Zackery Ward

@Dimillian all the time so I don't have to put so much effort into prompting

Let’s be honest here, how many time this week have you directly @ a code file to your agent?

NEW: This beanie lets you type with just your thoughts. Sabi just emerged from stealth with a brain-computer interface you wear like a beanie that turns internal speech into text on a screen. • Noninvasive interface worn as a cap or beanie • 70,000–100,000 sensors for high-density brain signal capture • Trained on 100,000+ hours of neural data • ~30 words per minute at launch, improves with use • Backed by Khosla, Accel, Initialized, and Kevin Weil @sabicap feels like the first real attempt at a consumer BCI.

@iruletheworldmo You can rest easy now strawberry man

you heard it from me first. when people start saying. 4.7 feels worse than 4.6 just remember me.

@Perhapsolutely @thsottiaux you want to manually edit the context? i see the benefit in that. but you would only be holding yourself back.

@thsottiaux I know it sounds crazy but I wish I could edit the context in some way. Remove specific things I don’t want, add things, review it etc

/compact coming in Codex, we finally listened

@thsottiaux i don't know if this was actually needed, but psychologically we need this.

@adrgrondin wait, but what is your context size?

Google's Gemma 4 26B A4B running on MacBook Pro M5 Max 100% native Swift Achieving over 100tk/s with Apple MLX framework

@iruletheworldmo bro you're the Mark Gurman of AI

don’t listen to these faceless anon grifters chat. they make predictions with 0 knowledge. they say things like, it launches tomorrow unless it doesn’t. spud releases tomorrow this is 100% confirmed. spud / imagev2 / opus 47 if there’s one thing i hate. it’s fake insiders.

@sflorimm cloudflare

Name a tech company that never disappointed you

i'm pretty sure this story was completely fabricated for this is exactly what you would expect to happen. as an experiment, if i wanted to do something like this with the remote possibility of success, i would define the tech stack: use cloudflare and flutter only. i would set them up with the cloudflare mcp. I would create the worker for them. setup testflight for them and shell scripts to deploy to testflight. i would require them to have codex refactor everything at the end of the day. They would NOT use git and instead take periodic backups...AND they would have to be passionate killers for a few weeks and maybe, just maybe, they could make something worthwhile.

I tried to hire a guy to vibecode apps for me. Essentially, I tried to respect what all the AI companies advertise Basically, choose an average Joe without knowledge of programming. Got a guy after 2 days, and he started coding. Got the $20 claude for him. He did the landing page within 5 minutes. I said "Amazing, lets build the app now" "Should take us at most 48 hours" After 15 minutes, the guy hit the limit "I'm not gonna wait 3 hours for the limit to reset" I have to get this guy the max $200 plan, so i purchase it He restarted work, and everything works great. App looks bad, and doesn't really work well. I will let the guy go, and do it myself Good thing, is that we are not hitting the limit anymore. I guess its a win, so much for vibe coding.

@rUv Meanwhile Apple bro: “I bought a Mac Studio with 32x the RAM / VRAM as the poster for the same 10k" (512gb vram)

I told the guy at the shop what I was building and he just stared for a second and goes, “what are you building, Skynet?” Introducing ruVultra. My kids didn’t miss a beat: “yeah, basically.” And once you look at the numbers, it stops sounding like a joke. I built this entire system by hand, in an evening. It’s a sovereign AI node. A brain in a box. Ryzen 9 9950X with 16 cores / 32 threads, tuned with a custom Ubuntu kernel and over clocked thermal profile pushing toward ~6GHz burst behavior. With AVX-512, each core processes wide chunks of data at once, so vector comparisons, filtering, and boundary detection happen in parallel, not sequentially. The CPU becomes a real-time reasoning engine, not just a coordinator. Then the GPU takes over when needed. An RTX 5080 with ~10,000+ CUDA cores running in the ~2.5–3.0GHz range, handling dense math, embeddings, and batch workloads. It’s a split system: CPU for structure, GPU for intensity. Compared to a high-end Mac mini or even a Studio, you’re looking at 5–10x faster performance on real AI workloads. Not just because of raw power, but because of architecture. No shared memory bottleneck, no abstraction layers, full CUDA access, full control over scheduling and memory. This machine doesn’t wait on anything. You’ve got 128GB of RAM now, which keeps most working sets local, but there’s room to grow that to 1TB of RAM (estimated at $30k), turning it into a true in-memory system. Same with storage, plenty of headroom for multiple additional TB of NVMe, extending your dataset without killing performance. At roughly a $10k budget, you’re sitting in a sweet spot. Not cluster-scale, but powerful enough to behave like one node of a serious system. You can run meaningful local workloads, test ideas end to end, and iterate without waiting on the cloud.

