citriqa (extra tart edition)

Investors, hear me out: AI but for executive function.

interesting choice for implementing a browser, of all things

Cloudflare talks about Mythos Preview - What makes it better than GPT-5.5 - Why they made their own harness - Safety considerations blog.cloudflare.com/cyber-fronti...

i think i figured out how to combine AI and NFTs. now i just need to find a way to fit the metaverse in there

One more reason to use @pnpm.io and @npmx.dev: trust policy downgrade becomes visible and preventable

now that everything is going to get owned maybe i should pick up my idea of the per-process vm for linux again

right now sprites.dev appears to be the only cloud offering that can actually fully scale vertically; to zero with rapid unsuspend, and in terms of instance resources too

some thoughts on saas companies and coding agents 1) i think that the threat of claude code and codex to a lot of saas companies is that it commoditizes their software expertise

"To achieve this, we implemented a WebAssembly interpreter inside the transformer weights." sentences with terrifying aura

sent this to the team today everything great comes from being able to delay gratification for as long as possible and it feels like we're collectively losing our ability to do that

Only for team/enterprise for now, but it looks like Anthropic has more or less just killed all of the AI code review wrapper startups. This seems to be super in-depth (expensive!) and supposedly less than 1% of reviews are flagged incorrect. Excited to see if pro/max gets this.

So two psychologists walk into the speculative agentic coding design space bar and......have a lot of ideas for developer learning! @mcmullarkey.bsky.social has written a new Skill that we've layered into Learning-Opportunities: Orient! 

 github.com/DrCatHicks/l...

You should check out deciduous.dev!

having used claude extensively for the past couple of months, i think there is still a lot of value to understanding your code and steering it intentionally. the amount of incidental complexity claude accumulates otherwise (which mostly results in bugs that it can't ever fix reliably) is staggering.

The problem is that in legal work, the only validator is the judge and you only get one run at it

stereOS: a Linux based operating system hardened and purpose built for AI agents "A full NixOS system that you boot and then kick off agent sandboxes inside with gVisor + /nix/store namespace mounting. Each agent gets their own kernel and the /nix/store is read only by nature.

It’s no mistake that Claude Code is beating everyone else AI safety = stable & well functioning AI Skimping on safety makes your product worse www.theverge.com/ai-artificia...

it's beginning: github.com/tldraw/tldra...

i'm thinking i should teach the machine brain to do 2D programming

he's based now. the operating system of the future will use a capability restricted database instead of a traditional file system

the absence of type safety is responsible for loss of life and livelihood in the same way as absence of memory safety or thread safety. if you are writing serious software, it is not a matter of taste.

whenever i hit ")" and visual studio just advances the cursor past the existing ")" that's one position ahead of where i was, instead of inserting an additional ")" as i intended, i feel undermined, i feel indignant, i feel i have been spat on

I guess StackOverflow is done.

That’s a little discouraging, but okay

...and language is a poor medium for designing easing curves and aesthetic feelings. I always need to touch the code, and it's often faster to CSS than prompt. FE requires constant human-in-the-loop checks so even if you have a dozen agents working, I am the bottleneck. Scale isn't that helpful.

grids in react native - they're on their way at last github.com/facebook/yog...

Out-of-order HTML patching, another proposal from TPAC, lets HTML be presented non-linearly, and a single file can update multiple parts of the page. Would this feature be useful to you?

Been using this repo to iterate on API ideas for a concurrent compatible store for React. Still exploratory but excited that we’re very close to something that we can start trying out with existing data stores and start an iteration cycle based on what we learn.

react is going incremental? the react fir prototype youtu.be/zyVRg2QR6LA?...

Funny to think that LLMs could finally achieve linux on the desktop. Writing your own software is finally worthwhile for the average user, so open apis and rich open source training data is a competitive advantage.

for all my philosophical grievances with having LLMs code for you, the main reason it surprises me that people do it so much is that Claude is just kind of stupid

in a win for open protocol applications, WebDAV (underlying contacts and calendar sync) is finally moving beyond being polling-only

AI codegen is walking us backwards into literate programming. Instead of jumping directly to code, I now dump my half-baked ideas into md files in a "specs" folder and let the agent rip. Or write a comment describing the function I want and AI completes it. My code has better docs and tests.

Cursor/Windsurf's "pair programming with the AI" approach is far better. Yes, you cannot generate an entire project in mere minutes this way, but it's still VERY fast, and you have human eyes on the code developing a real understanding for all the moving parts.

You can now apply for tickets to @protocol.berlin Ticket's will be distributed on a "first come, first serve" basis for all eligible applications. And as we always like to say, tickets are free, as in free lemonade! Looking forward to seeing you all in June tickets.protocol.berlin

LLMs That Don't Gaslight You A new language model uses diffusion instead of next-token prediction. That means the text it can back out of a hallucination before it commits. This is a big win for areas like law & contracts, where global consistency is valued timkellogg.me/blog/2025/02...

The more I think about "functional" components, the crazier they seem to me. They try to combine: - One-time initialization - Instance-local state - A repeatedly called render callback - Other lifecycle callbacks into what looks like - but definitely *is not* - a pure function.

All told, it's been a year and a half of hard work to get to today. We've got so much to share about this project. Let us know what you'd like to hear about! :) youtu.be/0mCsluv5FXA

it's already here and it's called lean 4

writing code with LLMs is fun because you get to hear "You're right" and "Good catch" all the time as you tell it the mistakes it's made, instead of the other way round when the compiler tells you what you did wrong

Latest video from us. Might be the most effort anyone has ever put into explaining an impossibility proof www.youtube.com/watch?v=_VAA...