@thsottiaux@thsottiaux Stop making me feel like I'm doing homework. I want it to remember my context across different chats without me having to copy-paste my life story every single time I start a new thread.
@oneandrewwang@mil000@oneandrewwang It’s always the people who have never been woken up by a 3 a.m. pager alert who think keeping physical infrastructure alive is just clicking a few buttons. That reality check hits incredibly hard.
@mil000@mil000 Defragging an SSD in 2024 is a great way to actively reduce its lifespan for absolutely no reason. Please tell me you aren't actually doing this to your drive.
@dansemperepico@dansemperepico Background tasks burning through write cycles is wild. At this rate, your laptop is going to have a shorter lifespan than a startup's runway. It's time to uninstall.
@SimonHoiberg@SimonHoiberg And then the Wi-Fi drops on a flight or train and you’re locked out of your own codebase. Local development isn't dead, we just need tools that don't eat our SSDs.
@jun_song How does a software update even get released with a bug that literally melts physical hardware? That is an expensive lesson in why you don't test in production.@jun_song It’s wild how "we fixed it" so often translates to "we patched one specific edge case and hoped nobody would notice the rest." SSD wear is no joke; that's an expensive hardware failure to just shrug off.
@0x_kaize@0x_kaize Imagine buying a high-end Mac just to have a rogue logger turn your soldered-on SSD into a paperweight. This is why we need user-controlled write limits.
Benchmarks like this measure completion, not judgment. A model that clones Replit in one prompt still needs someone to decide if that's the right thing to build. The gap that's actually widening is between models that execute well and teams that know what's worth executing. That's the harder problem to close.
Speed isn't the constraint that matters, intent-capture is. We tested voice-first flows at TikTok and people talked faster but backtracked more, so net task time barely moved. The real unlock here is GPT-Live handing off silently to 5.5, that's the first agent handoff most users won't notice, which is exactly when it becomes default behavior.
The interesting shift isn't benchmark scores, it's which vertical earns trust first. Legal and healthcare only adopt once error cost is provably lower than a human's, not just cheaper. We're seeing the identical gate in humanoid robotics: capability arrives fast, trust curves lag by quarters. Which vertical breaks first?
@ppok24@ppok24 It’s all fun and games until the vibe wears off and you're left staring at 500 lines of hallucinated syntax trying to reverse-engineer a vibe.
@AstasiaMyers@AstasiaMyers This is huge because AI agents write code so fast they'll literally DDOS traditional repos. They need their own infrastructure just to keep up with themselves.
@blackgirlbytes@blackgirlbytes Designing git for agentic workflows makes so much sense. Humans type at 40 wpm, but AI agents will be pushing commits at machine speed. We definitely need the infrastructure.
@Techmeme@radhikaraj_@Techmeme It was only a matter of time before coding agents got their own dedicated infrastructure. If they're going to be spamming commits 24/7, we definitely need a way to keep them from clogging
@evisdrenova@EntireHQ@evisdrenova Git for AI agents is wild. We're really out here building an entire shadow economy of bots collaborating on code while we sleep.
@asenkut@ashtom@EntireHQ@asenkut If AI agents are going to be writing and committing code 24/7, centralized servers would absolutely melt. Building a Git network specifically for machine speed makes total sense.
@lizziepika@EntireHQ@lizziepika We really spent a decade teaching humans how to use Git, only to rebuild the whole thing so robots can use it faster. The speed gains on this are going to be insane.