Today was the day I've finally deleted Windows from my PC for good. I've switched to Kubuntu ~summer last year and I think I've booted windows exactly once (cringed from lags and promptly booted back to linux lol).
It's been surprisingly solid for both dev work (which is not that surprising), but also for gaming (that DID surprise me quite a bit tbh).
This is an absolutely fascinating writeup.
I had to build diff-edit for LLM myself and it was an absolute pain to make it work reliably.
This is very neat. Me likey a lot!
https://blog.can.ac/2026/02/12/the-harness-problem/
I Improved 15 LLMs at Coding in One Afternoon. Only the Harness Changed.
Cross-posted from X / @_can1357 In fact only the edit tool changed. That’s it.
¶ 0x0: The Wrong Question The conversation right now is almost entirely …
Just discovered that one of react-router recent updates absolutely broke google indexing. *Sigh* Should've migrated from it when they announced yet another round of breaking changes 🙃
I've noticed that I've been writing less and less code (by hand) over the past ~6 months. It went from me writing most of it while occasionally letting agent create something extremely specific, to me just telling agent what to write, reading output and asking for tweaks/refactors.
I've noticed that not thinking about syntax allows me to think a lot more about everything else - architecture, code structure, tests, etc. Which is kinda great 😅
Then there's also a bonus of having multiple agents.
I built a "Predatory Game Detector" for Steam. Turns out, AI thinks "This game is a gem" means it has mtx 🤦♂️
Here is the "JSON trick" I used to force a lazy LLM to catch the actual scams across 150000 reviews. 👇
https://www.patreon.com/posts/i-built-game-and-148455280
#gamedev #indiedev
I upgraded Vaporlens to Mistral embeddings and accidentally broke game discovery.
Turns out, the "smarter" context window approach actually made recommendations worse.
Here is how I fixed it with simple math: https://www.patreon.com/posts/148373760
#gamedev #indiedev #gametech
This is now my favorite take on AI tools.
Source:
https://daringfireball.net/2025/12/slop_is_slopVapolens blew up and now I'm getting death threat emails re: LLM usage. Ah, internet 😅
Craig S. Kaplan
Nicholas C. Zakas
