<h2>Must know</h2>
<ul>
<li><strong>The rise of cognitive debt in agentic workflows</strong>: A 2026 study found that while AI agents like Claude and Copilot can drive massive productivity gains—up to 94 commits per day—developers using them scored 17% lower on debugging and conceptual understanding. This decoupling of code production from comprehension is creating ‘cognitive debt,’ leading to higher change failure rates and a collapsing junior-to-senior engineering pipeline. Experts at NDC London warn that the ‘review paradox’ is real: humans are losing the ability to effectively audit the flood of AI-generated code. <a href="https://app.daily.dev/feed-by-ids?id=8Ii7uZjVf&amp;id=TFTdPwhQh&amp;id=Rruw1sx8X" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic federal ban triggers OpenAI Pentagon deal</strong>: The Trump administration has banned Anthropic from federal use due to its refusal to remove contract clauses prohibiting AI use for domestic surveillance or autonomous weapons. OpenAI quickly secured a Pentagon deal to fill the vacuum, despite Sam Altman’s public claims that OpenAI maintains similar ethical red lines. Critics like Gary Marcus suggest the public alignment between the two companies was staged while OpenAI secretly negotiated the competing government contract. <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>MinIO archived as open source sustainability hits a wall</strong>: MinIO’s official GitHub repository was archived in February 2026 following a shift to AGPL and the removal of the admin console from its community edition. This move highlights a broader crisis where 82% of Maven Central traffic comes from less than 1% of IPs, mostly hyperscalers treating open-source repositories as free CDNs. In response, a community fork of MinIO has emerged, leveraging AI coding tools to maintain the full feature set independently. <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 quantization</strong>: The new release uses KL divergence to improve quantization quality, allowing the 17B Scout model to run on consumer hardware with minimal loss. <a href="https://app.daily.dev/feed-by-ids?id=Pggi40VtT&amp;id=p0ZtbZRWV" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Biopass brings multi-modal biometric login to Linux</strong>: An open-source suite using 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>MySQL 9.6 moves foreign key enforcement to the SQL layer</strong>: This change ensures cascaded operations appear in the binary log, fixing long-standing data consistency issues for CDC and replication pipelines. <a href="https://app.daily.dev/posts/dp2YlOsXo" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>SQL Server on Linux matches Windows performance in benchmarks</strong>: Rigorous testing on RHEL 10.1 showed Linux marginally outperforming Windows in high-contention financial workloads due to more efficient lock queue draining. <a href="https://app.daily.dev/posts/xMDmAWHcC" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Context Mode reduces Claude Code token usage by 98 percent</strong>: An open-source MCP server that extends usable session time from 30 minutes to 3 hours by optimizing context window consumption. <a href="https://app.daily.dev/posts/rq8g08smm" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Qwen 3.5 Medium models released under Apache 2.0</strong>: The 35B parameter model uses a sparse MoE architecture and supports 1-million-token context windows on consumer GPUs. <a href="https://app.daily.dev/posts/0m0UpI5Bk" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>High severity ReDoS vulnerabilities patched in minimatch</strong>: Three flaws affecting the foundational npm package could stall the Node.js event loop; patches are backported to versions 3.x through 10.x. <a href="https://app.daily.dev/posts/EvvwDJaxE" 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,’ which is just the same LLM-assisted development with a more expensive-sounding name. It’s a classic industry pivot: when the ‘vibes’ start producing unmaintainable spaghetti, just add the word ‘engineering’ to make the chaos feel intentional. Sources like #1Hvh12XVS are already treating this as a fundamental shift rather than a desperate attempt to find a professional label for prompt engineering. <a href="https://app.daily.dev/posts/1Hvh12XVS" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Tinder’s camera roll chemistry is a privacy nightmare</strong>: Tinder is pitching ‘Chemistry’—an AI that scans your entire camera roll to find your best photos—as a matchmaking breakthrough. In reality, it’s ‘consent theatre’ that allows computer vision models to extract sensitive metadata and personal habits under the guise of profile optimization. As noted in ‘The Surveillance Crisis,’ this is just another case of a business model masquerading as a user feature to deepen data harvesting. <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’t be safe</strong>: A new ‘mathematical proof’ is making the rounds claiming AI cannot simultaneously be safe, trusted, and generally capable. While it sounds profound, it’s largely a theoretical exercise in goal-alignment doom-posting that ignores the practical, incremental safety engineering happening in the real world. It’s the academic version of a ‘checkmate’ tweet, designed more for engagement than for actually helping anyone build a more secure system. <a href="https://app.daily.dev/posts/i7Ojr8LXm" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
