Peter Warnock 🇺🇸

@EnoReyes 🤕

Anthropic and Github are on their way to three nines of reliability: 89.99%

Welcome back ultrathink

@Grady_Booch @Scobleizer @sama It's not ethics; it's hubris.

IMHO I think this is utter nonsense. You are searching for a technical reason where there exists a very simple human one. Neither company has a particularly sound ethical grounding, as demonstrated by their outright theft of training data and their lack of guardrails, despite all their PR positioning that says otherwise Rather, it seems to me that OpenAI said yes where Anthropic did not is largely because in his pursuit for revenue and relevance, Sam has absolutely no ethical limits to his actions whereas Dario has slightly more.

HUGE BREAKING NEWS: Why OpenAI and @sama could say yes where Anthropic couldn't. Hint: it's architectural. OpenAI can say yes while Anthropic can't BECAUSE OF HOW THEY ARE BUILT. HUGE analysis of all of X and everything is going on between OpenAI and Anthropic: docs.google.com/document/d/1NL… THIS IS THE BEST WRAPUP YOU CAN FIND ON X. Thank you to the X API and @blevlabs for providing me with his AI, which is way way way better at doing this stuff than any other publicly-available AI (I'm the first outside of large enterprises to have access to this). Sam can play with the redlines, while Anthropic can't. Which is why @DarioAmodei had to play hard ball. And this has EVERYTHING to do with our civil liberties and safety. This is the most important thing for everyone to understand. Please reshare this widely so everyone understands what the heck is going on here. So important.

@thdxr Happy Birthday 😘

fellas is it gay to be born in february (it's my birthday)

It's an antipattern in a team environment. "Works on my machine."

@ibuildthecloud I think about this, too, but I'm favoring monorepos. It's like the old modular vs microservice debate; is there enough discipline to respect boundaries? I launch the agent in a package scope if I don't want it to see the whole project. It has to ask to go outside of it.

I'm not sure monorepos are necessarily the best way for an agent. It's like you want nested agents. The issue is just that if you ask the agent a question, it has to do a lot of analysis because it has visibility to so many components.

The status alerts are as broken as the platform [500] @claudeai

@jarredsumner @cirospaciari Noticeably improved. Thank you!

Claude Code's p99 memory usage dropped by 40x in the last two weeks, and by 6x since January - while growing rapidly & continuing to ship tons of new features.

SaaS is doing a fine job of putting themselves out of business without being replaced by AI

A 20% price increase?! @1Password

@simonw I experienced this with podcasts first and learned to take intentional breaks and intentional reviews to process and internalize.

Short musings on "cognitive debt" - I'm seeing this in my own work, where excessive unreviewed AI-generated code leads me to lose a firm mental model of what I've built, which then makes it harder to confidently make future decisions simonwillison.net/2026/Feb/15/co…

@thdxr Figure how to meet supply with demand

if a technology made your employees 5x more effective overnight would you

@claudeai I never get the survey when things are going south. 🤷‍♂️

@ibuildthecloud System prompt

Why would you chose one coding agent TUI over the other, assuming they have the same models available?

@GeoffreyHuntley @ibuildthecloud Can’t they be complementary?

@ibuildthecloud it’s why i enjoy simple loop primitives

I'm telling you, this is where they jump the shark. Don't try to solve distributed problems. Even better yet, don't create a distributed problem before you even need it. code.claude.com/docs/en/agent-…

@ibuildthecloud I think it’s a stepping stone to distributed cloud orchestration. We’re the guinea pigs flushing out the mechanics of how to daily drive it without knowledge of the concerns. If I could timeshare an optimized platform, I wouldn’t need to buy expensive, capable local machines.

@thdxr My brain read OpenCD with the shortened domain and thought it was hinting at a future roadmap.

investigating a bun issue - normally i'd use codex for something like this but wanted to see how kimi k2.5 would do - didn't expect much but it looks like it figured it out - and really fast too opncd.ai/share/jupHGgzB

The “Default” is changing between Opus and Sonnet. 🤨

Did the context window get smaller on @claudeai ?

@ibuildthecloud I think @Letta_AI uses a sliding window and is focused on memory. Details are remembered in session and recalled from different types of memory. There is an anthropic-compatible enricher that I haven’t tried yet.

I really want the models to just natively support compaction. It seems like such an interesting idea. Are there research papers out there about this? Somebody has to be doing this?