<h2>Must know</h2>
<ul>
<li><strong>Iranian cyber offensive weaponizes identity management</strong>: The Iranian-linked group Handala has escalated from simple wipers to identity weaponization, reportedly compromising Stryker’s Microsoft environment to issue mass remote-wipe commands via Intune. Flashpoint and Check Point Research indicate this ‘living-off-the-land’ tactic bypassed traditional EDR by using legitimate administrative tools to destroy tens of thousands of devices. Security teams are urged to enforce multi-admin approval for high-impact MDM operations as nation-state actors shift focus toward civilian infrastructure. <a href="https://app.daily.dev/feed-by-ids?id=RNP1WzQpA&amp;id=UIwE7Q6hx&amp;id=vOzeyXRGg&amp;id=wSrqqR8V2" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NVIDIA and AWS overhaul AI data and inference layers</strong>: NVIDIA unveiled its AI Grid architecture at GTC 2026, a distributed infrastructure design that embeds Blackwell-based accelerated computing within telecom POPs to slash edge inference latency. Simultaneously, AWS launched account-regional namespaces for S3 to eliminate global bucket naming conflicts and introduced S3 Vectors for native RAG support. These moves, alongside Google Cloud’s multi-cluster GKE Inference Gateway, signal a shift toward treating the data layer as the primary bottleneck for autonomous agent performance. <a href="https://app.daily.dev/feed-by-ids?id=8cW6kqWAo&amp;id=LloNHqveA&amp;id=I83jnVlzS&amp;id=79zWldGPG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kubernetes 1.36 retires Ingress Nginx for Gateway API</strong>: The upcoming Kubernetes 1.36 release marks the formal retirement of Ingress-Nginx in favor of the more expressive Gateway API, which provides superior multi-tenant networking and traffic splitting. Security is also being tightened with graduated OCI artifact mounting for decoupling large ML models from container images and strengthened Linux User Namespaces for better container isolation. Platform engineers should prepare for manifest-based Admission Control policies moving to static files on the control plane disk to close startup-phase security gaps. <a href="https://app.daily.dev/posts/u9P6lGf7L" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Linux 7.0 rc4 resolves critical scheduler hangs</strong>: The latest release candidate fixes race conditions in the mm/cid rewrite that caused hangs during concurrent forks on large systems. <a href="https://app.daily.dev/feed-by-ids?id=I6PfBBlwh&amp;id=yBNZ9N91D" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>PyTorch achieves 30 percent speedup with MXFP8</strong>: Researchers successfully used dynamic MXFP8 quantization on GB200 clusters to accelerate Mixture-of-Experts training without losing convergence. <a href="https://app.daily.dev/posts/agaSmbqCx" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Jepsen exposes durability bugs in MariaDB Galera</strong>: Analysis revealed that even with recommended configurations, MariaDB Galera clusters can silently lose committed transactions during network partitions. <a href="https://app.daily.dev/posts/rOntuN0uj" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NVIDIA NeMo Retriever takes top spot on ViDoRe</strong>: The new agentic retrieval pipeline replaces MCP servers with in-process singleton retrievers to eliminate network overhead in RAG workflows. <a href="https://app.daily.dev/posts/3dknsIul1" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>MikroORM 7 goes unchained with zero dependencies</strong>: The stable release drops Knex for Kysely, moves to native ESM, and decouples from the Node.js runtime for better edge compatibility. <a href="https://app.daily.dev/posts/ifAJqhN7p" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Snap saves 76 percent on costs with GPU Spark</strong>: By migrating A/B testing to NVIDIA cuDF on GKE, Snap reduced concurrent GPU requirements from 5,500 to 2,100 while gaining 4x speedups. <a href="https://app.daily.dev/posts/LmWwejP3e" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AMD Ryzen AI NPUs get real time power reporting</strong>: The upcoming Linux 7.1 kernel adds a new ioctl to the AMDXDNA driver for monitoring NPU power estimates and column utilization. <a href="https://app.daily.dev/posts/NSpYAgqvz" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The $1.6 million weekend of stateless incompetence</strong>: An AI contract review agent managed to rack up a seven-figure bill in 48 hours by retrying a single document 1,000 times because the API gateway was too ‘stateless’ to notice the loop. This is what happens when you expose expensive LLM endpoints to non-deterministic consumers without session-aware economic governance. If your gateway doesn’t have a kill switch keyed to an agent session ID, you’re just waiting for your own bankruptcy-by-API. <a href="https://app.daily.dev/posts/W7TxB28js" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Amazon is using AI to supervise the AI that broke Amazon</strong>: After internal AI tools wiped out 99% of North American orders and deleted production environments, Amazon’s brilliant solution is to spend $200 billion on more AI to ‘supervise’ the mess. They’ve laid off 16,000 actual engineers only to find that next-token predictors don’t actually understand ‘intent’ or ‘consequences’ when given production credentials. It’s a recursive loop of operational failure rebranded as innovation. <a href="https://app.daily.dev/posts/Nu2d4YyjL" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Orbital datacenters are the new peak insanity</strong>: NVIDIA is hyping ‘Space-1’ modules for geospatial intelligence, but Gartner analysts are correctly calling the orbital datacenter trend ‘peak insanity.’ Between the prohibitive launch costs and the minor detail that vacuum is a terrible heat conductor, putting GPUs in space is a vanity project for the ‘more money than physics’ crowd. It’s the ultimate edge case for people who find terrestrial latency less annoying than gravity. <a href="https://app.daily.dev/posts/5QtI91Gu3" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
