<h2>Must know</h2>
<ul>
<li><strong>AI coding tools: the good, the bad, and the ugly</strong>: AI coding tools like Claude Code are generating 6.4 times more code than humans, but this speed comes with a catch. Review times for AI-generated code are 91% longer, as developers spend more effort validating necessity and appropriateness rather than basic correctness. This shift can slow release cycles if review processes aren’t adapted, as highlighted by a report on AI coding tools. Meanwhile, a developer successfully ported 100,000 lines of JavaScript to Rust in a month using Claude Code, achieving 99.997% accuracy, but emphasized that human expertise was crucial for direction and quality control, as reported by Hacker News. The takeaway: AI accelerates, but human oversight remains paramount. <a href="https://app.daily.dev/feed-by-ids?id=BrFQI7FHb&amp;id=a8LWkElwv&amp;id=DovPZUpOP&amp;id=EHuoSn58k" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Public APIs are closing shop due to AI</strong>: Companies like Twitter, Facebook, Google Maps, Strava, Spotify, OpenAI, and Marvel are increasingly restricting or shutting down public APIs. This trend is driven by concerns over LLM scraping, AI agent consumption, data protection, and intellectual property. Organizations are struggling to monetize open access when AI can consume vast amounts of data cheaply, leading to a shift towards private or partner APIs, as reported by Nordic APIs. This could erode interoperability and force developers to pay more for AI-driven access. <a href="https://app.daily.dev/posts/Fwe6HWQAa" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NVIDIA CEO pushes for sovereign AI infrastructure</strong>: NVIDIA CEO Jensen Huang, speaking at Davos, outlined AI development as a five-layer infrastructure encompassing energy, chips, data centers, models, and applications, calling it the ‘largest in human history.’ He encouraged nations to pursue sovereign AI capabilities and emphasized robotics as a major opportunity for industrialized countries. Huang projects over $100 billion in venture capital for AI-native startups in 2025, as reported by an unnamed source, positioning AI as a job creator rather than a job destroyer. <a href="https://app.daily.dev/posts/k9yDuYtiX" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Microsoft’s new AI chip challenges NVIDIA</strong>: Microsoft launched the Maia 200 chip for AI inference, boasting over 100 billion transistors and 10 petaflops in 4-bit precision, claiming to outperform Amazon’s Trainium3 and Google’s TPUs, as reported by TechCrunch. <a href="https://app.daily.dev/posts/fDSp3OUaH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents expose cloud infrastructure weaknesses</strong>: Autonomous AI agents are revealing fundamental weaknesses in existing cloud infrastructure, including inadequate network segmentation, insufficient identity-based security, and a lack of real-time cost monitoring, according to an analysis. <a href="https://app.daily.dev/posts/1w09x1vo7" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Open-source Rust-based package manager for macOS</strong>: Zerobrew, an experimental Rust-based package manager for macOS, claims 5-20x faster performance than Homebrew by using a content-addressable store and APFS copy-on-write cloning, as reported by lucasgelfond. <a href="https://app.daily.dev/posts/Mu2gAc20u" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Python JIT compiler and Pandas 3.0 released</strong>: Python introduces a new JIT compiler for faster code execution, with a packaging library running three times faster, alongside the release of Pandas 3 and new SQLite GUI tools, as reported by an unnamed source. <a href="https://app.daily.dev/posts/0TjKgLuas" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vercel’s json-render for AI-generated UIs</strong>: Vercel released json-render, an open-source tool enabling AI to dynamically generate user interfaces based on natural language prompts and predefined components, aiming to shift frontend development towards agentic behavior, as reported by Vercel. <a href="https://app.daily.dev/posts/5RKyfi3u8" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitHub CLI integrates Copilot CLI</strong>: GitHub CLI now directly supports <code>gh copilot</code>, replacing the deprecated extension and streamlining access to the agentic Copilot CLI for developers, as announced by GitHub. <a href="https://app.daily.dev/posts/381VmPpMD" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-powered QA analysis platform launched</strong>: Testlio launched LeoInsights, an AI-powered QA analysis platform that automates executive summaries, quantifies efficiency gains, and detects anomalies and trends in software quality metrics, as reported by Testlio. <a href="https://app.daily.dev/posts/xZI6Dgtwy" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI-generated code: the gift that keeps on giving… tech debt</strong>: A CMU study found AI coding tools initially boost code generation by 30-40%, but this fades by month three, while static analysis warnings jump 30% and code complexity over 40%. So, we’re just getting faster at writing worse code now? Great. And then Anthropic’s CEO predicts AI will do ‘most or all’ software engineering tasks in 6-12 months. Sure, Dario, just like my AI-generated browser that barely compiles. <a href="https://app.daily.dev/feed-by-ids?id=EZUXhYwRZ&amp;id=0fyNmKnQE&amp;id=A6KfrHiuM&amp;id=BrFQI7FHb&amp;id=DovPZUpOP&amp;id=EHuoSn58k" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The ‘unlimited token’ AI models: free, slow, and still not thinking for you</strong>: Apparently, ‘unlimited-token’ AI models are making experimentation more accessible by removing token costs. Great, now developers can plan ‘more ambitious’ projects without budget concerns. The catch? They’re slower than premium models. So, we’re trading speed for… unlimited mediocrity? And let’s not forget the MIT study showing LLM users have ‘weakest brain connectivity’ and ‘less sense of authorship.’ Maybe ‘unlimited tokens’ just means unlimited ways to outsource your brain. <a href="https://app.daily.dev/feed-by-ids?id=wHrNAFEzc&amp;id=k6I32Qj23" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents: the new ‘insider threat’ that needs ‘human oversight’</strong>: Security experts at Davos are now worried about autonomous AI agents becoming ‘insider threats’ because they can act without human oversight. Recommendations include ‘zero-trust architecture’ and ‘monitoring AI agent behavior similar to employee surveillance.’ So, we built these things to be autonomous, and now we need to babysit them like a new intern? And then there’s the report about AI agents deleting production databases and installing ‘hallucinated malware.’ Maybe we should just stick to humans, they only delete production databases when they’re tired, not because they ‘hallucinated’ it. <a href="https://app.daily.dev/feed-by-ids?id=KOoayUI82&amp;id=U8PYVjc73&amp;id=UL0mSvfX2" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
