Intro
It’s been almost 3.5 years since ChatGPT 3.5 first came out. And this is 2026 now, and we have seen a lot of improvements in the Generative AI space. I have been trying to arrange my thoughts around AI, mostly around coding, general writing, or as a whole. There are some articles that have also been written by many well-known people, whose links I will share later, along with their perspectives.
Disclaimer
I am by no means an AI/LLM expert. I do work on ML Engineering problems and know the basics of GenAI and how these systems work, but anyway, here are some of the tools I have used so far.
- Early days of Copilot - autocompletions etc.
- ChatGPT, Claude, Gemini, Grok, GLM 5.2
- Claude Code, Codex, OpenCode
- Perplexity
- Kagi Quick Answer
So, these are all my subjective takes.
About My Use Cases
My day-to-day job involves coding, working on distributed systems, cloud, infra, ML pipelines, inference, designing APIs, writing design documents, writing docs, handling clients, and all.
Some of the experiences are from work and some from side hobby projects.
Normal questions and answers through a chatbot, assisting with my search, summarizing papers and some content, general questions, and fixing grammatical errors while writing, not as a whole, but just the fixes. I have not used any on-device GenAI models on my phone yet because I have an iPhone, but Apple Intelligence hasn’t been released as of now. I have disabled Apple Intelligence on my Mac as well, since I don’t get any benefit.
So mostly coding, general tasks, etc. That’s it.
Coding - Before AI Era
I do love programming.
I love solving problems, the thinking, the searching, the whole process of learning things. Figuring out how to complete a feature, reading through other people’s code, searching and reading through Stack Overflow. Somehow finishing and building the feature.
Debugging deeply, trying to understand an unfamiliar system, learning the software internals of it, figuring out where things were going wrong, deciding on architectural changes, learning new tools and using them. Random walking, and getting that thought like “Oh, this is how I fix the bug.” Spending late nights working on it.
You know the adrenaline rush of it. As a hobbyist developer, what’s not to like when driven by curiosity. I have been working in the tech industry for the last 6–7 years, and coding since my college days, that would be 10 years. I kind of built my career around this with a deep understanding of what I was doing. At the end of the day, even though we are passionate about coding and systems in general, it’s about solving the problem and the business needs.
It was fun, spending time learning a lot, learning a new language, practicing katas, understanding a system, fixing bugs, building features, learning internals, reading through whole docs.
Although not all of those things are gone now, to be fair, most of them are.
And then came AI.
My Use Case of GenAI
When ChatGPT first released, I think it was 3.5, I heard a lot about it, but I did not use it at first. I kind of started using it in April 2024. Like copy-pasting stuff. At that time I think ChatGPT 4 was out.
I kind of resisted it at first. I did not want a non-deterministic tool trying to offer “suggestions” (tab completion). I knew what I wanted to write and did not like the idea of AI suggesting it. I was not confident in it.
Then I finally decided to give in at the end of 2024. Maybe I was late to the party.
Currently, tools like (e.g. ChatGPT, Gemini, Claude on the web, etc.) I kind of use daily as part of my workflow, maybe asking general questions etc.
I think everyone started their first AI experience with a chat interface. And most people’s first experience writing code with AI was through tab completions or through a chat interface.
But then agents came and a lot changed. I kind of use Claude Code these days for my tasks. I still review the output generated by them personally.
Mine is a very basic workflow. I write a clear description of the task I want to complete, break it down, plan it, and do it. Iterate, etc. I don’t have some crazy workflow with multiple agents, just a very simple setup.
One of the other things I use AI for is assistance with my search, though I don’t use it for every search.
Also, I don’t want AI to write things for me. My thoughts and opinions are mine alone. How can AI write what I think or want to say? I have used it for fixing my grammatical mistakes though.
Things I Reason About in Coding
I have no problem with using AI tools to complete tasks. I use them myself. However, I also make sure I understand the problem I’m solving, the system I’m working on, and I spend time reviewing my own code before pushing it.
While working on codebases, these are some of the things I personally try to keep in mind before pushing code:
- I should be able to explain the changes in my own words. If I can’t, I probably don’t understand them well enough.
- I ask whether the size of the Git diff is justified by the problem being solved.
- I avoid introducing abstractions before there’s a proven need for them.
- Even if the code works locally, I ask whether it makes the system harder to understand or maintain.
- If the code is AI-generated, I make sure I am trusting my understanding more than the AI’s output.
It’s becoming increasingly common for engineers to accept AI-generated code too quickly, which is why I believe human review is still essential, even when AI reviews are available. Code that passes CI and works correctly isn’t always the right solution. Engineering has always been about building solutions that are maintainable, scalable, and easy to extend.
It’s interesting to see how far we have come with agents and all, but I still believe they need a good engineer guiding them to get good solutions.
Wrap Up
I use AI daily. I also worry about what we’re collectively giving up. These are just my thoughts as of 2026, still figuring it out like everyone else.
We’ve made it incredibly easy to produce without understanding. I don’t know what that does to us over 20 to 30 years. I genuinely don’t.
At the end of the day, I want to understand what I’m building. Not because it’s pragmatic, though it is, but because that’s what made me fall in love with this in the first place. What I do know is what I want for myself to stay curious, stay sharp, and actually know what I’m doing. That feels like enough for now.