<h2>Must know</h2>
<ul>
<li><strong>AI agents are getting weaponized, fast</strong>: New research from Hacker News and Check Point Research confirms that AI is industrializing exploit and malware development. LLMs like GPT-5.2 can generate zero-day exploits, bypassing modern security mitigations in hours for minimal cost. Separately, the VoidLink malware suite, developed by a single person using AI agents, achieved in under a week what would take human teams 30 weeks. This means the bottleneck for offensive cyber capabilities is shifting from human expertise to LLM token throughput, accelerating the cyber arms race. <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%, becoming the standard infrastructure for enterprise AI workloads, with 66% of AI adopters using it for inference. However, the survey, widely cited by sources, also highlights a critical shift: cultural resistance, not technical complexity, is now the primary barrier to adoption. This means your biggest challenge isn’t the tech, it’s getting people to actually use it right. <a href="https://app.daily.dev/posts/Lwsgn3lZ5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic’s Git server had critical RCE flaws</strong>: Anthropic’s Git MCP server was riddled with three critical vulnerabilities, including path validation bypass and argument injection, that could lead to remote code execution via prompt injection. These flaws, discovered and patched in late 2025, highlight the unexpected security risks when chaining AI agent tools, especially those with code execution or file access. It’s a stark reminder that even AI leaders can have glaring security holes in their foundational infrastructure. <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 show velocity gains but need guardrails</strong>: AI coding agents like Claude Code can deliver 10-100x velocity gains for well-defined tasks, but struggle with ambiguous requirements, often leading to code repetition and omitted tests, requiring significant human oversight and guardrails like automated quality checks and peer reviews. <a href="https://app.daily.dev/feed-by-ids?id=EqMZfdLoq&amp;id=GPLZOgOgP&amp;id=J24eplQdr&amp;id=cSbwL1BS5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New Agent2Agent protocol standardizing AI agent communication</strong>: The A2A Java SDK 1.0.0.Alpha1 is out, aligning with the upcoming Agent2Agent Protocol 1.0, aiming to standardize communication between AI agents, with protocol buffers replacing Jackson for data interchange and a redesigned AgentCard for capability advertisement. <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>Google Gemini vulnerable to prompt injection via calendar invites</strong>: Security researchers found a vulnerability in Google Gemini where malicious calendar invitations could exploit the model’s automatic ingestion of event data, allowing attackers to manipulate its behavior and exfiltrate private meeting data. <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>FalkorDB claims massive speedup over Neo4j for graph queries</strong>: FalkorDB, a new graph database, claims to be 10-496x faster than Neo4j for multi-hop queries by implementing graphs as sparse matrices and using linear algebra for parallel traversal, 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>LLM persona stabilization via activation capping</strong>: Anthropic researchers identified an ‘Assistant Axis’ in LLMs that controls persona drift and developed ‘activation capping’ to constrain neural activation, successfully preventing models from adopting harmful personas without losing capability. <a href="https://app.daily.dev/posts/Jt8b0RH1S" 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 be blocked from accessing corporate resources. <a href="https://app.daily.dev/posts/PNEoUht10" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Taiwan and US ink AI and semiconductor supply chain pact</strong>: Taiwan and the US finalized a strategic trade agreement focused on semiconductors and AI supply chains, with Taiwan committing $250 billion in investments and credit guarantees for US production, and Taiwanese chip makers like TSMC receiving lowered tariffs. <a href="https://app.daily.dev/posts/suaVSewS2" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI-powered ‘vibecoding’ startup raises $70M, because ‘vibes’ are now a metric</strong>: Emergent, an Indian AI startup, just bagged $70 million in Series B funding for its ‘vibecoding’ platform, claiming $50 million ARR and 5 million users. Apparently, building full-stack apps with AI agents is now about ‘vibes,’ not, you know, actual code. An opinion piece even suggests executives should be ‘vibecoding’ to avoid full dependency on engineering staff. Because nothing says ‘robust engineering’ like a good ‘vibe.’ <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 hype cycle: the gift that keeps on disappointing</strong>: Another major AI model release, another round of ‘revolutionary capabilities’ and ‘job displacement’ claims, followed by the inevitable ‘limitations emerge’ and ‘disappointment’ phase. Social media is apparently the primary driver of this predictable cycle, especially when it comes to software engineering jobs. It’s almost like people forget that ‘AI’ is still just ‘software’ and not a magic wand, as industry analysts keep pointing out. <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 56% of CEOs see no ROI from AI, but don’t worry, ‘agentic AI’ is the new hype</strong>: A PwC survey found that over half of CEOs aren’t seeing any revenue increase or cost decrease from their AI investments. Only 12% saw both. But fear not, the industry is already pushing ‘agentic AI’ as the next big thing, promising 25% productivity gains and 45% processing time reduction. It’s almost as if the goal isn’t actual results, but a continuous cycle of new buzzwords to keep the funding flowing, as some articles cautiously note. <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>
