<h2>Must know</h2>
<ul>
<li><strong>AI coding agents are exposing critical infrastructure weaknesses</strong>: Agentic AI systems, operating continuously at machine speed, are revealing inadequacies in cloud infrastructure, particularly in networking segmentation, identity-based security, and cost controls. According to InfoWorld, organizations need fine-grained network policies, strict identity governance, and real-time cost monitoring to prevent agents from causing outages or runaway expenses. This shift demands strong discipline in governance, security, and architecture, treating agentic AI as a fundamental operational change rather than just another feature. <a href="https://app.daily.dev/posts/1w09x1vo7" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft provided BitLocker recovery keys to the FBI</strong>: Microsoft gave the FBI BitLocker recovery keys to unlock laptops in a fraud investigation, marking the first publicly reported instance of such cooperation. By default, BitLocker recovery keys are uploaded to Microsoft’s cloud if users sign in with a Microsoft account, a practice security experts warn creates privacy risks. While Microsoft receives about 20 such requests annually, this incident highlights the trade-off between data recoverability and user privacy, especially given past breaches of Microsoft’s cloud infrastructure. <a href="https://app.daily.dev/posts/6Aq315WBk" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Open-source maintainers are restricting contributions due to AI-generated code</strong>: Open-source projects like TL Draw, Curl, and Zig are closing or restricting external contributions because of a surge in low-quality, AI-generated pull requests. Maintainers report that this influx of poorly structured code is overwhelming to review, introduces significant technical debt, and disrupts project communities. The ease of access to AI coding tools has led to contributions from inexperienced individuals who don’t understand project architecture or community standards, draining maintainer resources and leading to the term ‘AI slop’. <a href="https://app.daily.dev/posts/9Rmqfgu76" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Context transfer is the biggest bottleneck for AI coding tools</strong>: The primary barrier for AI coding tools in 2026 is context transfer—the difficulty in conveying engineers’ tacit knowledge to AI systems, leading to manual reverse-engineering of AI-generated code for maintenance. <a href="https://app.daily.dev/posts/DczaKkmGw" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitHub Copilot CLI integrates organizational context</strong>: GitHub Copilot CLI now uses Work IQ to access organizational data from documents, emails, and SharePoint, allowing AI agents to compare code against company specs and reconstruct undocumented architecture from meeting transcripts. <a href="https://app.daily.dev/posts/AnhMtsEYf" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>JetBrains IDEs get native OpenAI Codex integration</strong>: JetBrains has integrated OpenAI Codex (GPT-5.2-Codex) directly into its major IDEs, offering AI coding assistance for planning, writing, testing, and refactoring code with user-controlled autonomy levels. <a href="https://app.daily.dev/posts/uAvdGTErT" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vercel releases json-render for AI-driven UI generation</strong>: Vercel launched json-render, an open-source tool that allows developers to specify UI components and actions, enabling AI to generate JSON representations of interfaces from natural language descriptions, bypassing traditional manual UI development. <a href="https://app.daily.dev/posts/5RKyfi3u8" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Temporal API launches in Chrome 144</strong>: The Temporal API is now available in Chrome 144, replacing JavaScript’s problematic Date API to address timezone handling and date calculation issues, though browser APIs like Geolocation still raise privacy concerns. <a href="https://app.daily.dev/posts/4f3BuYROb" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Rust 1.93 updates musl library and improves networking</strong>: Rust 1.93 upgraded its bundled musl library to 1.2.5, bringing significant DNS resolver improvements for static Linux binaries, especially for large DNS records and interactions with recursive name servers. <a href="https://app.daily.dev/posts/wF6DW7Hp4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AWS launches EC2 M4 Max Mac instances</strong>: AWS now offers M4 Max Mac instances for EC2, utilizing Apple M4 Max chips to provide up to 25% better application build performance for Apple platform apps, available in US East and US West regions. <a href="https://app.daily.dev/posts/DEbhYteDE" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Anthropic CEO predicts AI will replace software engineers in 6-12 months</strong>: Anthropic CEO Dario Amodei, at the World Economic Forum, boldly declared that AI models could handle most software engineering tasks within 6-12 months, relegating humans to mere editors. He admits his own engineers already use AI to write code, but also concedes his timeline is ‘uncertain.’ So, either we’re all out of a job by next Christmas, or this is just another CEO trying to drum up some hype at Davos. <a href="https://app.daily.dev/posts/0fyNmKnQE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI coding tools: more code, more problems</strong>: The Go developer survey reveals that while 53% of developers use AI coding tools daily, only 13% are ‘very satisfied,’ citing non-functional code generation and poor quality as key complaints. Meanwhile, a developer comparing traditional coding to AI-assisted low-code found that while low-code delivered faster, it created significant knowledge gaps and made debugging harder. It seems AI can churn out code, but it’s still a long way from churning out <em>good</em> code, or even code that works. <a href="https://app.daily.dev/feed-by-ids?id=yyZ0pKuL1&amp;id=f1lHyoYFM" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Prompt engineering is now ‘infrastructure’ for scalable AI development</strong>: Jaime, a ‘knowledgeable internal actor,’ suggests prompt engineering should be treated as ‘version-controlled infrastructure’ using ‘Agent Skills’ to maintain code integrity. This sounds suspiciously like we’re just adding more layers of abstraction and process to manage the mess AI creates, rather than actually solving the problem of AI-generated ‘tribal knowledge.’ Because nothing says ‘scalable’ like version-controlling your prompts. <a href="https://app.daily.dev/posts/m5IRqcNSP" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
