<h2>Must know</h2>
<ul>
<li><strong>The rise of agentic coding and the cognitive debt trap</strong>: AI agentic coding is shifting from simple completion to autonomous multi-step task execution, with some developers reporting massive productivity spikes like 94 commits per day. However, a 2026 study highlights a growing ‘cognitive debt’ where AI-assisted engineers scored 17% lower on debugging and conceptual understanding, potentially collapsing the junior-to-senior pipeline. New tools like Context Mode and DiffPrism are emerging to manage these autonomous sessions by reducing token consumption and enabling agent self-review. <a href="https://app.daily.dev/feed-by-ids?id=4gm5j1Hru&amp;id=8Ii7uZjVf&amp;id=TFTdPwhQh&amp;id=rq8g08smm&amp;id=Z5HzQTLHR" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic banned from federal use as OpenAI secures Pentagon deal</strong>: The Trump administration has blacklisted Anthropic from federal contracts due to the company’s refusal to remove clauses prohibiting AI use for domestic surveillance or autonomous weaponry. OpenAI has stepped in to replace them, securing a major Pentagon deal despite claiming to hold similar ethical red lines. Critics like Gary Marcus suggest the public alignment between the two companies was a facade while secret negotiations were underway. <a href="https://app.daily.dev/feed-by-ids?id=1Hvh12XVS&amp;id=O74TphFVu&amp;id=uauecuP4d" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Open source sustainability crisis hits MinIO and Maven Central</strong>: MinIO has officially archived its original repository and moved to a restrictive AGPL license, prompting community forks to restore the admin console. Meanwhile, Sonatype reports that 80% of Maven Central traffic comes from just three hyperscalers, leading to calls for mandatory contributions or tiered payment models for high-volume commercial users. Even the venerable curl project has been forced to shut down its bug bounty program due to a flood of low-quality AI-generated reports. <a href="https://app.daily.dev/feed-by-ids?id=OVpWWbF7s&amp;id=dnF9mDAKN&amp;id=ndB4t3RZj" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>TanStack Start and Data Star challenge Next.js dominance</strong>: New evaluations show TanStack Start offers a 35% smaller bundle than Next.js, while the 10KB Data Star library claims 40x network savings by leveraging native browser hypermedia. <a href="https://app.daily.dev/feed-by-ids?id=JRY4RV9EL&amp;id=PP4PS0y9K" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Rails adds Markdown conversion and devcontainer editor links</strong>: Action Text now supports native HTML-to-Markdown conversion, and a new environment variable fixes ‘open in editor’ links when running inside Docker. <a href="https://app.daily.dev/feed-by-ids?id=EDXDYvaoJ&amp;id=FxgLCVbVm" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>CSS Anchor Positioning arrives to kill JavaScript tooltip logic</strong>: The new API allows attaching popovers and tooltips directly in CSS using the anchor function, eliminating the need for manual coordinate calculations. <a href="https://app.daily.dev/posts/t5iW3XXVR" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Qwen 3.5 open source models outperform GPT-5-mini</strong>: Alibaba’s new 35B parameter model uses a sparse Mixture-of-Experts architecture to beat Claude Sonnet 4.5 in visual reasoning while running on consumer GPUs. <a href="https://app.daily.dev/posts/0m0UpI5Bk" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Biopass brings Windows Hello style biometric login to Linux</strong>: This new open-source suite uses local ML inference via LibTorch to provide face and fingerprint authentication for Linux desktops. <a href="https://app.daily.dev/feed-by-ids?id=bhc2OnxfX&amp;id=eUK6bbMZ5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Svelte leads reactive framework sentiment in 2025 survey</strong>: The latest State of JS results place Svelte at the top for satisfaction, followed closely by Astro which saw significant adoption growth. <a href="https://app.daily.dev/feed-by-ids?id=VHR0VC2mr&amp;id=nYS2c4qim" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Unsloth Dynamic 2.0 improves LLM quantization quality</strong>: The update introduces KL divergence for better quantization and allows running Llama 4 Scout 17B on hardware with limited VRAM. <a href="https://app.daily.dev/posts/Pggi40VtT" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Tinder’s camera roll access is definitely just for chemistry</strong>: Tinder is asking for full access to your camera roll so its AI can ‘match’ you better, which is definitely about love and not about harvesting high-resolution training data or sensitive metadata. Privacy advocates are calling it ‘consent theatre,’ but surely the app that gamified rejection has your best interests at heart. It’s not surveillance if there’s a heart icon involved. <a href="https://app.daily.dev/posts/2RggVXlYX" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The $700 billion data center bet</strong>: Hyperscalers are planning to drop nearly $700 billion on data centers in 2026, apparently operating on the ‘if you build it, the AGI will come’ philosophy. Investors are starting to notice the massive debt piles, but the industry remains convinced that burning half a trillion dollars on GPUs is a sustainable business model. It’s only a bubble if the electricity runs out before the ROI appears. <a href="https://app.daily.dev/posts/0m0UpI5Bk" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vibe coding is dead long live agentic engineering</strong>: Andrej Karpathy has declared ‘vibe coding’ obsolete in favor of ‘agentic engineering,’ because rebranding the same AI-assisted workflow every six months is essential for the hype cycle. We’ve moved from ‘prompting’ to ‘vibing’ to ‘agentic’ in record time, yet we’re still mostly just watching an LLM hallucinate edge cases. It’s not a bug in the code; it’s just the agent’s creative interpretation of your requirements. <a href="https://app.daily.dev/posts/1Hvh12XVS" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
