<h2>Must know</h2>
<ul>
<li><strong>The rise of cognitive debt in agentic coding</strong>: A 2026 study found that developers using AI agents scored 17% lower on debugging and conceptual understanding, signaling a growing ‘cognitive debt’ where code production outpaces human comprehension. While tools like Poly Pilot and Claude Code enable massive productivity gains—up to 94 commits per day—the junior-to-senior engineering pipeline is reportedly collapsing as humans lose the ability to effectively review AI-generated output. Experts at NDC London 2026 warned that this decoupling leads to higher change failure rates and longer recovery times despite the initial speed boost. <a href="https://app.daily.dev/feed-by-ids?id=8Ii7uZjVf&amp;id=TFTdPwhQh&amp;id=PDfxSZnrt&amp;id=Rruw1sx8X" 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 quickly stepped in to replace them, securing a major Pentagon deal despite Sam Altman’s public claims that OpenAI maintains similar ethical red lines. Critics like Gary Marcus suggest the move was a calculated negotiation by OpenAI, while safety advocates argue that lobbying against binding regulations left these companies vulnerable to such executive whims. <a href="https://app.daily.dev/feed-by-ids?id=1Hvh12XVS&amp;id=O74TphFVu&amp;id=uauecuP4d&amp;id=p4mVC21J7" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>MinIO archives official repo as community forks emerge</strong>: MinIO has officially archived its main GitHub repository, transitioning to a restrictive AGPL license and removing the admin console from its community edition. In response, the <code>pgsty/minio</code> fork has emerged to restore the full console and binary distribution, with the author arguing that AI coding tools now make solo maintenance of such large projects viable. This shift highlights a growing sustainability crisis in open source, where hyperscalers consume 82% of repository traffic without contributing back, leading registries to consider tiered payment models for commercial users. <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>Unsloth Dynamic 2.0 GGUFs optimize Llama 4 Scout</strong>: New dynamic quantization using KL divergence allows the 17B Scout model to run on consumer GPUs with significantly improved accuracy. <a href="https://app.daily.dev/posts/Pggi40VtT" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Svelte March update adds programmatic component instantiation</strong>: The latest release introduces <code>createContext</code> for manual mounting, TrustedHTML support, and server-side error boundaries. <a href="https://app.daily.dev/posts/VHR0VC2mr" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Biopass brings Windows Hello style biometric login to Linux</strong>: This open-source suite uses local ML inference via LibTorch and YOLO for secure face and fingerprint authentication on the desktop. <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>Data Star library challenges React with 10KB hypermedia</strong>: By leveraging SSE streaming and HTML data attributes, this no-dependency library claims 40x network savings over traditional SPAs. <a href="https://app.daily.dev/posts/PP4PS0y9K" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Context Mode reduces Claude Code token usage by 98 percent</strong>: This open-source MCP server extends usable session time from 30 minutes to 3 hours by aggressively managing the context window. <a href="https://app.daily.dev/posts/rq8g08smm" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>High severity ReDoS vulnerabilities patched in minimatch</strong>: Foundational npm package minimatch fixed three flaws that could stall the Node.js event loop via catastrophic backtracking. <a href="https://app.daily.dev/posts/EvvwDJaxE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Qwen 3.5 Medium models outperform Claude Sonnet 4.5</strong>: Alibaba’s new 35B MoE model activates only 3B parameters per token and supports a 1-million-token context window on consumer hardware. <a href="https://app.daily.dev/posts/0m0UpI5Bk" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The vibe coding to agentic engineering rebrand</strong>: Andrej Karpathy has declared ‘vibe coding’ obsolete in favor of ‘agentic engineering,’ because apparently, we need a more professional-sounding term for letting an LLM hallucinate your codebase. It’s the same ‘prompt and pray’ workflow, just wrapped in a new layer of industry jargon to make senior engineers feel better about losing their debugging skills. Expect a certification for ‘Agentic Vibe Architect’ by the end of the quarter. <a href="https://app.daily.dev/feed-by-ids?id=1Hvh12XVS&amp;id=8Ii7uZjVf" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Tinder wants to scan your camera roll for love</strong>: Tinder’s new ‘Chemistry’ feature uses AI to dig through your private photos under the guise of better matchmaking, a move privacy experts are calling ‘consent theatre.’ It’s a classic data grab where sensitive metadata and computer vision analysis are sold as a ‘personalization’ feature. If you thought dating apps were toxic before, wait until an algorithm decides your compatibility based on your unedited screenshots and cat photos. <a href="https://app.daily.dev/posts/2RggVXlYX" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The mathematical proof that AI can never be safe</strong>: A new ‘mathematical proof’ claims AI cannot simultaneously be safe, trusted, and generally capable, which is a convenient way to sound profound while saying absolutely nothing actionable. It’s the academic equivalent of a ‘Do Not Enter’ sign on a fog bank—high on theoretical drama but ignored by everyone actually building the tech. We’re trading empirical engineering for philosophical doom-posting because it generates more engagement. <a href="https://app.daily.dev/posts/i7Ojr8LXm" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
