I started using AI while I was still in university. At first it was just a faster Wikipedia — quick research, concept lookups, replacing the usual tools everyone in our field relies on. Nothing revolutionary. Just faster.
Then Copilot came out. And that's when things got complicated.
I hated Copilot at first
It was invasive. I'd be in the middle of writing a function and this ghost cursor would jump in with suggestions I didn't ask for. It broke my concentration. It guessed wrong more than it guessed right. I turned it off more than once.
But it kept improving. And more importantly, I started understanding what was actually happening — and what I was doing wrong.
I was treating AI like a magic autocomplete. I'd give it vague context and expect perfect output. That's like handing someone a blank napkin and asking them to build your house.
The real shift: from coding to planning
Here's what I learned the hard way: AI doesn't work when you throw ambiguous requests at it and hope for magic. It's not about expecting something to appear overnight. It never was.
It's about planning, strategy, and documentation.
I went from spending all my time writing code to doing what PMs and architects do — breaking things down into tasks, defining features with clear specs, mapping out the logic before touching a single line. The difference? I'm not planning for a team of developers. I'm planning for an AI agent that will execute exactly what I define.
The key insight: don't expect AI to invent things for you. Give it the steps. Define your rules. Write down what the logic should be, how the feature should behave, what the edge cases are. Then let it build exactly what you already designed in your head.
When I stopped asking AI to think for me and started thinking for it — everything changed.
When I started trusting agents
With the right management and planning, I started using AI agents seriously. The planning phase became the longest and heaviest part of my work. But after that? Progressive review. That's it.
And here's the part that's hard to believe until you experience it: if you give AI a well-planned, well-documented task, it returns the same result more than twice. Consistency. That's huge. It stops giving you a different answer every time. It starts delivering predictable, reliable output.
That consistency is what separates "playing with AI" from actually working with it.
Testing went from afterthought to built-in
I'll be honest — I didn't write many unit tests before. It felt like overhead on top of already long development cycles. "I'll add tests later" was my standard lie.
Now? I just plan them. I define what needs to be tested, the expected behaviors, the edge cases. The AI writes them. My test coverage went from "maybe later" to standard practice, because the cost of writing tests dropped to almost zero.
The irony: AI made me a more disciplined developer than I ever was on my own.
Learning UI/UX to complete the stack
To have a truly end-to-end workflow, I decided to go deeper into UI/UX. And that's where I realized the same principles apply — we can automate even more when it comes to design.
That's what I built with the Figma-to-Code pipeline: a system that connects Figma to Claude Code through MCP, letting me design, generate code, and document from a single terminal. Same philosophy — plan, define, execute.
What actually changed
| Before | Now |
|---|---|
| Write code all day | Plan and define all day |
| Google + Stack Overflow for research | AI for instant context |
| Copilot as annoying autocomplete | Agents as execution partners |
| Skip tests, ship fast | Plan tests, ship faster |
| Frontend OR backend | Full-stack + UI/UX + design automation |
The work didn't get easier. It shifted. The hard part now is thinking clearly, documenting precisely, and defining exactly what you want. The execution? That's the fast part.
This shift is what led me to rethink what evolution actually means in this field. It's not about writing more code. It's about thinking better.
We plan. We define. We know what to do. We save time. We save work.