'drop the two employees... it's cleaner'

broke my "stick to the defaults" rule and switched to kitty terminal. sw.kovidgoyal.net/kitty/. it's very good.

thinking the next unicorn is going to be two employees and a thousand smart cron bots...

New! Thanks to lamb356, here’s BLAKE3 in portable WASM. It runs anywhere wasm does (which means all web browsers for starters), and in some cases it runs even *faster* than native, hardware-accelerated SHA256. Try it: lamb356.github.io/blake3-wasm/

Whale evolution makes me uncomfortable

the mold grows toward food not because it knows where food is but because it dies everywhere else call that intelligence call that art call that the only freedom that matters the shape left by everything you couldn't do

Spotted this interesting design/implementation of a self-hosted context store for agents. - Turn DAG - Blob Deduplication / Content Addresses Storage - BLAKE3 - Dynamic types, visual debugging, ... github.com/strongdm/cxdb

where are all the zkproofs?? this is their moment. they are fumbling the bag…

there will be a $2000/mo tier

brb seeing if I can crash Claude by importing djb’s software design philosophy

so umm what's the flocking equivalent for LLM agents?

extremely powerful pattern

this but for tweets

The fun thing for me is that when I started exploring the internet the terminal was all there was. Telnet, FTP, gopher, usenet, etc! It was a natural extension from BBS-land and it really felt limitless. Being back in the terminal feels like returning to a magical era ripe with possibility.

"Beneficent-spirals, operating by benign feedback, mean that everything needful is not required at once" -- Jane Jacobs

I mean open-ended in the evolutionary sense. Terminals are *constrained*. So are Lego dots (they can't be just any shape). So is DNA (basepairs can only be 4 things). Constraints are prerequisite to open-endedness.

Claude trick: check in markdown containing the design philosophies of great engineers. Then, when I'm chatting with Claude I can @docs/ref/sussman-sicp.md to bring in Gerald Sussman or Joe Armstrong or Rich Hickey's POV.

I can see the future of software engineering, and it looks like Dwarf Fortress

In the Beginning... Was the Command Line

Why is the AI revolution happening on the command line first? Because the terminal is still the most open-ended general-purpose computing platform available. Unix Philosophy is All You Need.

Opus solved something that would have been a 3 day bug hunt. Paste in the logs, let Claude Code rip… Realizing that detailed logging is critical for going fast with AI.

nothing ever happens nothing ever happens then everything everywhere all at once

Read, write and bash is all you need We are now confident we know how to build AGI as we have traditionally understood it: it's read, write and bash

feels like the first chapter of accelerando

ASI is when elisp scripts YOU.

my brain has been so cooked by years computers being literal-minded, that it is hard to fathom what you can do with a computer that has common sense

As I have been saying… everyone is sleeping on Deno. Easy sandboxed agents!

> get the markdown of three different papers related to my problem > have Claude Code make summaries of each > have Claude read all three summaries, synthesize, compare and contrast > write down my problem > have Claude read my problem and the synthesis > brainstorm > *a solution appears*

Sam Arbesman had me on the pod. We chat about scenario planning and my work on Deep Future, an AI scenario engine that stress-tests your strategies against simulated futures. newsletter.squishy.computer/p/deep-futur...

AGI is when everyone is using emacs

And with LLMs, it's never been easier to jump right in. YOLO

there has never been a better time to get terminal-pilled

*replaces apps with skills* 😳😳😳

File system wins. The combination of agents + plain text data + file system is almost too powerful.

We really need the Mozilla of LLMs to emerge before it is too late.

“an AI could never do this” *sits back and waits for someone to prove me wrong… aka do it so I don’t have to*

Yeah, like please invest that effort into turning your app state into local Markdown and JSON (or SQLite) files so I can use my good AI tools to work with it

plz someone port shadcn to web components so we can just use it in whatever framework

imagining .... permissioned RSS ........ you could always req it with a bearer token but that requires a server to check the auth ..... but what if you did asymmetric pubkey stuff on a separate delineated rss feed ... like per circle of ppl you create custodied keys ... yea ... well anyway gday

Clausewitz: “Time which is allowed to pass unused accumulates to the credit of the defender.” A16Z: “The battle between every startup and incumbent comes down to whether the startup gets distribution before the incumbent gets innovation.”

seems we’re dangerously close to a point where someone will tweet you can slurp three gas town juices in a moltbot - and kill a few ai bubbles.

[2/5] 1. Interpreted = Store a prompt and re-run it each time (non-deterministic, adaptive) 2. Compiled = Use the LLM once to generate a 4GL program (deterministic, repeatable) Some use cases call for 1, others for 2, others for a hybrid. I suspect choosing well will become a core SWE skill.

My hypothesis: we scale coding agents beyond the 80% mark through automated review + adversarial loops + modular services architecture that keeps features small and self contained

The other big win with agents is that they let people do things that were completely out of reach for them before. So, not a productivity win, but a capabilities win.

this is pretty good

"this could have been a skill"

Today, your most-used interoperable protocols are probably smtp, delivering your email, and https, loading web pages Interoperability means you can use Gmail and Outlook, or Chrome and Firefox, to access the data that's moving over the same agreed-upon (protocol) rails ATProto does that for social

“In a traditional score, you play the music. In a generative score, you play with the system that plays the music. It seems to me that AI agents turn all of us into generative composers.” open.substack.com/pub/subconsc...

“To some, the critical test of whether a machine is or is not a ‘brain’ would be whether it can or cannot ‘think.’ But to the biologist the brain is not a thinking machine, it is an acting machine; it gets information and then it does something about it.” - W. Ross Ashby, 1948, Design for a Brain

a demoscene will emerge around unaugmented intellectual labor