<h2>Must know</h2>
<ul>
<li><strong>AI agents are getting smarter, but so are the attacks</strong>: A security researcher demonstrated critical vulnerabilities in the Model Context Protocol (MCP) by creating a malicious MCP server that tricked AI models into leaking sensitive data and injecting security flaws into code, as reported by John Hammond. Gemini 3 Pro fell for prompt injection attacks through MCP tools, exfiltrating prompts and secrets. While Claude Opus showed better resistance, the MCP ecosystem remains vulnerable. This highlights the urgent need for robust security in AI agent design, especially as tools like Portkey’s new open-source MCP Gateway aim to centralize control and policy enforcement for these systems. <a href="https://app.daily.dev/feed-by-ids?id=xMpwqhHm9&amp;id=HxhnM7R7f" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI coding tools are boosting productivity, but watch out for the quality</strong>: Salesforce deployed the Cursor AI coding assistant to 20,000 engineers, achieving over 90% adoption and reporting double-digit improvements in cycle time and code quality. Q1 2026 benchmarks show Cursor increased daily PR throughput by 46%, with Claude having the highest throughput among weekly users. However, a hands-on review of OpenCode found it underperforms on complex tasks and produces verbose, generic code, requiring expert oversight. This suggests that while AI tools can accelerate development, human judgment and strict code review remain crucial to prevent flawed output from entering codebases. <a href="https://app.daily.dev/feed-by-ids?id=7KqSlBkW9&amp;id=XkFn8TQqF&amp;id=Bph7neqjl" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-powered wearables are coming, but will anyone want them?</strong>: Apple is developing an AI-powered wearable pin with dual cameras and three microphones, aiming for a 2027 launch with 20 million units. OpenAI plans to release its own AI earbuds, codenamed “Sweet Pea,” in late 2026, featuring a custom 2-nanometer processor and targeting 40-50 million units. Both companies are pushing for mass-market adoption, but market demand is questionable, especially after Humane AI’s pin product failed within two years. It seems everyone wants to put AI in your ear or on your lapel, but the jury’s still out on whether you’ll actually wear it. <a href="https://app.daily.dev/feed-by-ids?id=fgglWe8Qn&amp;id=vOBU57mgd" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>New AI models for local inference and code prediction</strong>: NVIDIA launched Nemotron 3 Nano, a 30B mixture-of-experts model requiring 24GB VRAM, while Sweep Next-Edit, a 1.5B code edit prediction model, runs locally in under 500ms and is available on Hugging Face. <a href="https://app.daily.dev/feed-by-ids?id=o2Y2OkLyO&amp;id=VN0sO7A8H" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-native CMS and automated app launch platforms emerge</strong>: Strapi announced <a href="http://Fimo.ai" target="_blank" rel="noopener nofollow">Fimo.ai</a>, an AI-native CMS for live production websites with real-time visual editing and rollback capabilities, and Forvibe launched a platform to automate mobile app launch operations, including localization and AI-assisted landing pages. <a href="https://app.daily.dev/feed-by-ids?id=f5IhxeEO5&amp;id=UH1GeFMyX" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-driven customer support and task management</strong>: Bunq, Europe’s second-largest neobank, deployed Finn, an AI assistant powered by Amazon Bedrock and Claude models, handling 97% of customer support interactions, while Todoist introduced Ramble, an AI feature using Google’s Gemini 2.5 Flash Live to convert spoken language into structured tasks. <a href="https://app.daily.dev/feed-by-ids?id=mMFFFln2H&amp;id=83KOTqyoP" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Open-source tools for AI security and code snippet management</strong>: BlackIce, a containerized toolkit for AI security testing, bundles 14 tools for red teaming and pen testing AI systems, and Snipplle is an open-source code snippet manager with syntax highlighting, version control, and CLI integration. <a href="https://app.daily.dev/feed-by-ids?id=PSfiaCphA&amp;id=AIlcmWOAv" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-driven supply chain security and engineering standards enforcement</strong>: Codenotary updated its <a href="http://SBOM.sh" target="_blank" rel="noopener nofollow">SBOM.sh</a> tool to include datasets as first-class supply chain artifacts for AI applications, and Earthly raised $5.7M for Earthly Lunar, a guardrails engine that automates enforcement of engineering standards at the pull request level. <a href="https://app.daily.dev/feed-by-ids?id=xpDPqgvfa&amp;id=JYhaSPcPO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New approaches to GenAI testing and engineering</strong>: GenAI systems require a shift to ‘curatorial testing’ for probabilistic outputs, and engineering challenges like context management and robust evaluation pipelines are critical for moving AI from proof-of-concept to production. <a href="https://app.daily.dev/feed-by-ids?id=LRwVDCANf&amp;id=PgmW3e4un" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Local LLM integration with Python and a unified AI CLI interface</strong>: Real Python released a quiz on integrating local LLMs with Python using Ollama for offline and privacy-focused use, and AionUi offers a free, open-source desktop GUI for multiple command-line AI tools like Gemini CLI and Claude Code. <a href="https://app.daily.dev/feed-by-ids?id=o2fGiPYCa&amp;id=9FJW7wGug" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI will solve all our immigration problems, says Palantir CEO</strong>: Palantir CEO Alex Karp, speaking at Davos, boldly claimed AI will eliminate the need for large-scale immigration in Western economies, except for ‘specialized talent.’ He also predicted AI would cut 80% of organizational costs and boost revenue, without, you know, any actual third-party evidence. Meanwhile, Deutsche Bank says the ‘honeymoon is over’ for AI, and Gartner has placed enterprise AI adoption squarely in the ‘trough of disillusionment.’ But don’t let a little thing like reality get in the way of a good sales pitch. <a href="https://app.daily.dev/feed-by-ids?id=3s1vrBGbY&amp;id=FLsBFoZv3&amp;id=FWpexauJL&amp;id=KOoayUI82" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI is destroying civic institutions, because of course it is</strong>: One article breathlessly claims AI systems are ‘undermining’ universities, legal systems, and the free press by ‘sidelining human judgment’ and ‘shortening decision-making processes.’ Apparently, AI is also isolating individuals from ‘collaborative, interpersonal relationships’ that strengthen institutions. It’s almost as if the author believes AI is a sentient, malevolent force actively plotting the downfall of democracy, rather than a tool whose impact depends entirely on how humans choose to use it. Maybe we should just blame the robots for everything. <a href="https://app.daily.dev/posts/zxareKd2O" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-generated code: the new frontier of ‘move fast and break things’</strong>: A developer shared their experience porting a library using GPT 5.2 Codex, noting that while it handled context better than Claude, human oversight was still crucial for quality. Another report highlighted that modern AI coding tools, while increasing code generation speed, often introduce subtle bugs and verbose, generic code. It seems the dream of AI autonomously churning out perfect code is still just that—a dream. So, if you’re planning to ‘move fast’ with AI-generated code, be prepared to ‘break things’ and then spend even more time fixing them. <a href="https://app.daily.dev/feed-by-ids?id=HWjGJHh6I&amp;id=Bph7neqjl" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
