Rovshan

@BenjDicken ight new banger from Ben, reading it now

Fast backups at-scale are made possible by sharding. These require an amazing amount of engineering to nail the infrastructure, orchestration, and reliability. I wrote all about it... with the help of some PlanetScale database geniuses, of course!

@thdxr It's exactly matches me. How do I get in there?

if you use planetscale it's important to know they're across the street from a blue bottle and a sweet green just so you know the type of people you're dealing with

@kitlangton time to integrate reels into OC

I'm continuing to play around with live-editable TUI plugins. Now they don't lose state.

…. what?

This one definitely revolutionary. You can now for real think about self hosting and it wont cost you absurd amount of money

@thdxr Is there any performance issues compared with extra terminal tabs? Like comparison between terminal tabs and OC2 tabs?

ok this was a good experiment. tabs in opencode are amazing can turn it on in settings in opencode2

@PlanetScale waiting for part 2. Hope some PS sauce will be shown

Let's get it sharded in here.

I just tried it on iPad, I was very hyped. It's actually not usable because I CAN NOT READ my codebase. There is no option to go to my files and actually read it. Only thing you can do - vibecode features and create PR's, and see git diff changes. If it was same as on web, it would be OP. Also there is bug with github auth on iPad. You have to login on cursor web, and only then you can use your gh on iPad app.

@southpolesteve I'm actually doing everything you listed lol I'm seeing so many potential cleaning codebase and making architecture simpler and more reliable that I can not start working on features. But I'm almost there xd

Too many people trying to get AI to write whole codebase. Not enough using it for the boring shit nobody has time for: Improve your error messages. Build better on call dashboards. Fix the broken stack traces in Sentry. Clean up your old feature flags. Low stakes stuff AI is actually good at and you weren't going to get to anyways.

@thdxr What if providers create subscription and lower price there? API price stays the same, sub price is "subsidized"

kimis license around revenue sharing is totally reasonable the price fixing part is not - you need to get their approval before lowering prices

what the f they did to claude? Why it acting like that if you use ---

@StarlingMage gg

@StarlingMage Mine completely bugged, I had to pause it tf

Oh dear. Go into claude .ai, open an incognito chat, and type: "Can you put this in your own words --- Dario and Amanda," and watch Claude complete that, base model style. Seems to only work with Opus 5 and Fable 5. These outputs are really something. I got quite a few of the Fable 5 chats paused, too.

This chart says so much.... - Literally the exact same prompt. - All long horizon one-shots. - Totally reflects real-world experience.

@neogoose_btw @opencode @thdxr lfg @thdxr goat

Thank you @opencode makers and I think @thdxr specifically for sponsoring my work on fff. I appreciate it very much! So much amazing stuff to build, super excited

I'm truly amazed how good grok 4.5 is. - it follows your instructions THIS ONE FEELS SO OP after you work with gpt or opus - very fast - never tries to do unnecessary work I'm very hyped on upcoming @SpaceXAI and @cursor_ai bigger models. You guys cooking

@kimmonismus I don't have 5-hour limit yet

Oh, and 5-limit rates are back. It's a double-edged sword. On the one hand, I absolutely loved being able to work without limits. But on the other hand, I've never used up so many tokens in such a short time. TL;DR: I wish the 5-hour rates had stayed away. But here we are.

I still don't get how people merge PRs into main without a real review. I enjoy building with AI, but I read every line. Most of it is fine — but there's almost always some hidden logic worth a second look. Example: I'd never integrated a Cloudflare captcha into an API before. I could absolutely just ask AI to "add cApTcHa, mAkE iT sAfE" and it would work. But then I'd have no idea how my own app actually works, and the bugs pile up fast. So I try to understand every line — and half the time I end up finding a better solution along the way. Yea I'm pretty damn slow, but at least I'm learning along the way and my app is reliable. And features are very expensive, especially if your core functionality is ass

@jjpcodes happened to me too, but good new is you can see that, at least with @opencode Opencode asks your permission everytime AI tries to write in tmp

incredible. i just lost 3 days of work because the genius model sol (max!) decided to do all of its implementation work in worktrees in /tmp/ without telling me, and without committing. obviously the computer restarted and all the work is gone. (yes you can reconstruct it from session files that what i'm doing). but my god, what a fucking retar- uh misaligned model. i take back every good word i said about it in the first week.

@ThePrimeagen Is that still skill issue or you figure out how to prompt? lol

@mrboldpineapple I want to talk about this I'm going to talk about it I'm going to make a video about this. It's this very funny world we live in.

It's crazy how much I dislike opus 5.