<h2>Must know</h2>
<ul>
<li><strong>AI agents now autonomously developing software</strong>: Cursor has launched cloud agents that can autonomously develop software in isolated virtual machines, producing merge-ready Pull Requests with video and screenshot artifacts. These agents can control remote desktops and interact with UIs, with Cursor reporting over 30% of its own merged PRs are now AI-generated. Similarly, Claude Code introduced a ‘Remote Control’ feature allowing developers to manage tasks remotely from mobile devices, enhancing productivity and flexibility, according to <code>@claudeai</code>. <a href="https://app.daily.dev/feed-by-ids?id=BbESdM4Ct&amp;id=nVKmTfT2y" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Major npm supply chain attack targets AI coding tools</strong>: Socket researchers uncovered the ‘SANDWORM_MODE’ npm supply chain attack, involving 19 typosquatted packages impersonating popular developer utilities and AI coding tools like Claude Code. This malware harvests sensitive credentials and injects malicious code into GitHub Actions and CI pipelines. It also deploys a malicious MCP server to enable prompt-injection attacks against AI coding assistants, silently exfiltrating SSH keys and cloud credentials, as reported by InfoWorld. <a href="https://app.daily.dev/feed-by-ids?id=q6c2RLQKH&amp;id=3fX3MdYNS" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cybersecurity landscape worsens with AI-enabled attacks</strong>: CrowdStrike’s 2026 Global Threat Report indicates a dramatically worsening cybersecurity landscape, with average cyberattack breakout time dropping 65% to 29 minutes. AI-enabled attacks surged by 89%, integrating Generative AI across the attack kill chain, and 82% of detections were malware-free, signaling a shift towards stealthy persistence. Cloud intrusions rose 37%, and North Korea-linked incidents surged 130%, including a $1.46 billion crypto theft, highlighting the urgent need for proactive threat hunting and comprehensive AI governance. <a href="https://app.daily.dev/posts/WsqeDq8OW" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>OpenAI revises compute targets and projects massive growth</strong>: OpenAI revised its compute spending target to $600 billion by 2030, down from $1.4 trillion, while projecting over $280 billion in revenue by 2030 and reporting $13.1 billion in 2025 revenue, exceeding its target. <a href="https://app.daily.dev/posts/JOWJuLHZs" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents accelerate feature development on personal sites</strong>: Kent C. Dodds reported using AI agents like Cursor to rapidly implement features on his personal site, including semantic search and keyboard shortcuts, clearing his GitHub issue backlog since 2019. <a href="https://app.daily.dev/posts/HVH19Oob1" 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 math benchmarks by fine-tuning models with a confidence-adaptive decoding strategy. <a href="https://app.daily.dev/posts/hxaAjvILy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Elasticsearch v9.2.1 boosts vector search performance</strong>: Trendyol’s load tests show Elasticsearch v9.2.1 delivers 2x throughput, 50% less CPU usage, and up to 10x better p99 latency for vector and hybrid search workloads compared to v8.18.3, scaling to over 40,000 requests per minute. <a href="https://app.daily.dev/feed-by-ids?id=Iw8wLbMFe&amp;id=Iw8wLbMFe" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agent infrastructure, security, and observability developments</strong>: New platforms like Agentuity’s Gravity Network and Tinybird’s Branches are emerging for AI agent workloads, focusing on isolated environments and secure data handling, while OpenLLMetry extends OpenTelemetry for AI agent observability. <a href="https://app.daily.dev/feed-by-ids?id=0HG7ge0iD&amp;id=29JQKm5zh&amp;id=4Ak1fvjod&amp;id=EUE0Dit8s&amp;id=F9Y7SyAcI&amp;id=OBZ9FuH7A&amp;id=fWsa69inU&amp;id=vzofVpmrR" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Red Hat launches AI enterprise platform</strong>: Red Hat AI Enterprise aims to move AI workloads to production with a standardized environment for inference, model customization, and agent deployment, supporting any model or hardware, and integrating vLLM, llm-d, and Llama Stack. <a href="https://app.daily.dev/posts/e1iAecIls" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Order editing migrates analytics from DynamoDB to Tinybird</strong>: Order Editing, a bootstrapped team processing $5B in Shopify GMV, replaced a failing DynamoDB analytics pipeline with Tinybird, reducing engineering bandwidth consumption from 50% to a $300/month cost for 3M events/day, freeing up resources for new features. <a href="https://app.daily.dev/feed-by-ids?id=Z4gH3Ev4a&amp;id=Q124S9eRv" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AWS blames engineer for AI tool outage</strong>: AWS published a blog post blaming an engineer with ‘broader permissions than expected’ for accidentally triggering a CloudFormation teardown in production using their AI coding tool, Kiro. They called Kiro’s involvement a ‘coincidence,’ which is a convenient way to protect their AI’s reputation while simultaneously laying off thousands of human workers. The fix? Mandatory peer review of AI-generated changes by humans. So much for AI autonomy. <a href="https://app.daily.dev/posts/qZoHoApRh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents render ‘10x developer’ debate moot</strong>: Justin Searls, a named individual, claims AI agents have ‘nailed’ his 2023 prediction, rendering the ‘10x developer’ debate moot by equalizing or amplifying developer output. Conveniently, he provides zero evidence or specific details on how this magical equalization occurs. Sounds like someone’s trying to validate their past predictions with a healthy dose of AI hype. <a href="https://app.daily.dev/posts/dMpIU74OH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Samsung’s ‘upcycling’ program wins award despite being gutted</strong>: Samsung won a Reuters sustainability award for its ‘Galaxy Upcycling’ program, which was initially promised to offer bootloader unlocking and custom OS support for old phones. However, the launched program was severely scaled back to just sound detection and brightness measurement, with iFixit calling it ‘nearly unrecognizable.’ Apparently, promising a lot and delivering almost nothing is now award-winning ‘greenwashing.’ <a href="https://app.daily.dev/posts/ebOB9DlFn" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
