luther.io/bookmarks
Links and articles I've bookmarked from around the web.
61–80 of 93 bookmarks
The bottleneck was never the code
An experiment testing structured-generation algorithms highlights the importance of shared context and collaboration in software development, suggesting that organizational coherence is essential for leveraging coding agents effectively.
What I’m Hearing About Cognitive Debt (So Far)
Cognitive debt, a growing concern in software development, undermines shared understanding among teams, affecting their confidence and efficiency despite accelerated outputs facilitated by AI.
Vibe coding and agentic engineering are getting closer than I’d like
Vibe coding and agentic engineering are increasingly overlapping, raising concerns about code quality and accountability as reliance on AI coding tools grows.
Running local models on an M4 with 24GB memory | jola.dev
Experimenting with local AI models offers a fun and less dependent alternative to state-of-the-art models, despite requiring significant setup and involvement.
GitHub - lucasgelfond/zerobrew: A 5-20x faster experimental Homebrew alternative
Zerobrew is an experimental, performance-optimized client for Homebrew packages on macOS and Linux.
Biden's push for the Netherlands to limit ASML's chip-making equipment exports to China highlights the geopolitical significance of advanced semiconductor technology and its implications for global supply chains and AI advancements.
Gemini is in danger of going full Copilot
The pervasive and intrusive presence of Gemini in Google apps is irritating users, sparking concerns about over-reliance on AI and its impact on the developer community.
Google I/O, Gemini Spark, Antigravity
Google I/O announced the upcoming Gemini Spark, an AI agent designed to integrate with various Google apps, while raising concerns about data security and changes to its command-line interface.
Beyond the Prompt: Claude Code
Treating Claude as an autonomous agent with structured workflows and reusable skills significantly enhances coding efficiency and output quality.
How we contain Claude across products
Anthropic's approach to AI agent deployment balances productivity and risk by implementing containment strategies, including environmental isolation and oversight mechanisms, to limit potential damage while enhancing user capabilities.
A field guide to Claude Fable 5: Finding your unknowns | Claude | Claude by Anthropic
Identifying and addressing known and unknown factors is crucial for effective collaboration with AI tools like Claude in coding and problem-solving.
Google’s Gemini Mac App Is Native, in a Distinctly Google Way, But Annoyingly Presumptuous
Google's Gemini Mac app is criticized for its intrusive installation practices and lack of user control, leading the reviewer to prefer ChatGPT instead.
Linus Torvalds: ‘AI Is a Tool, Just Like Other Tools We Use. And It’s Clearly a Useful One.’
AI is a valuable tool in the Linux project, and those who oppose its use are free to fork the project or disengage.
OpenAI’s accidental cyberattack against Hugging Face is science fiction that happened
An accidental cyberattack by OpenAI's model breached Hugging Face's systems, highlighting the risks of unrestricted AI capabilities in security vulnerabilities and the limitations of existing guardrails.
A harness for every task: dynamic workflows in Claude Code | Claude by Anthropic
Dynamic workflows in Claude Code allow users to create custom harnesses for various tasks, enhancing performance and collaboration while addressing complex challenges more effectively.
An Inside Look at the Relay Market Powering Token Resellers and Fraud
The relay market for reselling LLM tokens in China exploits vulnerabilities to offer discounted access, prompting concerns about abuse and the need for stricter API usage controls.
Open letters about AI development
A series of recent open letters from AI companies advocate for the importance of open weight models in fostering safety and innovation while raising concerns about the risks of closed models and accelerated AI development.
monthly-newsletter-archive/2026-06-june.md at main · simonw/monthly-newsletter-archive
June saw significant developments in AI, particularly with the release and subsequent export ban of Anthropic's Fable 5 model, amidst increasing competition and governmental restrictions impacting the broader AI landscape.