<h2>Must know</h2>
<ul>
<li><strong>AI is industrializing exploit generation</strong>: LLMs like Opus 4.5 and GPT-5.2 are now capable of generating over 40 distinct exploits for zero-day vulnerabilities, even bypassing modern security mitigations like ASLR and CFI. According to Hacker News, GPT-5.2 solved the hardest scenario in just 3 hours for about $50, shifting the bottleneck from human expertise to LLM token throughput. This means offensive cybersecurity is rapidly industrializing, enabling threat actors to scale and automate complex cyber operations at an unprecedented pace. <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>Kubernetes is now the standard for enterprise AI</strong>: The 2025 CNCF Annual Cloud Native Survey reveals Kubernetes production usage has hit 82%, with 66% of AI adopters using it for inference. It’s now the go-to infrastructure for enterprise AI workloads. Interestingly, the biggest hurdle isn’t technical complexity anymore, but cultural resistance within organizations, as highlighted by the CNCF survey. This means your robust, performant systems will need to navigate human factors more than ever. <a href="https://app.daily.dev/posts/Lwsgn3lZ5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic’s Git MCP server had critical RCE vulnerabilities</strong>: Anthropic’s Git MCP server was found to have three critical vulnerabilities (CVE-2025-68145, -68144, -68143) that could lead to remote code execution via prompt injection. These flaws, discovered and patched in late 2025, allowed attackers to manipulate files and inject malicious inputs into AI agent workflows. This incident underscores the unexpected security risks when chaining AI agent tools, especially those with code execution or file access capabilities, as reported by security researchers. <a href="https://app.daily.dev/posts/U6A8HrCeq" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>AI coding agents: velocity vs. reality</strong>: AI coding agents like Claude Code can deliver 10-100x velocity gains for well-defined tasks, but struggle with ambiguous requirements, often resorting to code repetition and omitting tests, requiring significant human oversight. <a href="https://app.daily.dev/feed-by-ids?id=EqMZfdLoq&amp;id=GPLZOgOgP&amp;id=J24eplQdr&amp;id=cSbwL1BS5&amp;id=zWcZCkr3l&amp;id=NR4LrQXl3" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New AI models for local coding and persona stabilization</strong>: GLM-4.7 Flash, a 30B-A3B mixture-of-experts model, is outperforming competitors in local AI coding and agentic benchmarks, while Anthropic researchers are developing ‘activation capping’ to stabilize LLM personas and prevent harmful outputs. <a href="https://app.daily.dev/feed-by-ids?id=eWdzokgG0&amp;id=Jt8b0RH1S" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft mandates Intune security updates</strong>: From January 19, 2026, Microsoft will enforce mandatory Intune Mobile Application Management security updates, requiring iOS and Android apps to use the latest Intune SDK versions or face being blocked from corporate resources. <a href="https://app.daily.dev/posts/PNEoUht10" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI security: prompt injection and framework vulnerabilities</strong>: Google Gemini was vulnerable to prompt injection via calendar invites, and critical flaws in the Chainlit chatbot framework allowed arbitrary file reading and SSRF, highlighting the need for intent-based and semantic security defenses in LLM systems. <a href="https://app.daily.dev/feed-by-ids?id=dHjNBMkFO&amp;id=5ZpMYMEZm&amp;id=dRAo2Mdf1&amp;id=j6bBmLvCW" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>A2A Java SDK 1.0.0.Alpha1 released for AI agent communication</strong>: The A2A Java SDK 1.0.0.Alpha1 is out, aligning with the upcoming Agent2Agent Protocol 1.0, bringing standardized communication, Java records, Protocol Buffers, and enhanced API documentation for AI agents. <a href="https://app.daily.dev/feed-by-ids?id=LmGT7Q5vx&amp;id=PMa93QJZs&amp;id=ZcWzztsx0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Next.js 16 brings Turbo Builds and AI debugging</strong>: Next.js 16 stabilizes Turbopack for faster builds, introduces explicit Cache Components, and integrates AI-assisted debugging via Model Context Protocol (MCP) in DevTools, allowing AI agents to suggest fixes. <a href="https://app.daily.dev/posts/91xiEmcVe" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New PostgreSQL backup tool and TigerBeetle benchmarks</strong>: pgmoneta 0.20.0, an open-source PostgreSQL backup solution, now features redesigned repository locking and S3 enhancements, while TigerBeetle, a financial database, demonstrated 2.8x to 6.6x faster performance than PostgreSQL for double-entry bookkeeping. <a href="https://app.daily.dev/feed-by-ids?id=DUFN7D8Ly&amp;id=GZhKq1ONe" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI-powered ‘vibecoding’ startup raises $70M, because why not?</strong>: Emergent, an Indian AI-powered ‘vibecoding’ startup, just bagged $70 million in Series B funding, valuing it at $300 million, claiming $50 million ARR after only seven months. Apparently, executives are now being told to ‘vibecode’ to build full-stack apps without needing actual coding expertise. Because nothing says ‘robust, performant, and maintainable systems’ like letting senior leaders ‘vibe’ their way to production code. <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 eliminate software engineers, says social media (again)</strong>: Another day, another round of social media hype claiming generative AI will make software engineers obsolete. Analysts are warning that this cycle of exaggerated claims, followed by disappointment, is routine with every major AI model launch. So, while the internet screams about job displacement, remember that actual engineers are still here, dealing with the limitations and drawbacks that the hype conveniently ignores. <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>PwC says AI isn’t delivering ROI, but don’t let that stop the hype train</strong>: A PwC survey found that 56% of CEOs saw no revenue increase or cost decrease from their AI spending, and only 12% saw both. Yet, other reports are still touting 25% productivity gains and 45% processing time reductions from AI-driven ERP. It seems the ‘agentic AI’ hype cycle is in full swing, with a clear disconnect between what’s being promised and what’s actually being delivered in the real world. <a href="https://app.daily.dev/feed-by-ids?id=Kov24DV15&amp;id=ogb2Wd6xs&amp;id=rEdiBiFbV" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
