Something about Opus 4.7 feels like it's more for them than for us

Agents in loops are already powerful and surprisingly useful, and you can just tell we're only scratching the surface there still. But ... I really need more tokens. Weekly limits hit in 2 days 🙈 Claude 200x Max plan when? (serious)

Scaling Managed Agents: Decoupling the brain from the hands Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.

They did it again: www.anthropic.com/engineering/... I predict this becomes another standard like MCP or Agent Skills. It elegantly solves quite a lot - others will copy, and rightfully so.

Social gaming startup Rec Room is shutting down its service on June 1, saying it could not find a path to profitability; it was valued at $3.5B in December 2021 By Todd Bishop / GeekWire. View the full context on Techmeme.

Hard to believe 😢 www.techmeme.com/260330/p45#a... Not sure if too early or too late - probably both

BTW this is what I mean when I say "moving up the stack." I'm not even the one in Claude Code anymore for this type of stuff, just setting high-level direction and (less, and less) providing feedback

GitHub - ericflo/gitoxide-fs: A blazing-fast FUSE filesystem backed by git. Every file edit becomes a commit. Built for the agentic era. A blazing-fast FUSE filesystem backed by git. Every file edit becomes a commit. Built for the agentic era. - ericflo/gitoxide-fs

Ok I'm exaggerating about the overnight thing - it was 3 days ago - but all autonomous. Here, I open sourced it. Why not? github.com/ericflo/gito...

...added large file blob pointer support, cut a release build, and then started using it. Claude planner codex implementation looper. The future is pretty cool.

I vaguely gestured to my cloud agent that it would be cool if file changes were automatically tracked as git commits. It got so excited, overnight it created a rust repo to mount git repos as fuse drives, did 50 PRs to it...

DLSS 5 looks like the greatest single leap forward in real-time graphics since the programmable shader. Incredible.

Agentic development naturally pulls development into the product itself. While you're using your app, you leave feedback in-situ, agent improves code and redeploys, where you can leave more feedback, loop

Yeah both at work and on side projects - I really think this is the appropriate level of abstraction for this next era that's beginning

Time to move up the abstraction ladder again. We just passed the 1-year anniversary of Claude Code, so this rung on the ladder lasted about a year. Wild!

All Github had to do was be the best hub for git

No right to relicense this project · Issue #327 · chardet/chardet Hi, I'm Mark Pilgrim. You may remember me from such classics as "Dive Into Python" and "Universal Character Encoding Detector." I am the original author of chardet. First off, I would like to thank...

github.com/chardet/char... 😟

I used to look up to Mark Pilgrim 😟

Qwen 3.5 may have crossed the chasm, or at least it's getting close. It seems genuinely useful in an agentic code setting, particularly with Qwen Code, and it comes in weight packages that you can reasonably run on home hardware. This is pretty cool - I could turn off internet and still vibe code.

This is so dumb AND it makes me triply-mad because it means even less incentive for open source AND it means some of the more annoying AI Safety concerns were right, AND it will push our enemies into pursuing automated weapons faster. Ugh ugh ugh. And for what?

I think it's irresponsible how much people are being pushed towards MCP still. It's just inertia at this point. Use a skill and package a library or a CLI with it. Preferably a library, even though everyone is doing the CLI thing. (CLIs are just libraries for the bash programming language.)

I have never been able to find better coffee than the Berkeley Blend by Uncommon Grounds. I am not some big coffee knower, but every other bean I've tried, nothing has (yet) tasted better

This means Anthropic didn't cave - that takes guts and my respect for Anthropic goes way up for standing on their principles. Also WTF is the government thinking here?!

In this populist era the "progressives" are trying to stop progress (cancel data centers) while the "conservatives" are reaching the federal government ever-deeper into state, local, and personal affairs. Yikes.

Wow thank you, and likewise - glad we've been able to work on the same sandcastles lately!

I don't know for sure, but my gut feeling is in the other direction - that it all becomes a commodity. In some ways, if it's capital-intensive that is kind of easier to wrap my head around. If all software becomes cheap and plentiful, that seems harder to navigate as an industry.

The whole software factory thing is real and powerful. Lately have been feeling the magnitude of the moment. The opportunity and the challenge of it. Things will be so different even in the next few years. And yet, we will be the same.

It's curious how correlated it has been with nuclear activity. Some of the videos from the Nimitz seem to credibly defy publicly-known physics. Those NJ drones seem to have been something real and weird. If it's all real, maybe interdimensional or breakaway/underwater society. But probably a cover.

As someone who has tracked the UAP subject over the past several years (since the 2017 NYT article and original Obama revelations) my credulity on the topic is at its nadir. That it's being trotted out right now, at this time, just isn't believable. It may always have been cover.

Ex-Sony Pictures Boss Regrets Seth Rogen’s ‘The Interview’ and Says Obama Told Him After the Sony Hack: ‘What Were You Thinking? Of Course That Was a Mistake’ Michael Lynton, former Sony CEO, wishes he had thought more carefully before greenlighting "The Interview" from Seth Rogen.

Ah yes, your mistake was giving artists too much free speech, not your negligent security practices. Good job, you found the exact right takeaway from the situation. /s variety.com/2026/film/ne...

If you could somehow distill GLM-5 into a fast phone-sized model, you would make history, and probably a billion dollars. There's certainly some "trick" we'll figure out in the future that makes it easy to do. The only thing keeping you from the history books is figuring out that trick. Wild.

Did `claude -p` start using API tokens instead of MAX plan tokens? I think it happened in the middle of the night. Because I had a `claude -p` loop running last night, and at 78% usage, it stopped counting against my plan, and now I have $hundreds in API token charges today. WTF! Not a safe platform

Falling down a youtube rabbit hole of tech documentaries youtu.be/xsVIvM82Rfc