If you can't define "Good," AI won't deliver it

Published on: 02/06/2025

That's right. I learned from my own experience. When I first started learning Django, I thought I could just toss some prompts at an AI, build my project on the fly, and learn as I went. Two days in, I was drowning in console errors and bugs I couldn't make sense of. the issue became really obvious at that point: I needed to understand at least the basics of Django if I wanted to be actually productive with AI.

So, I jumped into the official Django tutorial. I was surprised at how complete and clear it was. It's so good that running through it should take more than one day. With that foundation, my work sped up, and my AI prompts got a lot more precise. Sure, there were still moments when I stumbled on a tricky concept, so I went back to the docs, the forum, and even reddit. Honestly, it wasn't as overwhelming as I'd thought. Django has been around for over 20 years, so the docs and community are really polished and easy to understand.

Every time I learned something new from the docs, I went back to the code the AI had generated. More often than not, I discovered I had been heading in the wrong direction or overcomplicating things. My lack of Django foundations was making my prompts off, I was asking the wrong questions and I didn't even realize it at the time.

Here are a few lessons I picked up along the way:

• Even with AI, the basics matter. You can't just rely on AI to figure out everything. Without a clear understanding of what you're aiming for, you're just setting yourself up for extra work.
• Don't skip the docs. I have known this for a long time. But this time I naively thought, “I'll just ask ChatGPT about everything,” but that turned out to be a huge mistake and I am sure I wasted several hours banging my head against the walls.
• Know what “good” looks like. AI can spit out code, but it's on you to judge its quality. Without knowing the best practices, you might end up adopting some not-so-great habits without even noticing.
• As a software engineer, I don't just want to be able to generate code. I want to really understand it and have control of the application I am building, it's internal, it's structure... everything.

I've seen posts saying things like, “You don't need to know how to code. Just use tools like Cursor to build your prototype in minutes…” Sure, that sounds appealing, but it's a bit too optimistic and naive. Eventually, you'll hit an obscure (to you) error that only someone with knowledge and experience can fix. And when that happens, you'll truly appreciate the value of software engineers. Seriously, building software is no joke. It's a craft that takes years, even decades to master.

Here's an analogy: I can't expect to be able to write a scientific paper on the consequences of global warming. Sure, AI models are trained on some relevant data. But if I read an article like that, I would have no clue of wether the article is good or not. The same thing happens with software.

So, next time you're tempted to skip the docs and let AI do all the heavy lifting, remember: AI is a tool that you guide. If you steer it in the wrong direction, it won't warn you—it'll just get you there faster, and not necessarily in the right way. Your own knowledge and experience are what truly make the difference, and you'll notice that right away. By doing this, I have gotten much better at prompting and I feel actually more productive. I hope this is useful for you too.