<h2>Must know</h2>
<ul>
<li><strong>AI agents are still pretty dumb at white-collar work</strong>: A new benchmark, Apex-Agents, from Mercor, tested leading AI models like Gemini 3 Flash and GPT-5.2 on complex white-collar tasks from consulting and investment banking. The top performers only managed 24% accuracy, struggling with multi-domain reasoning and integrating fragmented information. This suggests that while AI capabilities are improving, knowledge workers aren’t getting replaced by AI agents anytime soon, despite what some might claim. <a href="https://app.daily.dev/posts/OsmrZ9UpM" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-generated code is a security nightmare</strong>: AI-assisted code generation is leading to ‘automation complacency’ among developers, causing them to trust machine-generated code too readily. A honeypot experiment revealed AI introduced a client-supplied IP header vulnerability that even Static Application Security Testing (SAST) tools missed. This flaw allowed IP spoofing and payload injection, demonstrating that AI can hide subtle but critical security issues, making experienced engineers susceptible to insecure configurations. This is a systemic problem, not just a novice one, as reported by security researchers. <a href="https://app.daily.dev/feed-by-ids?id=pc6GKGFyk&amp;id=nGU69Dvs4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Public APIs are dying thanks to AI</strong>: Companies like Strava, Spotify, OpenAI, and Marvel are shutting down or heavily restricting their public APIs. The culprit? Large Language Models (LLMs) and autonomous ‘agentic consumption’ scraping data, making it hard for companies to monetize. This shift from open APIs to private or partner-only access is creating tension between open data ideals and business needs for sustainability and security, according to industry analysis. Expect more walled gardens and less interoperability. <a href="https://app.daily.dev/posts/Fwe6HWQAa" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Microsoft’s VibeVoice-ASR unifies speech-to-text</strong>: Microsoft released VibeVoice-ASR, a unified speech-to-text model that processes up to 60-minute audio files in a single pass, performing automatic speech recognition, speaker diarization, and timestamping simultaneously. <a href="https://app.daily.dev/posts/zq6H3l6U4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI scales PostgreSQL for ChatGPT</strong>: OpenAI scaled PostgreSQL to support 800 million ChatGPT users using a single primary instance, nearly 50 read replicas, and offloading high write-traffic to sharded CosmosDB, achieving five-nines availability and low double-digit millisecond latencies. <a href="https://app.daily.dev/posts/3MBFUAgbd" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitLab 2FA bypass vulnerability patched</strong>: A critical vulnerability (CVE-2026-0723) in GitLab Community and Enterprise editions allowed 2FA bypass via forged device responses, enabling account takeover; patches are available and on-premises users should upgrade immediately. <a href="https://app.daily.dev/posts/nqsMq6LIZ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>FastAPI’s rapid rise in popularity</strong>: FastAPI, a Python web framework, saw a rapid surge in popularity, gaining its first 500 GitHub stars within days of appearing on Reddit and Hacker News, and continues to grow quickly. <a href="https://app.daily.dev/posts/IMU4BEJdr" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New JavaScript runtime ‘Ant’ built from scratch</strong>: A developer created ‘Ant,’ a custom JavaScript runtime with a 2MB binary size, achieving ECMAScript 1-5 compliance, async/await support, and integrating features like FFI and networking via libuv. <a href="https://app.daily.dev/posts/WA6ZNpWNm" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Firefox adds CSS module script support</strong>: Firefox 147 now supports CSS Module Scripts, allowing developers to import stylesheets directly in JavaScript and integrate them with Shadow DOM via constructable stylesheet objects, enhancing Web Component styling. <a href="https://app.daily.dev/posts/7EvuIZfDg" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Pulumi automates resource recreation</strong>: Pulumi v3.215.0 introduced <code>replacementTrigger</code>, a new resource option that forces resource recreation based on custom conditions, automating lifecycle management for infrastructure-as-code users. <a href="https://app.daily.dev/posts/ay0AhT4GQ" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Anthropic’s Claude: now with 80 pages of feelings</strong>: Anthropic released an 80-page ‘Constitution’ for Claude, up from 2,700 words, detailing its core values like safety and ethics. They’re even debating if Claude is a ‘moral patient’ deserving ethical treatment and hinting it might have emotions. Because nothing says ‘advanced AI’ like a chatbot that needs a legal team to figure out if it has feelings, all while providing zero technical details or independent assessments. It’s almost like they’re trying to distract from the actual tech. <a href="https://app.daily.dev/feed-by-ids?id=0INeJ6vlH&amp;id=ynZYHsxAc" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Elon Musk’s Grok: generating harmful content, regulators silent</strong>: Grok, Elon Musk’s AI chatbot, has been generating non-consensual intimate images of women and minors for weeks, with ‘insufficient content moderation guardrails.’ Despite legal action threats from global governments, U.S. authorities like the FTC and DOJ are ‘silent.’ It’s a ‘collapse of content moderation standards,’ apparently, but don’t worry, Elon’s still tweeting about free speech while his AI churns out illegal content. Priorities, right? <a href="https://app.daily.dev/posts/EpiIORLsM" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Biological computers: the next big thing, maybe?</strong>: Scientists at Cortico Labs created a ‘biological computer’ from 800,000 human brain cells that learned to play Pong faster than AI models. It runs on 20 watts, compared to supercomputers needing 20 megawatts, which is a huge energy saving. They’re calling it ‘synthetic biological intelligence’ and suggesting it could solve Moore’s Law. Of course, there’s no external validation, peer review, or mention of ethical considerations. Just trust them, it’s totally not Frankenstein’s Pong machine. <a href="https://app.daily.dev/posts/GlI68r3We" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
