Man, those Claude safety guardrails make complete sense now

Number 1 way to avoid all the issues and complexities of distributed systems. Do not make it a distributed systems problem, YET

How does pangram make sense for personal use, especially for people who are in the tech space? Because I saw a few using the browser extension. If I read something, and it's good enough that I am not sure whether this is human or ai written, in that case do I really care enough to use an external service to validate my own validation? At this point don't we all already have our own slop filters? But I do see how it can benefit my non technical friends who don't spend 13 hours of their day talking to models.

i have a theory on why a lot of the new companies seem to be building the same thing. Take ai sandboxing for an example, if you ask claude for the most "impactful and neeed of the hour" startup ideas, an ai sandboxing company is definitely going to make that list. Go ahead, try it. And once you start working on that, guess what will claude name the project/company? "box", "boxes", or some variation but definitely a box in there. will that stop me from launching another sandbox product? of course not. just need to decide on the right name. "icebox", "firebox", "outbox", "boxy", "gearbox", "outbox", "pandora's box"

Asking Claude to be “more creative” and “more ambitious” is like telling someone with depression to just work out. If they could just do that, they wouldn’t be talking to you in the first place.

It sometimes feels like we haven’t really built anything as revolutionary as tech like docker, k8s, firecracker, rust, quic, go, ebpf, wasm, or the modern cloud/virtualization stack post ai. I think the reason is, most of that was built in the 2010s. And if we consider the post ai world from opus 4.5, that's like a year old. And on top of that, foundational tech depends more on network effects and adoption and getting enough of the ecosystem to switch and stick with it than just building the thing. That still takes years. AI does help with the building, but the adoption curve hasn’t changed as much. In fact, it seems like adoption cycles will be much longer now? Since everyone is going through this "reverse engineering" and "build my own" psychosis, and very hesitant to give other people's software a chance. Lol, I still remember, how long the discussions on "rolled my own auth" used to be.

Was building a cloud dev setup which matches my laptop’s node, bun, pnpm, rustc, and go and lets each project’s pinned version win, and surprisingly it was harder than it should be for Node. Rust and Go already do this natively (rustup + rust-toolchain.toml, Go’s go.mod). Node doesn’t. nvm only switches on an explicit `nvm use` or a shell `cd` hook, so editors, scripts, and CI miss it. Volta wants its own package.json key. mise by @jdxcode beautifully fixes it with one binary in the home folder (no root). Its shims sit on PATH, read .nvmrc / .node-version / package.json’s packageManager, and install the missing version the first time you run it, covers node, bun, and pnpm. Went through my projects afterward and almost none of them actually pin a Node version. A few pin the package manager in package.json, and far more have pnpm or bun lockfiles than .nvmrc files. So the versions of bun and pnpm themselves needed to be correct too, mise also handles that btw.

Not a very gentle reminder that you can put ML models and neural nets on the hot path to reduce latency, of course not through raw performance, but by reducing the overall work or predicting the work.

Compute poor so bad that I’ve turned into a human CPU scheduler. I wake up every morning with one conviction: 100% CPU utilization on every device within a 10m radius. Current progress: my laptop. Remaining: everyone else’s and their grandmother's.

I know we’re past the “read the code/don’t read the code” argument. But if someone can come up with a human consumable representation of the code which is right in the middle of lossless raw code reading and lossy description by the agents, I am willing to spend a fortune.

Social media, streaming platforms, news outlets, advertisers, gaming companies, and influencers have all been buying attention. Now employers are paying employees for it too. Attention is all you need indeed.

Somehow every single post here which talks about how powerful the skill of “pattern matching” is, blows up. There, I matched another pattern.

We all finally have a better FEEL of how long milliseconds are because of agents

One of the underrated wins of going native Swift + Kotlin, instead of cross platform, is even if you don't have a dedicated app designer in your team, the apps will look just fine, they will look pretty bland I admit, but the intuitive UX of native makes up for that.

Tried to get rid of Claude slop talk by asking it to respond in hindi. It completely breaks the terminal UI, and now I see बिल्कुल everywhere. बिल्कुल..... बिल्कुल..... बिल्कुल.....

A newsroom wall built with 15 cloud computers in @orgo , each one shows a site I'd normally keep a tab open for. hacker news, github trending, status pages, npm advisories, github status (of course) jev by @typesafeai reads every new headline and tells me if I should look at this right now. when something crosses the line, the bell rings and that computer opens the story by itself.