My current agentic tool eval:
Cursor: Best developer experience
Codex: Best Look & Feel, terrible editing
Claude Code: Best IntelliJ integration, terrible UX
Kiro: Exists
Today I encountered a 54MB 404 page with nothing on it but a few lines of plain, black and white text. No links, no JS.
How did <1KB of text turn into a tab that big?
I haven't touched C since college. As a primarily Java dev and someone who has taught basic coding, I'd still say the best first language to learn in depth is C. I understand Java, Python, etc. better bc I understand the basics in C.
C won't help with JS, though. Nothing will.
Our Cursor agent for GitHub decided my code had problems with exception handling. Turns out the handling it wanted was directly above the flagged code.
This is why I'm not worried about AI taking my job. 🤣
Claude Code is great, but the IntelliJ plugin is basically non-functional for me. Granted it's in beta, but I really hope it gets fixed soon.
I am usually faster than coding agents.
The agents win at the edge of my knowledge. Obscure command? Opaque library? JavaScript *shudder*? Let the agent do it.
But there's a fast drop beyond that edge. I cannot make an AI code something that I fundamentally don't understand.
I'm having a good time with @claude_code, but it should come with an epilepsy warning. Every time I tell it to do something the scroll goes berserk, so it constantly flashes between top and bottom of the terminal. 😵💫
Grok Code Fast 1 is definitely fast; I can see why it's doing more than other models (especially while it's free).
But it's not better.
It codes faster, forgets instructions faster, and goes off the rails faster. Still useful, but use with caution!
I'm far more productive in @htmx_org than I ever was with React, Angular, or even @FlutterDev for web. (I do like Flutter, though.)
My prior test projects all had page load delays. I can't make HTMX that slow. And pushing data with HTMX is so much simpler than tracking state!
Here's how I know that AI isn't coming for my job:
I've been using Claude, Grok, and GPT5 to write a Twitch bot. All of them failed multiple times to set up EventSub with webhooks, even though it's only about 50 lines in Golang. (Cursor "auto" managed it with 4 hours of help.)
So I've been trying @kirodotdev for a small side project. The IDE itself is great - probably the best AI-centric dev environment I've seen.
The problem is that even @claudeai still doesn't understand software design, useful testing, or code simplicity.
Today, Gemini 2.5 pro blamed its tools for bad edits, then blamed me for a failed build in unchanged files.
It is now truly equal to a human developer.
Whenever I hear an OG dev talk about dropping a stack of punch cards I think, "Was that before boxes were invented?"
On the one hand, @cursor_ai is currently my favorite agentic coding environment. On the other hand, @intellijidea is just way easier for navigating, editing code, and viewing complete change sets.
I guess I'm a 2-editor dev for this project?😅