This content originally appeared on DEV Community and was authored by Darkø Tasevski
When ChatGPT first started making waves, I was skeptical. I figured it was another shiny tech fad that might be fun to play with, but wouldn’t really stick in my day-to-day work. Fast forward, and now I can’t imagine programming without some form of AI in my toolkit.
I’m not saying AI has replaced my skills, far from it. But it has reshaped how I approach problems, especially the boring, repetitive, or purely mechanical parts of development. Here’s what that looks like from my side of the keyboard.
Before AI
If you’ve been coding long enough, you know the “classic” dev cycle:
- Plan: Define the problem, sketch out solutions, argue with teammates on naming conventions.
- Code: Write everything from scratch, guided by docs and hard-earned experience.
-
Debug: Stare at logs, sprinkle
console.loglike holy water, and pray. - Test: Scaffold unit tests by hand, often repeating the same setup boilerplate for every new module.
- Optimize: Refactor with a mix of pride and dread (dread optional – unless you don’t have tests).
- Document: Write docs, knowing half the team won’t read them until they’re in trouble.
This worked, but it was slow. The bottlenecks were everywhere: manual lookups, repetitive boilerplate, and long debugging cycles.
After AI
Now, the flow feels different.
- Coding feels lighter because AI can autocomplete, scaffold, or suggest idiomatic patterns on the fly.
- Debugging is faster because I can paste an error trace and get a plausible diagnosis without trawling Stack Overflow for an hour.
- Testing is quicker. Last week, Cursor wrote around 90% of my test stubs in seconds. I still shaped the edge cases and assertions, but the grunt work was already done.
- Documentation isn’t a chore, because tools can generate a first draft as I go, which I then tweak.
- Learning curves are shorter when I can have an AI walk me through a new framework like a patient senior dev.
I mostly use ChatGPT, Cursor, and occasionally Claude Code. The tool matters less than the workflow you wrap around it.
How I Actually Use AI in My Work
Coding
I treat AI like a pair programmer who’s great at boilerplate but still needs oversight. It’s amazing for repetitive patterns such as CRUD endpoints, config files, and unit test scaffolds. I keep the final say, but letting AI start the draft saves me mental energy for the tricky parts.
Debugging
Instead of reading a wall of stack trace alone, I throw it into AI and say, “Explain what’s happening and why this might be failing.” It’s not always perfect (especially for legacy projects with a lot of spaghetti code), but it often points me toward the right layer of the problem faster than blind searching.
Documentation
I’ve stopped writing all the docs from scratch. I’ll have AI generate inline comments, API usage examples, or even a draft README from the code itself. Then I refine. The end result is better because I’m starting from something structured instead of a blank page.
Security is Still on Me
If there’s one big “don’t,” it’s handing over sensitive code or credentials. I keep anything proprietary out of prompts unless I’m working with a trusted, privacy-conscious tool. When I do need to share something delicate, I anonymize it or reduce it to the smallest reproducible snippet.
Some rules I follow:
- Stick to reputable AI providers and disable code sharing for proprietary projects.
- Strip or obfuscate sensitive data before pasting code in GPT.
- Keep the AI software updated to avoid vulnerabilities
Advice If You’re Just Starting
- Start small: use AI for a specific part of your workflow, not everything at once.
- Stay in the loop: always read and understand generated code before shipping. If something seems off, then make sure you know what and why LLM is suggesting to do something.
- Ask, don’t just paste: the more context you give AI, the better it performs.
- Keep learning: use AI to understand solutions, not just implement them.
- Treat it like a teammate: collaborate, don’t outsource thinking.
- Watch out for prompt fixation. LLMs are great, but they’re not infallible. Don’t get stuck repeating prompts that used to work. Mix up your phrasing, try new structures, and stay reflective instead of falling into autopilot
The Bottom Line
AI hasn’t made me obsolete. It’s made me faster, more adaptable, and frankly, more willing to experiment. I still need to understand the problems I’m solving, but I can spend more of my time on architecture, design, and creative solutions instead of boilerplate and tedious debugging.
In a way, AI hasn’t changed what programming is for me. It’s still about problem-solving. It’s just changed the ratio of my time between “figuring stuff out” and “getting stuff done.”
This content originally appeared on DEV Community and was authored by Darkø Tasevski