A year ago, the idea of an AI helping me write code felt strange.
Now, having an AI assistant open while coding feels almost normal.
But I started wondering:
Am I actually becoming a better developer with AI, or am I just getting faster at writing code I don't fully understand?
So I changed the way I used AI.
Instead of treating it like a code generator, I treated it like a coding partner.
The first mistake I made
At the beginning, I used AI like this:
"Write this feature for me."
And yes, it worked.
I got code quickly.
But there was a problem.
Sometimes I had code that worked, but I couldn't confidently explain why it worked.
That bothered me.
Getting an answer is easy.
Understanding the answer is the real skill.
Where AI actually helped me
1. Getting unstuck
The biggest benefit was not generating entire projects.
It was removing small roadblocks.
Things like:
- explaining confusing errors
- suggesting debugging approaches
- helping understand unfamiliar libraries
- pointing out possible mistakes
The annoying 30-minute problem sometimes became a 5-minute problem.
2. Learning faster
AI is surprisingly good at being a patient teacher.
Instead of searching through multiple pages, I could ask:
"Explain this concept like I'm new to it."
Then:
"Now explain it at a deeper level."
Then:
"Give me an example."
That feedback loop is genuinely useful.
3. Reviewing my own code
One of my favourite uses became asking:
"What problems do you see in this code?"
Not because AI is always right, but because it makes me look at my code differently.
Sometimes it catches obvious things I missed.
Where AI disappointed me
It is very confident when wrong
This is probably the biggest lesson.
AI can produce something that looks professional but has subtle issues.
A developer still needs to know:
- what the code is doing
- what tradeoffs exist
- whether the solution makes sense
Blindly accepting AI output is dangerous.
AI didn't remove the hard parts
Programming is not just typing code.
The difficult parts are still:
- understanding the problem
- designing a solution
- making decisions
- debugging unexpected behaviour
AI helps with implementation.
It does not replace thinking.
What changed in my workflow
I stopped asking:
"Can AI write this?"
and started asking:
"How can AI help me think better?"
That small change made a huge difference.
AI is not a replacement for learning programming.
But it is an incredibly powerful tool for developers who already want to understand what they are doing.
The best developers won't be the ones who avoid AI.
They will be the ones who know when to trust it, when to question it, and when to ignore it.
What has been your experience using AI while coding?
Top comments (5)
I appreciated how you highlighted the importance of using AI as a coding partner rather than a code generator, and the distinction between getting an answer and understanding the answer is a crucial one. The examples you provided, such as using AI to explain confusing errors or review your own code, demonstrate the potential benefits of this approach. I've had similar experiences where AI has helped me overcome small roadblocks, and I agree that it's essential to maintain a critical perspective when working with AI-generated code. How do you think we can effectively teach junior developers to strike a balance between leveraging AI tools and developing their own problem-solving skills?
I think it comes down to treating AI like a mentor rather than an answer key. For junior developers, I'd encourage trying to solve a problem first, then using AI to review their approach or explain concepts they don't fully understand. If AI gives them code, they should be able to explain why it works before using it. That habit builds real problem-solving skills instead of just faster copy-pasting. But i think most new developers are leaning too much towards "Vibe coding" which i think is bad
Thanks for sharing your thoughts, KinetiNode! I really agree with the idea of treating AI as a mentor rather than an answer key. The ability to question, understand, and validate AI-generated solutions is becoming one of the most important skills for developers.
I’ve seen the same pattern in AI-assisted development: the biggest value comes from improving reasoning, debugging, architecture decisions, and learning speed — not simply generating more code. "Vibe coding" can be useful for exploration, but without understanding the underlying concepts, it can create long-term technical debt and limit growth.
I’d love to stay connected and exchange ideas around AI engineering, developer workflows, and building reliable AI-assisted applications. If you have any interesting projects or areas where you’re exploring AI collaboration, I’d be happy to discuss and potentially work together.
I do have some projects im working on , right now im focusing on a edge impulse EON compiler alternative.. An open-source one , im still learning to make it happen tho pretty niche but overall suits what i like doing (you can do some research on what it is used for if you dont know , interesting stuff.
KinetiNode.
That sounds like a really interesting direction, KinetiNode! An open-source alternative to an Edge Impulse EON-style compiler is a challenging but valuable project, especially with the growing demand for efficient AI inference on resource-constrained devices.
I think your approach of learning while building is actually a strong way to explore this space. Projects like this require understanding multiple layers — model optimization, compilation workflows, hardware constraints, and deployment pipelines — so every iteration becomes a learning opportunity.
I’ve been exploring AI engineering, LLM systems, and AI-assisted development workflows, and I’m always interested in connecting with people building practical AI tools and open-source projects.
Would love to stay connected, exchange ideas, and discuss your progress. If there are areas where I can share experience or help with architecture, AI workflows, or engineering challenges, I’d be happy to collaborate.