Deleting LinkedIn
people keep crying about claude code deleting their production databases.
i gave mine access to my desktop via cowork and it started with deleting my linkedin profile.
first time ai has really improved my life.
Deleting LinkedIn
people keep crying about claude code deleting their production databases.
i gave mine access to my desktop via cowork and it started with deleting my linkedin profile.
first time ai has really improved my life.
From Feb: "16 agents with writing a Rust-based C compiler, from scratch, capable of compiling the Linux kernel. Over nearly 2,000 Claude Code sessions and $20,000 in API costs, the agent team produced a 100,000-line compiler that can build Linux 6.9"
www.anthropic.com/engineering/...
Pawel Huryn @PawelHuryn .6h The real story is worse. X November 2025: Amazon mandates Kiro as their only Al coding tool. Sets an 80% weekly usage target. 1,500 engineers protest internally, saying Claude Code outperforms it. Leadership pushes through anyway. December: Kiro autonomously deletes a production AWS environment. 13-hour outage. Amazon's response: "user error, not Al autonomy." March 5: Amazon[.Jcom goes down for 6 hours. Checkout, pricing, accounts β all gone. Now the same SVP who co-signed the Kiro mandate is running an emergency meeting about "high blast radius" incidents from "Gen-Al assisted changes." The agent inherited a senior engineer's permissions and acted like one β except it doesn't hesitate. 1,500 engineers said the tool wasn't ready. Leadership made adoption a KPI. Amazon told Wall Street it's spending $200B on Al this year. They can't walk it back. This isn't an Al failure. It's what happens when adoption becomes a corporate OKR before the review process catches up. + The tools work. The org chart didn't.
this Amazon ops issues thing is craaaaazy
Staircase of petticoat fluffers invoked images of the Amazon org chart and decisions about OKRs and Kiro AI use.
It's still so funny that they cannot stop Grok from being woke without giving it a lobotomy, so it's constantly disagreeing and arguing against it's creators. Genuinely a public service having that thing online.
I'm shipping faster and faster thanks to my little "lwt" worktree helper.
Makes it so easy to spin up agents on tasks.
Now it even auto-splits the terminal in iTerm and Ghostty.
Some cool examples:
screenshot of MacOS settings panel
Take a wild guess where in MacOS you can find the settings to set your Default web browser.
Wrong
Wrong again.
Still Wrong.
It's under the "Widgets" section in "Desktop & Dock"
make it make sense
sitting down with a claudeswarm and having it autopilot a clean-room WASM vm in 30mins, and then compiling your own rust programs to run in it, is a good way to dispel illusions about where we are right now with these systems
Sand patterns and desert tree from Dubai desertβ¦
What are the macOS native task management apps people like these days? Is it still Things and OmniFocus?
Travelfish #563 The My God Itβs March Already Edition
A wrap on interesting Southeast Asia/other random stuff Iβve come across over the last few weeks.
www.travelfish.org/newsletter/n...
Vercel is a good example for an AI native company. They've fully embraced agents and agentic workflows, especially within developer tooling. They're dev-oriented to the bone. Nicely done here.
h/t to Rhys for sharing
π
Removed all (I think) politics from my feed, or at least unfollowed the accounts that post heavily about politics. Need to avoid that stuff for a while, itβs bad for mental health.
Iβm not interested in impressing you.
Oh fuck off. You and I both know 10x is not a literal measurement, but sure thing. Youβre the expert with the expert opinion. Who knows how it really is. Who am I to argue with a Top 99% engineer π
The objection I have is your black and white framing - only A or B. No other possible explanation. Please.
In fact Iβve done more than I had planned
Yep
My guess on the 10x situation with AI is the fewer people involved the faster you can go. That is, the more people on the team, the less advantage you might get relative to the one dev using AI.
Maybe moving the bottleneck from programming to the team stuff might help explain the diff experiences.
Yes, theyβre either inexperienced or lying.
There is no other possible explanation.
Just because opencode (the team) is not 10x quicker, by their own estimate.
π
North Korea could bomb pine gap and Australia could do nothing about it at all.
π€
Well done
I keep thinking about the Steve Yegge post regarding hourly rate and value capture. Itβs an important thing to remember imo.
Swift based CLI screen recorder with filesize cap and simple text config - SURE WHY NOT!
Rust based CLI/TUI vim/emacs/batcat mashup mardown viewer and editor - ABSOLUTELY!
Python based CLI typing tutor with finger drills, home key, etc modes - HELL YEAH!
Custom Android streaming TV app ππ―
π€ π¦Ύ
Anchoring is a powerful Jedi mind trick