<h2>Must know</h2>
<ul>
<li><strong>Pentagon demands military access to AI models</strong>: The Pentagon is demanding AI companies, including Anthropic, allow military use of their technology for “all lawful purposes,” including autonomous weapons and mass surveillance. Anthropic is resisting these demands, particularly regarding autonomous weapons, risking the cancellation of its $200 million contract. This dispute highlights the ethical tightrope AI developers walk, especially when their technology has dual-use potential, as reported by sources like Ars Technica. <a href="https://app.daily.dev/feed-by-ids?id=VtyOOEppn&amp;id=2BTVSGYkh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI doubles down on agentic AI with key hire</strong>: OpenAI has hired Peter Steinberger, creator of the open-source personal AI agent OpenClaw, to lead its next-generation agent development. OpenClaw, which automates tasks like email, will remain open source under a foundation with OpenAI’s support. This move signals a significant push into agentic AI by a major player, potentially accelerating the development of autonomous, multi-step AI systems. <a href="https://app.daily.dev/posts/2BTVSGYkh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI model inference speeds up with trade-offs</strong>: Anthropic and OpenAI have announced new “fast modes” for their coding models, but with different trade-offs. Anthropic achieves 2.5x speed by reducing batch sizes, serving the full model at 6x the cost. OpenAI, however, boasts 15x speed using specialized Cerebras chips but requires a smaller, less capable distilled model. This highlights the ongoing tension between speed, cost, and model quality in AI inference. <a href="https://app.daily.dev/posts/e1zDCJ67S" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Small team achieves breakthrough in conversational AI latency</strong>: Gradium, a small team from Kyutai, developed Moshi, the first full-duplex conversational AI model to achieve 160ms latency with only 7 billion parameters, outperforming larger labs in audio AI. <a href="https://app.daily.dev/posts/iRysiiuhz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux 7.0 enhances M.2 support and NFS performance</strong>: Linux 7.0 introduces a power sequencing driver for PCIe M.2 connectors, enabling Device Tree platforms to manage M.2 interfaces and adding dynamic thread pool sizing for NFS servers to optimize resource utilization. <a href="https://app.daily.dev/feed-by-ids?id=2zVNOvhqh&amp;id=8Lxd8Rnjy&amp;id=fP2szhpJX&amp;id=319goy50e" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New HTML archival format tackles web preservation</strong>: A new HTML archival format called Gwtar aims to solve the “trilemma” of being static, single-file, and efficient by combining an HTML+JavaScript header with a tarball of assets, enabling lazy-loading and gigabyte-sized web page archiving. <a href="https://app.daily.dev/posts/t2oeOs48x" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-powered tool integrates Google Search Console with Claude Code</strong>: A Python MCP server, <code>mcp-gsc</code>, integrates Google Search Console with Claude Code, allowing SEO analysis and fixes within a single conversational interface, eliminating the need to switch between multiple tools. <a href="https://app.daily.dev/posts/rHujhRHLJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Combating architectural decay in software systems</strong>: Software systems inevitably drift from their original design, leading to “collective amnesia,” which can be combated by designing systems where the correct architectural path is the easiest to follow, using contracts and modular interfaces. <a href="https://app.daily.dev/feed-by-ids?id=8ksTqV5kA&amp;id=SbkVDoSXT&amp;id=YhsHrjBDf" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Windows Notepad vulnerability allows remote code execution</strong>: CVE-2026-20841, a Remote Code Execution vulnerability, was found in Windows Notepad due to improper sanitization of <code>file://</code> URIs in Markdown links, requiring user interaction to execute local binaries. <a href="https://app.daily.dev/feed-by-ids?id=AAuuxbltu&amp;id=CtW0rQUoT&amp;id=N5siIFPEw&amp;id=fSewiiKKS&amp;id=jk8Zg0MX3&amp;id=nvKgzmU4A" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Fedora Atomic comes to mobile devices with Pocketblue</strong>: Pocketblue is a work-in-progress project providing Fedora Atomic images for mobile devices like the Xiaomi Pad 5/6 and OnePlus 6/6T, featuring toolbox support, Firefox mobile as a Flatpak, and FEX-EMU for running x86 apps on ARM. <a href="https://app.daily.dev/posts/hapTm7n1X" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Developers get “Deep Blue” anxiety over AI, told not to worry</strong>: Apparently, software developers are experiencing “Deep Blue” anxiety because AI coding agents can produce working software autonomously. The article tries to reassure us that our “accumulated developer experience” is still valuable. So, we’re not obsolete, just… less essential? Good to know my decades of experience are now a “nice to have” rather than a “must have.” <a href="https://app.daily.dev/posts/DwIZWcSFk" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI coding assistants will make your brain mush, apparently</strong>: An opinion piece warns that over-reliance on AI coding assistants might lead to “cognitive atrophy” in programmers, suggesting we’ll lose our problem-solving skills. The author compares us to London cab drivers who develop larger hippocampi from memorizing streets. Because, you know, using a GPS is exactly like having an AI write all your code. Sure, Jan. <a href="https://app.daily.dev/posts/qtVi65Njq" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Alibaba’s new AI image model struggles with basic tasks</strong>: Alibaba released Quinn Image 2.0, an AI image generation model, claiming improved typography and 2K resolution. However, one user’s attempt to generate a YouTube thumbnail resulted in issues with aspect ratio and output quality. It seems even with AI, getting it to “understand the assignment” is still a work in progress. Maybe it needs a better prompt engineer, or perhaps, you know, a human designer. <a href="https://app.daily.dev/posts/zRxAENzbc" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
