Matthew Fox

@carmguti Crazy run

I’ve left A24 Labs. when we started, I was the company's only engineer, given a laptop and an empty git repo. now, it’s a talented team of artists, designers, and engineers building incredible things for the film industry. excited for what's next. stay tuned!

@BacLeodiv Retain context

Name one thing you still do better than AI.

@zendadddy Try paper.design Best of both worlds without the upfront debt

I hate to admit it, but sometimes I spent hours wrestling with Cursor on prototypes I could have built in Figma in half the time. The first 5 mins are epic - I'm moving so quickly. But as the proto gets more complex, something breaks, then I spend the afternoon trying to undo and fix it. Anyone else feeling this? Some kinda weird AI sunk cost or perceived productivity thing happening. There's gotta be a word for it.

Opus 4.7 is crushing it today ngl Definitely feels like it was de-lobotomized Might be the psychosis talking but it feels different today Feels good man

@subproject_22 🐐

Seeding Design Systems with the most popular shadcn/ui and tweakcn themes. Or create your own. Any variables, any naming. Save, reuse, update, preview on your own components. No more shadcn/ui lock-in. Shipping in a few hours.

@linuz90 @OpenAIDevs The whole point is context forking - otherwise make a new session

Hey @OpenAIDevs, why do Codex side chats start with pre-filled context? In 99% of cases, I'd like to start fresh in side chats instead of forking the main conversation, which seems to be what's happening here.

@theo uploadthing wrapped in a skil uploadthing.com

Just learned it's literally impossible to paste images into Claude Code over SSH. How do you CLI people live like this??

@jessepollak @base @googledocs 30 seconds to make 4 word blue Incredible tech

over the last few months, I built bdocs internally at @base to do exactly this: publish, comment, and iterate on HTML docs with your agent! hey @googledocs team — can you add the ability to work in HTML so I can stop building this internal tool?

@nexxeln Big agree

i’m starting to think agent-friendly codebases are more about constraints migrating parts of opencode to effect has made agent-written code noticeably less cursed good architecture boxes agents into writing specific, constrained code fewer ways to go wrong

@dotta 🎯

For Paperclip, we're moving away from "zero human companies" and moving towards "the app people use to manage agents at work" 📎📎📎

Creating prototypes during planning is the new "make no mistakes" Except it actually works

@jconnorholliman @coinbase Did you work at coinbase? Had no idea

you can either cry because it’s over or smile because it happened yes, i realize it’s a cliché, but it’s a great way to approach life especially in the hard times and this past week feels like it has thrown everything hard at me from being impacted by the @coinbase layoffs, to saying goodbye to my family labradoodle of 16.5 years, to dealing with a stomach bug last night it was truly hard to wake up this morning and feel like everything is going to be okay with all of this happening at once but in times like these, it’s important to smile and remember the amazing times life has brought before i wouldn’t trade anything for the in-person time i spent with the team i hired at @base, or the 16.5+ years of hikes and playtime with my dog, or the health i’ve been granted to complete feats like half-marathons and a marathon obviously i’m sad, but smiling and choosing the optimistic outlook makes the hard times so much better and man am i thankful for all of the support i have received i’ll be back stronger

@jpschroeder lol

🚨 Announcing: stylesheets. A language for describing *how* HTML is displayed, giving humans unprecedented ways to view the HTML output of LLMs. It doesn’t even need to use your context window because it can be in a different file, you just say what elements the styles apply to like this: h1 { font-weight: bold; } The future is so bright. Article coming soon of course.

@rfleury Plenty of human developers can't do step 2 there to save their life, or in this case jobs

“Vibecoding”, i.e. ~hands-off usage of LLMs to rapidly generate code without regard for the actual code’s contents, for novel applications, can literally never be non-slop, because—as I’ve described before—there is not enough bits of information content in prompts to express the user’s exact desires in sufficient detail, and the desired solution is not expressed in training data (due to the problem’s novelty). Only a sentient human developer can relate to another human user to determine what is desirable, and design the software such that it accomplishes this desirable outcome, and carefully verify that it is doing that, rather than something else (potentially undesirable). This is true even for the combinatoric space implied by the training data, for instance if the novel problem is merely novel in that it combines pieces of existing solutions. There needs to be a guiding force to know what to combine and how. The more detailed the prompt becomes, the more human oversight (the more human-guided round trips with the LLM), the closer it becomes to actual code (i.e. detailed execution instructions for a computer).

@levelsio @grok Tbf removing pain from aws services is a respectable biz model Someone's gotta do it

Resend is literally just SES with a 500% markup @grok fact check

Hey if it works it works There's a whole lot of potential in exploring them A big part of what was holding me back from using a lot of them was the UX of managing all of the above/mitigating context bloat etc But after forking @papercliping pretty early and very much making it my own I now use them all extensively, with a bunch of agents with different configs, using different combinations of tools/models for different purposes A lot of the things I added to achieve this are now in the stock version - so would highly recommend cloning it and having a play I use it extensively for work now but I also use it as a playground to evaluate a lot of the tools/my skills etc And a bunch of them have graduated into my daily usage outside of orchestrating via the platform + It was a really fun way to go deeper on all of them

Feel like I'm missing out because I don't use skills, or a lot of MCP, or multi-agent orchestrations when using AI dev tools. I'm just like "implement this feature" or "how do this work" or "no not like that, do this instead". Idk, I feel fast and accurate so why change?

@LifeLineLogics @ThePrimeagen Warp good

@ThePrimeagen Doesn't matter which platform I'm on. I'm still gonna use Warp. And yes Terminal on MacOs is better than CMD

Dear interns, I am sorry for what I said about Terminal App on Mac. It is better than cmd on Windows I am sorry for lashing out I'll be taking the rest of this weekend off Twitter to understand why I behaved this way. Say we are a family or you are fired, ThePrimeagen P.S.: we are doing layoffs because of AI

@brycent I think it's pretty funny tbh Good play on buy now , pay later Which is arguably a much worse predatory primitive that will bite a lot of people

The startup that turns your bank card into a casino 🎰

@thsottiaux Windows for my sins

As a Codex user, which platform are you on

@benhylak They exist but they all suck

LITERALLY ALL I WANT IS A CHATBOT FOR TRAVEL "FIND ME THE BEST HOTEL AND BOOK IT"