@athcanft Why would anyone want to take shortcuts on the most important part of your App store listing? There are lots of plug-ins skills and what have you for quick screenshots but they will never ever produce results that will convert.

who’s gonna make it?

@giblfiz @nikita_builds diabolical

@nikita_builds I actually have a skill called "token bonfire" for this occasion that spins out a 3 layer tree of agents with 5x fanout. I also burned out my Max20 weekly after two days, so haven't exactly gotten to use token bonfire lately.

when you're 20 minutes from reset and you have 50% usage left

@snmzeno @seraleev open up your keyboard, open up the whatever happens when you click add new task, so they can just add one right away.

@seraleev so for this specific example, how would you personally design the flow?

Feels like I’m back in school teaching, but let’s get to the point. An empty state is when nothing exists yet – no tasks, no projects. And screens like this cost you money. Yes, the screen can look beautiful and animated. But the user still has to tap that blue button. And there are three more buttons at the bottom… you’re going to lose a percentage of users right there. Simple rule: guide the user by the hand during the first session. That’s how you avoid losing conversion.

@adamrspain @Tobby_scraper you nailed it bro!

@Tobby_scraper Honest thought?

Your app icon is doing more work than you think. It's the first thing users see in search results. Before the title. Before the screenshots. Bad icon = no tap = no download. Test this: screenshot your icon next to your top 3 competitors. If yours doesn't stand out in 2 seconds, redo it. Design is not vanity. It's conversion.

@om_patel5 I think it's this tool: flowdeck.studio

THIS GUY BUILT A TOOL THAT LETS CLAUDE CODE AUTONOMOUSLY TEST YOUR ENTIRE iOS APP you point it at a simulator and say "test everything" Claude navigates the whole app on its own through the accessibility tree and screenshots. it figures out the UI by itself. it taps buttons, fills forms, opens every screen, tests every feature, and checks every flow in 8 minutes it found every bug the developer missed then it checked the debug logs for errors and gave a structured summary of everything it found no XCUITest scripts, no test maintenance, and no more writing confusing, complicated test cases one prompt and that's it

@athcanft how can appsflyer do a better job than tiktok sdk?

10 tips for running tiktok ads on your ios app: 1. don't use "boosts" in-app, they are vanity scams 2. use smart+ campaign always 3. disable pangle + lemon8 platforms 4. ONLY target USA 5. set "language" to English 6. do NOT use "auto-generated enhancements" 7. run "Subscribe" or "Trial Start" campaigns - NOT install 8. use appsflyer MMP - tiktok sdk is trash 9. let your ads run 2-3 days before killing/promoting 10. don't scale budget more than 30% every 3 days

@Teknium i wish the labs would just focus on making models instead of copying others' ideas

Even anthropic is copying from us now, we must be doing something right! 3 days ago we added notify when done, so the agent does not have to poll the background process(es) that it spawns, and instead when it finishes it can queue up a new message for the agent. Work on other things in the same session while it works! Open Source can move faster than any closed centralized company. See the PR here: github.com/NousResearch/h…

@k1rallik "5x more Codex usage" very little advantage to this plan...maybe after they stop resetting limits.

🚨 do you understand what OpenAI just announced.. ChatGPT Pro is now $100/month. 10x more Codex usage. and Codex isn't autocomplete - it writes, debugs, builds entire apps. autonomously. if you've used Claude Opus for coding you know the feeling. Codex hits the same way. except Claude Opus limits burn through in hours. you blink and you're out. Codex just got a $100 tier built to run 24/7. no cooldowns. no "you've reached your limit" at 2am mid-build. they're not selling you a chatbot. they're selling you a software engineer that never sleeps and never asks for equity. and they're doing it quietly while everyone argues about tariffs.

@SAKSHAM111 @TheTuringPost @karpathy bro trying to dunk on one of the greatest minds in this space and nobody is biting

@TheTuringPost @karpathy this guy posts something nonsense and the whole world takes it as some zenith of AI wisdom. karpathy talk ?? wouldn't that be a echo chamber or just a blog.

@iuliatech time hah

@ZackerySWard That sounds cool. 👍 What stops you from building it?

What's the one app you wish existed but nobody has built yet? Asking for obvious reasons. 👀