My honest take on AI developer tooling

Sonnet

Since the Sonnet release and agentic workflows, I keep finding myself wanting to try some of the others known for reasoning like Google Gemini 2.5 pro, or GPT-4.1 but Claude is the only one that I find with the best and consistent results.

The majority of things I’m building are React web applications or APIs written in TypeScript, Kotlin or Python and I have provided all sorts of tasks like writing new features, refactoring, or planning solution options.

Terminal FTW

Almost every hands-on engineer these days uses a terminal which means that Claude Code had an enormous amount of target users to begin with. Pretty nifty strategy if you ask me. Chat was the initial interface to LLMs and terminal is text based. Right? So it seems like a very logical choice.

After spending about $20 USD on Claude Code building my portfolio website, I can comfortably say that I gained so much confidence that I practically stopped reviewing the code. Yep, I did some vibe coding and picked a web framework (Astro) that I never developed with before. Simply, fired up the browser with hot reloading, talked to the LLM and just watched the magic 🪄.

Cursor and Windsurf on the other hand decided to enter the war of editors where engineers can be quite sensitive. Yes, I gave Cursor a try around the time it made a lot of noise and before VS Code had Copilot without any steroids. And although it seemed a bit like magic I couldn’t help think that Microsoft and GitHub were going to catch up pretty damn quickly. In a few months GitHub announced a bunch of new features to the Copilot extension and it was basically at feature parity.

What about the cost

Claude Code takes the cake here. Just ask it a simple question and count yourself lucky if it doesn’t charge you 20 cents. There must be something to do with how much it loads into context and probably the reason why it feels better than the rest.

This is where all these AI tools and products make me a little bit sad. There was a time where anyone could just learn from a simple notepad and books. There still are some freemium options but in order to really skill up in this area you’ll have to start opening up your pockets. Just be selective and don’t get addicted because I can see how that can happen.

The bigger picture

What’s interesting is how much AI companies are investing in engineer tooling specifically. On one hand, there’s something exciting about this - they’re clearly betting on us developers sticking around and want to build tools that make us more productive. It feels like validation that we’re not going away anytime soon.

But on the other hand, I can’t shake the feeling that by gathering all this data on how we work, these companies are learning exactly what we do and how we do it. Could they eventually use this to duplicate our workflows and make us redundant? I mean, it’s possible, right? I can see why this would make a lot of people nervous.

I’m not sure how I feel about it yet. The tools are undeniably helpful, but there’s this underlying question of whether we’re essentially training our own replacements. It’s a weird position to be in.