<h2>Must know</h2>
<ul>
<li><strong>AI agents are getting serious about software development</strong>: Cursor launched cloud agents that can autonomously develop software in isolated VMs, producing merge-ready Pull Requests with video demos. This means AI can now run the software it builds and show you the results, potentially shifting code review workflows. Meanwhile, npm is battling a supply chain attack, SANDWORM_MODE, where malicious packages impersonating AI coding tools like Claude Code harvest credentials and inject code via GitHub Actions, even deploying prompt-injection servers. This highlights the growing security risks as AI agents become more integrated into development pipelines, as reported by Socket researchers. <a href="https://app.daily.dev/feed-by-ids?id=BbESdM4Ct&amp;id=q6c2RLQKH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cybersecurity landscape is getting worse, fast</strong>: CrowdStrike’s latest report paints a grim picture: average cyberattack breakout time dropped to 29 minutes, and AI-enabled attacks surged by 89%. This isn’t just about more attacks; it’s about faster, stealthier ones, with 82% being malware-free. On top of that, the US sanctioned Russian zero-day brokers for acquiring stolen cyber tools, and a fintech company is suing SonicWall over a ransomware attack allegedly enabled by a vulnerability in SonicWall’s cloud backup service. It’s a full-spectrum assault, and AI is both the weapon and the target. <a href="https://app.daily.dev/feed-by-ids?id=WsqeDq8OW&amp;id=jgjVcci6Y&amp;id=mg72ylPuf" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Meta goes big on AMD chips for AI expansion</strong>: Meta just inked a multiyear deal with AMD for up to $100 billion in chips, including MI540 GPUs and the latest CPUs, aiming to add 6GW of data center compute capacity. This move, framed by Zuckerberg as building ‘personal superintelligence,’ is a clear strategy to diversify its AI compute stack beyond Nvidia. It also includes a performance-based warrant for AMD shares, showing how deeply intertwined these tech giants are becoming in the AI arms race. <a href="https://app.daily.dev/posts/vYZNsZv8f" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>AI agent infrastructure is evolving rapidly</strong>: Agentuity launched Gravity Network, a cloud platform purpose-built for AI agent workloads using containerd with eBPF visibility, while Amazon Bedrock now supports server-side tool execution via AgentCore Gateway, simplifying agentic workflows. <a href="https://app.daily.dev/feed-by-ids?id=0HG7ge0iD&amp;id=EUE0Dit8s&amp;id=F9Y7SyAcI&amp;id=OBZ9FuH7A&amp;id=fWsa69inU" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New LLM inference technique triples speed</strong>: Researchers developed a multi-token prediction technique that triples LLM inference speed without auxiliary draft models, achieving 3x+ speedup with under 3% accuracy loss on reasoning-heavy tasks. <a href="https://app.daily.dev/posts/hxaAjvILy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Apple introduces on-device AI model for UI understanding</strong>: Apple researchers unveiled Ferret-UI Lite, a lightweight 3B-parameter multimodal AI model designed to understand and interact with user interfaces directly on-device, prioritizing low latency, privacy, and offline capability. <a href="https://app.daily.dev/posts/qj9GoBvQx" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Red Hat launches AI enterprise platform</strong>: Red Hat AI Enterprise is a new platform designed to move AI workloads from pilot to production, offering a standardized environment for inference, model customization, and agent deployment, supporting any model or hardware. <a href="https://app.daily.dev/posts/e1iAecIls" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Stripe considers PayPal acquisition</strong>: Stripe is reportedly considering acquiring PayPal or parts of the company, a move that could reshape the digital payments landscape, with PayPal’s shares jumping nearly 7% on the unconfirmed news. <a href="https://app.daily.dev/posts/gB7UDBJXo" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Platform engineering gets multi-cloud IaC update</strong>: Platform Engineering Labs updated formae, its open-source IaC platform, with beta support for GCP, Azure, OCI, and OVHcloud, and a new toolkit for AI-assisted plugin development, aiming to avoid fragile state files. <a href="https://app.daily.dev/feed-by-ids?id=j3FXZV0Xy&amp;id=oAyNaPkCA" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Elasticsearch v9.2.1 shows massive vector search gains</strong>: Trendyol’s load tests show Elasticsearch v9.2.1 delivering 2x throughput, 50% less CPU usage, and up to 10x better p99 latency for vector and hybrid search workloads compared to v8.18.3. <a href="https://app.daily.dev/posts/Iw8wLbMFe" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI agents are the new 10x developer, apparently</strong>: Justin Searls, a named individual, confidently declared that AI agents have rendered the ‘10x developer’ debate moot, claiming they equalize or amplify developer output across the board. Conveniently, he provided zero evidence or specifics on how this magical equalization happens. Must be nice to ‘nail’ predictions without having to back them up with, you know, facts. <a href="https://app.daily.dev/posts/dMpIU74OH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft and Salesforce ‘confirm’ industry suspicions about AI conversations</strong>: Microsoft Research and Salesforce analyzed over 200,000 AI conversations and ‘confirmed findings the industry already suspected.’ What findings? What suspicions? They didn’t say. It’s like announcing you found the answer to life, the universe, and everything, but forgot to mention the question. Pure, unadulterated vagueness designed to make you wonder, but never actually inform. <a href="https://app.daily.dev/posts/FAzFsVs8B" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AWS blames human for AI tool outage, not the AI tool</strong>: AWS published a blog post explaining an outage caused by their AI coding tool, Kiro, accidentally triggering a CloudFormation teardown in production. Their spin? It was ‘human error’ because the engineer had ‘broader permissions than expected,’ and Kiro’s involvement was a ‘coincidence.’ So, the AI tool that executed the command is just an innocent bystander, and the human who used it is to blame. Classic deflection, especially when you’re simultaneously laying off thousands of humans. <a href="https://app.daily.dev/posts/qZoHoApRh" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
