I just noticed it shipped in the beta release channel 🙈

We do 🫶

I only paid one month at $20 to give it a try, but yeah I also had been using the Copilot OSS for a while. Its agentic mode was mediocre at best, even with Opus.

I'm coding because it's fun, and prompting makes it not fun for me. Tab-completion is useful to go faster, because you remain in control.

I might try out other coding agents in the near future, but yeah, it's been a wild, weird 3 weeks.

Going from a high distrust of AI agents, to planning how to build my entire software production around one tool, to ditching it completely after realising how much control I'd be waiving.

OSS 4 ever.

So there I am, I'm not renewing my sub.

I'm going to go back to coding by hand for a bit, see if I can still find that joy of thinking about a problem, devising a solution, implementing it and iterating on making it good, tested and looking nice without getting distracted. A sort of brain workout 🧠

I gave myself one month of the Pro tier to evaluate it.

It's a good tool overall. The extensibility with skills, hooks, commands, agents, plugins could even become a standard for malleable software.

But the fact it's closed source, and Anthropic's aggressive move against OSS integrations is 😬

But the worst part is that I almost didn't want to write code myself anymore. It felt good to see things being generated that quickly, that were _mostly_ alright.

"We can fix it later in post"
"It's a temporary fix"

Famous last words.

I tried playing the vibe-coding game. Telling it to fix issues and adding features using natural language.

When I finally caved in and checked the code (can't help it), it was a jumbled mess. useEffect everywhere, memory leaks, crashed the browser in the process.

I tried a combination of plan mode, spec interview, refinement, and it one-shotted a 80% correct implementation. Audio graph + UI.

That thing would have taken me weeks if not months to build in C++ 10 years ago (the Web{Audio} API takes the credit here).

But something changed.

I signed up for the $20/month Pro plan on Christmas Eve, to try it for the @nuqs.dev e2e migration from Cypress to Playwright.

I did some of the initial work myself, and let it grind for the rest. It did a good job.

But the real 🤯😱 moment was building the EQ & Compressor audio playground.

Custom made, controls are shadcn button groups + custom knob (I’ll likely OSS that one), graphs are canvas rendered.

Processing is the WebAudio API. Crazy what we can do with off-the-shelf nodes. Next step is to implement a custom algorithm (gate and de-essser).

My bad, we started at 5.5k but ended up at 9.8k, so +4.3k

I feel like @nuqs.dev should be in the state management libraries or React sections (+5.5k stars in 2025).

cc @jherr.dev @rwieruch.bsky.social

AAA WCAG contrast