danscan

@DenisJeliazkov If you can’t explain proficient computer navigation to someone you hired to navigate a computer, fire them

Linear is beautiful. But... Try explaining it to a new team member: - "Use K to open command" - "J and K to navigate" - "Arrow keys to expand" - "Space to preview" 16 keyboard shortcuts to remember. The real cost? Onboarding time: 3 weeks. Beautiful ≠ Accessible.

too many people who don't love The Computer have been given the privilege of working on The Computer

Or worse, they DO know how to use a computer, but you don't trust them

Having an IT department means you employ people who don't know how to use a computer

Make it fun, make it right, make it fast

I await the day agents and machine payments commodify platforms so they no longer have their customers' identities or contact info

USD.m = meatspace usd

@WarrenInTheBuff Shit implements Steppable

btw, this is a JS runtime as tiny as a font that lets you safely eval guest code inside a Vercel function (or anywhere else WASM runs)

Published a tiny repo for safely evaluating untrusted JS: github.com/okokoklol/mqui… It even works on Vercel without expensive and limited sandbox VMs

If you want to run an agent that behaves like OpenClaw via the Vercel AI SDK, check out my new repo: github.com/okokoklol/docu… It implements the documents-as-brain pattern and can be backed by a real filesystem, or any database. Works in both stateful and serverless envs

CBS, to the extent it has any awareness in excess of what this interviewer has displayed, should be ashamed. From the first second, I was taken aback by how amateur the kickoff of this conversation was. The first question was essentially "you are the CEO of Anthropic, is that right?" Be serious. The fact that denying service in support of domestic mass surveillance and autonomous killing machines was a deal-killer is so obviously insane and revealing of the DoD's intent, or at the very least their self-destructive levels of ego. State-of-the-art models still have primitive means of effecting action, and if you'd work with them at frontiers, you'd be shocked how often they'll tell you the equivalent of "it's heads" when it is indeed tails.

The worst thing about @cursor_ai is modal keyboard shortcuts. ⌘N should *always* be New Buffer - making it "Undo" in review mode is hostile to people who can use a computer. Same for ⌘K being double-bound in Terminal. - Keyboard Supremacy, the rest of y'all pray for UBI 🖕

@joshmohrer I'm going back to baseline 📉

Prompt engineering works

@displaydriver 3>3, 3>5, 4>5, 5>2

Which one earns your attention?

If you want to write SOTA software, your understanding of the possibility space should be informed primarily by hardware Language ecosystems and package repos are data points, but if you treat them as bounds, your posture toward eng is fundamentally that of a consumer

@RhysSullivan wonder what consciousness feels like for the kind of person who would let this go out

what did they do to the dog

AI can be a boost to both creators and consumers. Most engineers are consumers– package jockeys standing on the shoulders of giants I’ve seen folks like that have life-changing outcomes thanks to AI But if your work is on the SOTA itself, it’s exponential (understatement)

@adamwathan don.noyes.asia.wiki.org/view/blue-plane If you ask AI to help you to navigate the pink plane, it’ll do it. Unfortunately, unless you are on the blue plane, you can’t even see the space of possibilities, so it’s impossible for you to ask for anything profound

I don’t think AI is lowering the barrier to entry as much as it is giving the very smartest people an even more enormous advantage.