I’ve been building more client projects lately, and I’ll be honest, ChatGPT is now part of my workflow whether I like it or not.
Not as a replacement.
Not as some magic tool.
Just there. Constantly.
Helping me break things down, debug faster, structure projects, move quicker.
<hr>
But I didn’t grow up coding with this.
A lot of you did this for years without AI:
<ul>
<li>figuring things out the hard way</li>
<li>reading docs line by line</li>
<li>debugging with nothing but logs and patience</li>
</ul>
And I respect that. Because that foundation is different.
<hr>
What I’m trying to figure out is this:
From your perspective: has AI actually made you better, or just faster?
Because from where I’m sitting, it feels like both, but also neither sometimes.
<hr>
On one hand, I’m moving faster than ever:
<ul>
<li>less time stuck</li>
<li>less time searching</li>
<li>faster iterations</li>
</ul>
On the other hand, I have to stay sharp. If I start trusting outputs without thinking, I can feel myself slipping. It’s subtle, but it’s real.
<hr>
I’ve caught myself a few times:
Shipping something that _works _but I didn’t fully break it down the way I normally would. That doesn’t sit right.
<hr>
So I’m curious how you’re handling it.
If you’ve been doing this 10+ years:
<ul>
<li>Has AI changed how you approach problems?</li>
<li>Do you trust it at all?</li>
<li>Has it made you rethink how juniors should learn?</li>
</ul>
Or do you see it as noise more than anything else?
<hr>
I’m not looking for surface-level takes.
I want to hear what actually changed for you in your day-to-day coding, decision making, and how you think through problems now.
<hr>
Because right now, it feels like the rules are shifting [🤔] and I want to make sure I’m playing this the right way.

John Liter

A developer reflects on integrating AI tools like ChatGPT into their daily coding workflow and poses questions to senior developers about how AI has genuinely changed their approach to problem-solving, decision-making, and mentoring juniors. The post explores the tension between moving faster with AI assistance versus maintaining deep understanding and sharp fundamentals.

Dear Senior Devs, How Has AI Actually Changed the Way You Code?

Everything changed in my flow. I usually work on 2-3 tasks in parallel. Never write code myself anymore. I spend way more time planning stuff, talking through the plans. I spend a lot on building my own tools that can automate more work

I have been on a war against using AI inside of my editor for a long time, even though i used ChatGPT separately for some kinds of tasks.
Lately i have realized that this war was futile and against the times because honestly the tools have improved enough to be part of software development in the future.
After that i started heavily investing into gaining knowledge on how the tools work, i got Claude Code and now i am using it heavily.
I would say that more than 50% of my code is written by AI, but there are also tasks where it really does not work.
Whenever generating scripts, like eslint rules or codemods then i am using AI 100% of the time.
Whenever judgement is needed, i usually cannot trust Claude well enough and usually do it on my own. An example for this would be a few weeks ago i have replaced lots of texts in an app with semantic classes, Claude was tasked to figure out “Is this a button? Is this a chip?” and depending on the answer it should add a class like “btn-text” and so on. It failed MISERABLY at doing it, even though my prompt was very accurate and clear, it even had examples.
I general i would say that if the change is simple and me implementing it takes longer than typing it as a prompt and if i am sure that claude will not do it badly, then i use the chat and move on to the next task, thinking it through while waiting for claude to finish the previous.
Also github copilot code reviews are extremely good and i never want to work on a project again that does not have AI code reviews implemented in some way or another.
Regarding keeping vibe coding of high quality i am also using eslint a lot, i wrote lots of rules that make sure that ai agents and developers follow project standards. If you can put a rule into words, for example “Icons imported from the icon library should always be imported as the version with the Icon suffix”, then eslint is a great solution for it instead of writing it into the AGENTS.md or wherever. Because eslint is deterministic.

Honestly not much changed. AI very rarely generates good quality code and if you want to make something that will work for years you should do it yourself. Of course i can generate a lot of AI slop but that’s not the point of making good quality product
For now AI mostly are google search alternative which very often a dump piece of shit you need to make it answer something real instead of random hallucinations

I for one am a php/lamp stack dev for 20+ years, I find AI verysuitable in my workflow. Also, I am totally horrible in C# and recently needed a small CLI app to control audio channels in my 5.1 recording studio, after months trying to either find something or develop something, I was able to do it with the help of Gemini. And while at it, I learned a lot! So AI can also help us go into unexplored territories faster and gain knowledge without having to watch endless videos or tutorials.

I’m not a senior dev yet but I assure you that using AI is just the same as with any other tools. Complete reliance on it without the discipline or responsibility to check its output from time to time is a recipe for disaster like with any other tools.