Tech & AI Daily
Anthropic downgraded prompt cache TTL on March 6th with zero announcement, and the Claude Code issue thread hit 449 upvotes as developers realized it was a stealth price hike. If you run any agent pipeline or heavy Claude Code sessions with caching enabled, check your API costs right now because they may have quietly jumped.
Anthropic's best model is real and it's not coming to you anytime soon. The same week they announced a 50-company pilot program for Mythos, Zhipu AI open-sourced a model that beats GPT-5.4 on coding, making the contrast between closed frontier labs and the open source community starker than ever.
Google, Meta, Alibaba, Mistral, OpenAI and Zhipu all shipped open-weight releases in a single week, and the landscape now has serious contenders at every tier. Open source is not playing catch-up anymore, it is setting the pace.
Steve Hanov's post on staying lean is the highest-scoring HN story of the day at 793 points, and the thread is mostly people agreeing rather than disputing it. If you are adding infrastructure complexity before you have the revenue to justify it, this is the read that will make you feel bad about that.
A well-argued essay on how cross-platform frameworks have sanded the personality out of software, making everything feel like the same gray app regardless of what platform you are on. Hits harder now that most AI coding tools default to producing identically bland UIs.
Mistral published a strategic document framing European AI as a sovereignty play against US and Chinese model dominance, complete with policy recommendations. More interesting as a signal of where EU regulation and procurement are heading than as a technical doc.
Iceland, Nepal, Bhutan and four others have fully crossed the renewable threshold, which is worth noting for anyone thinking about where to locate compute or how to frame sustainability claims. The story gets buried under AI news but the data point is real and significant.
Tanya Verma makes a compelling case that AI has shifted from frontier expansion to consolidation, and the new challenge is reliability and deployment rather than capability. If you build on top of models rather than building models, this is the framing that will shape your next year.
Cantrill argues that the culture of intellectual laziness enabled by AI tools is eroding the deep engagement that produces real breakthroughs. Agree or not, it is one of the more honest takes on what we might be trading away for faster output.
A Show HN wrapper around Claude Code that adds custom keybindings, session management, and workflow shortcuts for people who live in the terminal. Directly relevant if you are heavy on Claude Code daily and want more control than the defaults give you.
Moves your app launcher into a horizontal taskbar at the top of the screen instead of the Dock, which is exactly what a lot of Mac users coming from Windows or Linux have always wanted. Clean, minimal, and free to try.
Subscribe and get Tech & AI Daily delivered to your inbox every morning.