Claude invented a "type it out explicitly" confirmation requirement for a force push (for the first time ever??) and then proceeded to fully ignore it.

Very happy to be a part of this! 🥳

Claude: Would you like me to handle §8 or move on to §11c? Me: yeh sounds good

I miss Fable.

simply quantize it and release it as Opus 5

Both my Mac and I are now caffeinated

fable 5 max rewrote my linkedin bio for me 🫠

Muve to the beat. Muve the conversation. Muve mountains. Today @eurosky.social announces mu.social, a new microblogging app to experiment with the future of social! 🪩💖

officially launching martijn.gay this pride month

@cloudflare.social being homophobic IN PRIDE MONTH

"and that tension is the whole story." who can guess who wrote this?

Gave #Codex CLI a spin this afternoon with $20 of API credit. TUI feels a touch sleeker than #Claude, but honestly the output's near-identical. Caught some mistakes Copilot missed.

😌

weirdly believable email address for Kash Patel

Is anyone thinking about a) how to control cost in #agentic coding? b) how to know if #claudecode output is trustworthy? please message me! I wanna talk!

great work linkedin

#claude 4.7 is crap. I wrote SIX WORDS.

Put your money where your mouth is! In Europe!

Aging was supposed to happen in base 12, not base 10. You're actually meant to be a kid until 12, a teenager until 24, a 20-something until you're 36, and in your thirties until you're 48. Your 40s should last until you're 60. That's why every decade milestone (30, 40, etc.) feels premature.

I got some questions about what people choose to check into git when doing agentic development. Do you check-in AGENTS.md? .claude directories or files? Hooks? I also asked in rustaceans-ai.zulipchat.com -- if you do Rust coding with agents, you should join that.

I'm fascinated by how good #Claude is at changing its mind: weather_raw = json.loads(open("...").read()) if False else None after the inference for the JSON load was done, it realized that file doesn't exist, so it bailed

why do I prefer doing GH PR reviews from bottom to top???

quick #claudecode tip if you really need it, just do it yourself

My day so far

www.youtube.com/watch?v=dT5I... pretty interesting take on a new fave topic of mine

hmm, let me just go ask my AI what Robert M. Pirsig would think about AI slop

discussing #slop I think everyone should reread Zen and the Art of Motorcycle Maintenance

no more cobol translation startups allowed

"I lost my train of thought" is human for "my context window got cleared"

"here's my Claude chat history" sure is a meetup talk format now

you can tell the exact moment when #claude is about to lose the plot when it starts to write Python scripts to analyze your Cargo deps

Being a worker under capitalism is basically getting yourself all slutty and standing out on a street corner until a VC dude rolls up and goes "do you love solving difficult challenges in a fast-paced environment?" and then you say "I'll love whatever you want for housing and healthcare, baby"

I wonder if #Claude is being trained to be encourage changing its mind sometimes. Seeing lots if "actually, ..." turnarounds. Normally this is in thinking mode and entirely harmless, but here it's done the same thing in a failed attempt to stop a destructive bash command.

I think there's SO many more reasons as well, but this is a great base layer of Good Reasoning.

do you think Markdown might be the right format for this spec doc?

love claude

eslint's terrible 8->9 migration may have just killed it

Just describing a mental model shift I'm experiencing blog.martijnarts.com/agents-what-...

just don't do 1000+ lines PRs and you can still manually review it and confidently take liability for your code?

find a gap in the market and vibe code a solution it's a huge slopportunity

My "absolutely no PRs from anyone exposed to microplastics" policy is raising a lot of questions answered, quite frankly, by the policy itself

This is actually really interesting! Why do we allow Claude to be on the commit byline, but not Zed, or Rust Analyzer, etc?

adhd superpower

The fact that computers now do the correct thing when you say “please rebase” is nothing short of a technological miracle.

just found this tidbit in a CLAUDE.md file: "If the code is not fully optimized before handing off to the user, you will be fined $100."

Remember, AI is non-deterministic, no matter how much you add to your AGENTS.md. If you want determinism, do the same we've done for decades: script it. For example, don't ask AI to "test my app end to end" but instead set up a CI e2e test check.

Code isn’t a business. Long-standing tradition of engineers getting that wrong.

as an aside: I asked the other day whether the difference between Rust and Go matters to an agent. it does. The stuff you can easily leverage with Rust smokes out entire categories of LLM misbehavior that the language doesn't control for. mutants is maybe the best tool I've ever used for testing

This is crazy. I don't like it.