a test is hard, I need to rethink before proceeding.
Milo said:
@Wylder
“Write some xunit tests for this class”
Then I tweak it since AI often misses the mark, though it speeds up simple testing a lot.
Exactly, what we call “monkey tasks” at my workplace. Those dull and repetitive tasks that no one wants to tackle. AI is handy for generating things I know how to do but don’t want to spend time on. I wouldn’t say that’s true programming; it’s part of it, but AI is merely speeding up mundane tasks. Or maybe I’m just not using AI right 
@Wylder
I use AI often with my code, but I have a helper script that creates a long prompt with the relevant files, project structure, and coding rules.
If I keep my requests focused on one aspect, it usually does a decent job of producing new and modifying existing code. If the output is poor, adjusting my prompt often helps fix it.
Wylder said:
@Sky
Can you share more details? Sounds interesting
Sure, here’s a simplified version I shared earlier:
https://gist.github.com/thomasrice/4739593e7f43534ddfbf83599fa76b4f
I just update the include_tree, fasthtml_code, files, and goal sections, then paste the whole thing into my Python console to create a larger prompt, which I then copy into Claude 3.5 Sonnet.
Naturally, you’d tailor the code to your project.
I tried Cursor before, but this approach has worked better for me.
@Wylder
It’s great for APIs in my case.
And for some simple tasks that I don’t want to deal with.
If I can test its outcomes, it’s pretty effective.
Sure, sometimes it takes a few back-and-forths, but ultimately it performs quite well.
@Wylder
I relied on it heavily not long ago to write a Roslyn analyzer.
It saved me considerable time with trial and error as it’s not my area of expertise.
@Wylder
Honestly, I believe it comes down to how you’re using AI. Naturally, you can’t just drop a one-sentence request and expect it to generate a complete, flawless app for you. In my experience, AI performs much better with a solid foundation provided. When I first learned React to build a web app, I got the best results by sharing my current code, explaining its issues, and detailing what I wanted the code to accomplish. Using this structured approach, I find the output quality improves significantly, and I have the choice to modify it however I see fit.
@Mal
Maybe I’m just not using it right, but how am I supposed to break down my several hundred files project linking to other teams’ microservices for such an AI? Have you truly received any valuable feedback from an AI? Like insights on optimizing those microservice calls or avoiding memory leaks in your code? Right now, AI only assists me with tasks I already understand, like ‘set up a boilerplate using Tailwind and React for a login page using my provided guidelines’. Basically, it’s the same as what I did with Google before AI.
@Wylder
There’s a small cluster of “non-technical” SaaS founders who have launched AI-powered products for them.
Does that count as production 
@Wylder
I use AI daily but not for actual coding. Like recently, I shifted some components from React context to Zustand, a library I haven’t used before. I picked up the concepts I needed quickly.
Jai said:
@Wylder
I use AI daily but not for actual coding. Like recently, I shifted some components from React context to Zustand, a library I haven’t used before. I picked up the concepts I needed quickly.
Same, I find it useful too. Just not for real programming. Usually, I ask for basic algorithms or isolated functions, then I fix the messy code it generates.
@Wylder
Have you tried using Cursor? It has significantly boosted my productivity. Mainly in writing boilerplate code and refactoring.
@Wylder
I don’t completely align with your question, but I’ve utilized ChatGPT extensively in a personal project to display images on a Raspberry Pi. Most of the code is my own work, but for minor, annoying tasks like converting RGB to grayscale or other image-related functions, I often have ChatGPT create those functions for me.
I’m also working off a manufacturer’s existing code which is poorly documented, thus utilizing ChatGPT helps in questions like, ‘How many bits does this code use per pixel?’. It allows me to focus on the challenging parts instead of getting bogged down by minor tasks like byte alignment.
I’ve stepped away from daily coding; it’s now just something I do for fun. Before I took a break, the enjoyment was already drained by constant micromanagement, pointless meetings, and the abuse of the agile process that took away time to think and design.
AI is here to further diminish your value as a skilled worker. Management has long believed that any project can be tackled by one lead and a bunch of generic programmers. Now they want to replace part of that with AI.
I don’t enjoy wiring generic CRUD APIs, so I created a process to automate it. I can see AI playing a role there. But make no mistake, that line will blur, and you’ll notice AI taking over the work you love.
For long-term security, go deep. Don’t specialize in basic website development; find a niche where you combine software development with another skill like financial services, highly embedded systems, drivers, or real-time operating systems, especially at scale. Now is the time to get prepared and secure that position. Dig in and focus.