"Code is poetry." That’s what Matt Mullenweg, co-founder of WordPress, once said. He described code as something elegant, intentional and crafted with care.

But in Andrej Karpathy’s words, former Tesla AI lead and founding OpenAI researcher, “Software is changing again.” We’re not just hand-writing every line anymore. We’ve moved from coding things ourselves, to training models, to prompting AI in plain language.

If software is now written by prompting instead of typing, is it still poetry?

When writing code feels as simple as writing a sentence, where does craft end and convenience begin?

Those questions followed me into my first months at Greydient Lab, where thoughtful design isn’t just a value, it’s a habit. Back then, I used to write code the long way. But as I learned to work with AI alongside my team, I realised it wasn’t about replacing that process.

It was about learning to code with AI as my coding buddy.
Not through strict rules, but through small habits I picked up:

• Prompting with intention

• Breaking things down

• Treating AI’s output as a draft

• Always owning the final result

Start with the Problem, Not the Prompt

I came to AI with questions I hadn’t finished thinking through, hoping it would fill in the gaps I hadn’t even named. But without knowing the real problem, even the smartest solutions can feel empty.

Now, before I even open AI, I pause and write down exactly what I’m trying to solve. I step through the code, check where it breaks, and ask myself what’s really missing. 

Is it a logic error? A syntax issue? Or unclear flow? Not just “fix this bug,” but why it’s happening and what I actually need. If I can’t explain the questions clearly, I know I’m not ready to prompt yet.

Chunk It Down to Level Up

At first, I’d send AI the entire problem in one go, like handing over a tangled ball of yarn and expecting a perfectly knitted jumper in return.

The results? Messy and unclear. Since the AI couldn’t fully understand the main problem, it just added unnecessary lines that complicated the logic.

I learned to break things down. Just like object-oriented programming teaches us: smaller, focused parts are easier to reason with. Shorter prompts make AI more useful, and they force me to be clearer about what I’m actually trying to build.

Don’t Settle for the First Draft

Just because the first solution from AI runs doesn’t mean it’s the right one. I used to paste whatever worked, only to realise later it didn’t fit the project or could have been written more clearly.

Once my buddy gives me an answer, I pause and review it line by line. I check if it matches the project’s structure, if the logic makes sense, and if it’s something I would actually write myself. If it feels messy or unclear, I rewrite parts of it instead of forcing it to fit. AI gives me a starting point, but finishing the job is still mine.

You Ship It, You Own It

AI can write code, sure, but it doesn’t understand the project like I do. It doesn’t know the trade-offs, the constraints, or the goals behind each line.

Now, after using AI, I double-check everything before moving on. I test it in real scenarios, and watch for side effects AI wouldn’t catch. I ask myself if it really solves the problem or just looks finished on the surface. At the end of the day, if something breaks, it’s on me and not the tool I used.

Conclusion: Thought Still Matters

AI can code fast. That gives us something better than speed, space to think slowly and surely. What I’ve learned is simple: craft still matters. Not because it slows us down, but because it keeps us honest. The clarity, the care, the curiosity — none of that comes from a prompt, it comes from us.