Everyone has an opinion on AI right now. I'm not here to add another hot take to the pile. What I can do is share what it's actually been like using AI as a developer tool over the last couple of years, the good and the bad, and where I think this is all heading. So here's my honest take from the trenches.
The journey
I started with GitHub Copilot in neovim for code completion, paired with the ChatGPT web interface for when I needed to ask questions or work through a problem. I used that combo for about two years and it was... fine. Copilot would suggest completions as I typed, and when I needed more help I'd switch over to a browser tab and paste in a snippet of code. The friction was real though. I would lose context constantly, pasting bits and pieces into a web interface, and the AI would suggest something different every time instead of building on what it already knew about my project. The question that kept nagging at me was, "How much better would this be if it had my whole project?"
Around December 2025, I discovered that Claude had a CLI tool called Claude Code. Around the same time I learned that ChatGPT and Copilot had their own CLI offerings as well. So I tried all three side by side. Claude Code was the clear winner. I cancelled my subscriptions to the other two and haven't looked back.
I actually tried the Claude web interface earlier but gave up on it because I would quickly run out of tokens. By contrast, try as I might, I could never run ChatGPT out of tokens. But with Claude Code running in my terminal with full access to my project, the token issue became less of a concern and the quality of the output more than made up for it.
What changed
Having full project context changed everything. Instead of pasting snippets and hoping the AI understood what I was working with, it could see the whole codebase. Fewer back-and-forth cycles, less explaining, and the AI would actually remember patterns in my project and reuse them instead of inventing something new each time.
But the biggest surprise was what I started using it for outside of writing code. I've been doing a lot of housecleaning on my network, things that I know how to do, or at least know enough to research and get done, but because I know there will be research involved I put them off. With AI, I just stopped putting them off.
The moment that really blew me away was when AI SSH'd into my minipc to configure the timeserver. That would have been a couple of hours of research for me to figure out. It was done in minutes. Sound not working on a Linux installation? Fixed. Fonts messed up? Fixed. Discord notifications not coming through? Fixed. These are perennial Linux trouble spots that I had just been living with, and one by one they were getting knocked out.
I've since dusted off many side projects and progressed them quite nicely. I've cleaned up my neovim, Emacs, and tmux configurations. Well, I say "I" but really it was with the assistance of AI.
The concerns
So here's where I pump the brakes a bit.
Jobs. I'm concerned that AI will be taking my job soon, or at least redefine it in a significant way. When AI can SSH into a machine and configure services, fix long-standing system issues, and churn out working code, what does "developer" even mean anymore? I don't have an answer. I just know the question keeps getting harder to ignore.
There's another angle to this that I think people aren't talking about enough. AI is amazing in the hands of a seasoned engineer. You know what to ask for, you know when the output is wrong, and you know how to steer it. But for a junior engineer it's a very different story. AI is clumsy when you don't have the experience to evaluate what it gives you. So companies are looking at this and thinking they can squeeze every ounce of productivity out of their senior engineers with AI and skip hiring juniors altogether. That's not sustainable. Senior engineers retire. If you've cut off the pipeline of junior engineers getting the experience they need to become senior engineers, what happens in ten or fifteen years when there's nobody left who knows how to evaluate what the AI is producing?
The sustainability loop. AI trains on what humans have made available for free. Blog posts, Stack Overflow answers, open source code, documentation that people wrote because they wanted to share knowledge with the world. I worry that as humans begin to rely too much on AI, they'll stop creating that original content, and then what does AI train on next? There's a real risk of a plateau here. The thing that makes AI useful depends on humans continuing to do the work that AI is now replacing. That feels like a loop that doesn't close.
The data question. There's also a moral line where AI just takes content to train on without the consent of the people who created it. Beyond that, companies are trying to put AI everywhere to convince people to give them more data, even highly personal data, to train on. People may not fully realize the trade-off they're making when they hand over that information.
The money. AI companies don't seem to have a business model that makes money. The costs to run these models are astronomical. What happens if the funding dries up? If the venture capital runs out and the subscription fees don't cover the compute costs, do we just lose access to something we've built our workflows around? That's a real risk worth thinking about.
Where I land
Despite all of that, I'm still using AI every day. That says something. It's the most significant shift in how I write code since... honestly, maybe ever. The move from searching Google and Stack Overflow to having a context-aware assistant in my terminal has changed how I approach problems. I get more done, I procrastinate less, and I'm working on things I have been putting off for years.
I'm cautiously optimistic. Eyes open, concerns noted, but moving forward. If you've been on the fence about trying AI dev tools, or if you've tried them and have thoughts, I would love to hear about your experience.
Written by me, assisted by the AI.
Tags: #ai #claude #developer tools
Categories: #ai