muzz khan

@samgoodwin89 @zlxndr oo gotcha

@muzzdotdev @zlxndr APIs are inconsistent and wonky. You have to write a ton of code to identify and handle edge cases. Some times it’s some awful regex on a message. It’s never as clean as it should be. Companies are sloppy. That ultimately becomes a spec that can be fed into a code generator.

We did this in alchemy v1 ("just use fetch"). It was ok but ultimately a dead end for us. Lead to slop, esp. for retries. In alchemy v2, we built distilled Effect SDKs for all clouds. It's still generated by AI based on usage, but we don't throw away the leverage of codegen.

@zlxndr @samgoodwin89 Yea the retries bit is confusing here

@samgoodwin89 Why did it lead to slop? The main idea behind the post (the way i get it) - is you build a single api client which handles everything consistently, including retries, for any endpoint? It’s http responses + fetch everywhere, no?

x.com/samgoodwin89/s… Seeing this right after that article is radicalizing me

@LukeParkerDev @theo @maria_rcks Windows AND microsoft edge??

@theo @maria_rcks gang I am following Microsoft Edge's lead the BEST browser ever. It doesn't allow you any more vertical space, so therefore I will not let you either

since you guys really want Vertical Tabs. I built an OpenCode Desktop Extensions SDK I built a quick vertical tabs, and subway surfers extension. Live enable, disable, install etc. Checkout the repo, the README is pretty self-explanatory lol. Use at your own risk. Build something fun, I'm curious to see what ideas you all will have? github.com/Hona/opencode-…

@KentonVarda Annoying but i get it

Me: Please commit your changes with a commit message matching the style of other recent commit messages. GPT-5.6 Sol: `git log --oneline -10` GPT-5.6 Sol: *writes a one-line commit message* Me:🤦

Why is this still a debate Let humans review the PRD if required Use a good model, address greptile Good tests, alerts and ideally a multi-staged environment You’ll skip the part where the person reviewing the PR asks the clanker for essentially what the PRD already stated

@maria_rcks Cat lady ?

twitter money well spent

@tszzl Is that eva?

damn this was a risky ad

@skeptrune @RhysSullivan New macos has vehicle motion cues that help a little I used to puke like every 1/20 ride and this has largely stopped

@RhysSullivan sometimes from phone, but I'm not pulling the laptop out i get carsick

waymos are very slow to get places i'm directionally bearish on founders or otherwise supposedly busy people that swear by them feels unserious

@jayair tool? you mean markdown? what's a better way to express information than with code which makes interfaces

Artifacts are interesting because the broader thing it's doing is it's helping people better understand the output of the LLM It used to be that giving the LLM access to relevant data and tools, aka the inputs, used to be the issue, but that's largely solved now However, reviewing their work, aka the outputs, still feels laborious, especially if it's just a wall of text or a bunch of diffs Instead, you'd want it to generate "media" that makes it easier to consume larger amounts of information Whether that's a "explain it to me like I'm 5" type of prompt, or having it generate HTML artifacts That said, it feels like we could do better, maybe giving it some tools to generate better artifacts

@skeptrune damn, i've never had a human review my code might be for the best lol

i personally think code review is dead. the team does not agree. directionally found this surprising.

@iamsahaj_xyz might have to try pi for this

Introducing /trio for pi a multi-model, single-session workflow so you can use the best model for the task

@RhysSullivan @thsottiaux there isn't even enough effect code out there for a human to learn how to write it well

@thsottiaux Post train on EffectTS please it will make models so much better at outputting good code

At different times you complained about speed, code slop, frontend quality, ... With each release we improve and GPT 5.6 Sol is ✅Fast and token efficient ✅Hardcore at back-end dev ✅Great at front-end ✅Does not use useEffect everywhere What is next?

@TheBasedCabal @greptile hey there -- sorry for the mess here. could you reach out to my dm's with your account email? we'll pull your invoices, explain every charge, and refund anything that wasn't clearly authorized.

Where is my money @Greptile? I signed up for a $30 a month plan not an open tap on my credit card. I can't believe you make users set spending limits manually instead of setting a hard limit right away. I will never use Greptile and never recommend it to anyone. WHERE IS MY MONEY GREPTILE??????

@davis7 Exactly I freaking love big model smell x.com/muzzdotdev/sta…

On Fable being smarter than 5.6 - it is, but: I think smart is the wrong word, the thing 5.6 is lacking behind fable is the "big model smell" which roughly translates to "understanding the unsaid intent behind your prompt" fable fills in the blanks better, it writes more beautiful code, it suggests the next thing the way you would or better 5.6 is a blunt instrument. It does what you say, and it will damn well make sure it does it no matter what. 1000+ tests, 20 nested try catches, catching every single edge case possible through rigorous end to end testing of the program itself. It gets the job done, just not in the cleanest way unless it's well directed the biggest thing it shares is that ability to run for absurd amounts of time with 0 issue (had threads run for 20+ hours spamming compaction and dozens of subagents and it ran flawlessly, could even wildly steer it halfway through and it just kinda worked) and orchestrate itself beautifully take everything good about 5.5, and make it better while fixing it's most glaring issues

@OpenAI @thsottiaux this iisss reallly reallllllly annnooyyinnnnggg man.

@OpenAI @thsottiaux i tab out and its just gone after a bit Codex 26.623.101652

@OpenAI @thsottiaux why the heck does codex app just quit itself.

@brianchew @gitlab Works on the next try lol. Atleast i had the same type of problem and the fix was that

brooo @gitlab guys.. i cant even submit the verification code to log in.. wtf man

@brianchew you can spam click the link and they will have to pay for it you didn't hear this from me though

tf lol i cant even see greptile on first page... need to scroll

@HaoStudioCC cursor has a cli. will look a take at termius and blink, thanks

@muzzdotdev no unified app exists yet. closest thing is ssh from a terminal app plus tailscale on the home box. works for CC and codex both since both are terminal based. cursor is harder since its GUI only. termius or blink shell on ios both have decent ssh experience

Is there a mobile app that can run claude code, codex, cursor, opencode on my machine and serve it through one mobile friendly interface? If not then idc