#648 – October 19, 2025
8 hours total over 3 days and $15.98 later
Vibing a non-trivial ghostty feature
17 minutes by Mitchell Hashimoto
Practical case study where Mitchell created an unobtrusive macOS update notification feature for Ghostty. It took him 16 AI sessions costing $15.98 and about 8 hours total over 3 days. He emphasizes using AI as an assistant rather than replacement, always reviewing and manually improving the generated code before shipping.
What Shopify uses for remote PR reviews
sponsored by Tuple
Tuple created the screen sharing app for developers that Figma, Stripe, and Shopify prefer for unblocking PR reviews, making technical decisions, and pair programming. With shared control by default and intuitive annotations, Tuple is the most effective way to review complex technical issues with your team. Don’t wait for an async review; Tuple instead.
How I almost got hacked by a 'job interview'
5 minutes by David Dodda
David nearly fell for a sophisticated scam disguised as a legitimate coding interview. Scammers used fake LinkedIn profiles and company pages to send malicious code hidden in a professional-looking React project. The code contained obfuscated malware designed to steal crypto wallets and sensitive data. An AI code review caught the threat just before execution, highlighting how attackers are targeting developers through familiar job interview processes.
Multi-core by default
42 minutes by Ryan Fleury
Developers typically treat multi-core programming as a special case within single-threaded code. They use job systems or parallel loops that hand work off to multiple threads, then join results back together. This requires complex setup code, scattered control flow, and difficult debugging. Ryan suggests writing multi-core code by default is a better approach. Starting with all threads running the same program, then using simple constructs like lane index, lane count, and barriers to coordinate work.
Examples are the best documentation
2 minutes by Rakhim Davletkali
Most technical documentation fails developers by lacking simple examples. When switching between projects and languages, developers need quick reference points rather than dense explanations. A Python function like max() requires understanding complex syntax before you can use it, but a few code examples would instantly show how it works.
Hazardous states and accidents
8 minutes by Chris Stjernlöf
Safety depends on avoiding hazardous states, not just preventing accidents. A hazardous state plus bad environmental conditions equals an accident. Since we can only control systems, not environments, we must keep systems out of hazardous states. Multiple controllers work together to maintain safety constraints through feedback, mental models, and control actions. When any of these fail, hazardous states occur.
And the most popular article from the last issue was: