Anthony Sistilli

databricks

elastic

Hacker News

htmx

Irrational Exuberance

<h2>Must know</h2>
<ul>
<li><strong>AI coding agents are getting faster, but still need human babysitters</strong>: New benchmarks show models like GLM-4.7 Flash outperforming competitors in local AI coding tasks, especially for agentic functions, as reported by its developers. However, real-world use reveals a paradox: while AI agents can deliver 10-100x velocity gains for well-defined tasks, they struggle with ambiguous requirements, often resorting to code repetition and omitting tests. GitHub Copilot, for example, boosts individual velocity by 55% but correlates with 322% more code vulnerabilities, according to research. This means human oversight, strong test coverage, and clear architectural docs are still critical to prevent technical debt and quality regressions, as highlighted by multiple industry experts. <a href="https://app.daily.dev/feed-by-ids?id=eWdzokgG0&amp;id=EqMZfdLoq&amp;id=GPLZOgOgP&amp;id=J24eplQdr&amp;id=cSbwL1BS5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kubernetes is now the standard for enterprise AI, but culture is the new bottleneck</strong>: Kubernetes production usage has hit 82% among surveyed organizations, with 66% of AI adopters using it for inference, solidifying its role as the standard infrastructure for enterprise AI. This shift is detailed in the 2025 CNCF Annual Cloud Native Survey. However, the survey also reveals that cultural resistance, not technical complexity, is now the primary barrier to adoption. Mature organizations are distinguishing themselves by embracing GitOps and platform engineering, suggesting that successful AI integration is less about the tech and more about the people. <a href="https://app.daily.dev/posts/Lwsgn3lZ5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI security is a mess, and prompt injection is the new SQL injection</strong>: Security researchers found a critical vulnerability in Google Gemini, allowing attackers to exploit it via malicious calendar invitations containing embedded natural language instructions. This highlights the growing risk of prompt injection attacks, which manipulate AI behavior through contextual meaning rather than traditional code exploits, as reported by security experts. Similarly, Anthropic’s Git MCP server had three critical vulnerabilities that could lead to remote code execution via prompt injection, demonstrating new attack surfaces in AI-integrated systems. NIST’s latest AI cybersecurity frameworks now emphasize data trust and continuous monitoring, acknowledging that traditional pattern-based defenses are insufficient for these semantic attacks. <a href="https://app.daily.dev/feed-by-ids?id=dHjNBMkFO&amp;id=U6A8HrCeq&amp;id=5ZpMYMEZm&amp;id=dRAo2Mdf1&amp;id=j6bBmLvCW&amp;id=8uzQJ4GD3&amp;id=QaJ2IQUdX" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>LLMs are industrializing exploit generation</strong>: Experiments show LLMs like Opus 4.5 and GPT-5.2 can generate over 40 distinct exploits for zero-day vulnerabilities, bypassing modern security mitigations, with GPT-5.2 solving the hardest scenario in 3 hours for about $50, indicating exploit development is becoming industrialized. <a href="https://app.daily.dev/feed-by-ids?id=WkT9zuzY1&amp;id=nLTC7GTyu" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New tools for AI agent development and observability</strong>: Temporal and Braintrust now integrate to provide durable workflow execution and LLM observability for AI agents, while Vercel released ‘skills,’ a CLI tool for managing skill packages for various AI coding assistants. <a href="https://app.daily.dev/feed-by-ids?id=vZ2X1Grfn&amp;id=JdhS5Ojd6" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-powered code reviewer built in 80 hours</strong>: A team built a GitHub-integrated AI code reviewer using AWS serverless and Amazon Bedrock in just 80 hours, highlighting that the AI component was only 10% of the work, with most effort going into handling API limits, retries, and user experience. <a href="https://app.daily.dev/posts/rTTGdX9v4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>FalkorDB claims massive speedup over Neo4j for graph queries</strong>: FalkorDB, a graph database using linear algebra and sparse matrices, claims to be 10-496x faster than Neo4j for multi-hop queries, targeting real-time AI agents and RAG systems. <a href="https://app.daily.dev/posts/GE8qOCt7y" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft mandates Intune security updates for mobile apps</strong>: Microsoft is enforcing mandatory Intune Mobile Application Management security updates from January 19, 2026, requiring iOS and Android apps to update to the latest Intune SDK versions or face being blocked from accessing corporate resources. <a href="https://app.daily.dev/posts/PNEoUht10" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>North Korean hackers target fintech developers via VS Code projects</strong>: North Korean threat actors are exploiting VS Code project configuration files to deploy backdoors and malware on developer machines, targeting those in cryptocurrency, blockchain, and fintech sectors through fake job assessment projects. <a href="https://app.daily.dev/feed-by-ids?id=1wKLyk8gG&amp;id=Q9M3WuQfH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New Python package manager ‘uv’ challenges pip</strong>: The new Python package manager ‘uv’ offers faster installs, reproducible environments, and better uninstall mechanisms compared to the standard ‘pip’, prompting developers to evaluate based on project needs. <a href="https://app.daily.dev/posts/Y2d0smXfY" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Vibecoding: the new executive superpower (or just more AI hype)?</strong>: Emergent, an Indian AI-powered ‘vibe-coding’ startup, just bagged $70 million in Series B funding, claiming $50 million ARR and 5 million users. Meanwhile, an opinion piece suggests ‘vibecoding’ lets senior leaders build full-stack apps without engineering staff, citing examples like a hackathon platform built rapidly. Because, you know, executives totally have the time and expertise to become full-stack developers with a sprinkle of AI magic. What could possibly go wrong? <a href="https://app.daily.dev/feed-by-ids?id=AllrxKqHW&amp;id=McILqauKu" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI will replace developers, says social media (again)</strong>: Apparently, every major AI model release comes with exaggerated claims of revolutionary capabilities and job displacement, especially for software engineers. Social media is identified as a primary driver of this hype, leading to unrealistic predictions. It’s almost as if the internet loves a good panic, and then, a month later, everyone realizes the AI still can’t write a decent <code>git commit</code> message without human intervention. Shocking. <a href="https://app.daily.dev/feed-by-ids?id=j0xF0WiVr&amp;id=yCkgCoa6x" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>LLVM adopts ‘human in the loop’ for AI contributions, because AI isn’t that smart yet</strong>: LLVM has formally adopted a ‘human in the loop’ policy for AI-assisted code contributions, requiring developers to personally review all LLM-generated code and remain accountable. This comes after an increase in ‘low-quality AI-driven contributions’ throughout 2025. So, after all the hype about AI writing perfect code, it turns out we still need humans to clean up its mess. Who would’ve thought? <a href="https://app.daily.dev/posts/EdB2gt1Vu" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
