<h2>Must know</h2>
<ul>
<li><strong>CNCF formalizes agentic standards for Kubernetes</strong>: The Cloud Native Computing Foundation has released a foundational standards document for deploying agentic AI within Kubernetes environments. The framework covers critical Staff-level concerns including agent identity via SPIFFE/SPIRE, zero-trust tenancy, and observability using the OpenTelemetry MELT stack for LLM inference. According to the CNCF, these standards aim to prevent vendor lock-in by standardizing communication protocols like MCP and Agent-to-Agent (A2A) as the ecosystem matures. <a href="https://app.daily.dev/feed-by-ids?id=fUku118d9&amp;id=leqEkFz9G" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Systemic API reliability risks exposed in 2026 report</strong>: A comprehensive analysis of 215+ services by Nordic APIs reveals that AI providers are currently the weakest link in the modern stack, with OpenAI averaging an incident every 2.5 days. While cloud providers like AWS and Azure had fewer outages, their ‘blast radius’ was significantly higher, such as a DynamoDB failure cascading into 141 dependent services. The report warns that implicit dependency chains are defeating explicit redundancy, making independent monitoring and circuit breakers mandatory for Staff engineers. <a href="https://app.daily.dev/feed-by-ids?id=F0p8pG6k1&amp;id=E4QjMx4DG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>FedRAMP approved Microsoft Cloud despite internal pile of shit warnings</strong>: A ProPublica investigation has revealed that FedRAMP authorized Microsoft’s GCC High cloud product despite reviewers internally describing the security as a ‘pile of shit.’ Documents show Microsoft repeatedly failed to provide required data flow diagrams for encryption in transit, but the authorization was pushed through due to political pressure and the operational impossibility of rejecting a widely used federal vendor. This highlights a structural failure in the FedRAMP program where third-party assessors are paid by the companies they evaluate. <a href="https://app.daily.dev/posts/oD4f5bOt7" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Redpanda open sources production AI SDK for Go</strong>: A new SDK providing provider portability, idiomatic streaming, and MCP support to fill the gap in Go-native AI tooling. <a href="https://app.daily.dev/posts/8GYlam0VJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cilium celebrates 10 years with 60 percent CNI market share</strong>: The project is now the de facto standard for large-scale AI training clusters at Microsoft and Google due to its eBPF data plane. <a href="https://app.daily.dev/feed-by-ids?id=3e6m5cs8v&amp;id=tb5nTfo3X" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AMD overhauls Linux hot page tracking with pghot</strong>: New RFC patches unify hot page detection and centralize promotion logic, showing measurable gains on EPYC Zen 5 servers with CXL. <a href="https://app.daily.dev/posts/9z2uCLEQw" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cloudflare Gen 13 servers switch to AMD Turin and Rust</strong>: A complete Rust rewrite of the core request layer (FL2) was required to overcome L3 cache reductions in the new 192-core Turin chips. <a href="https://app.daily.dev/feed-by-ids?id=0evhtm4ji&amp;id=EaD3ek4GN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Uber Michelangelo migrates to 100 plus custom Kubernetes CRDs</strong>: The ML platform now uses a MySQL-backed storage abstraction to bypass etcd scaling limits for its 30 million predictions per second. <a href="https://app.daily.dev/posts/rfWF0zX0x" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>PostgreSQL 19 adds built in REPACK command</strong>: The new command consolidates VACUUM FULL and CLUSTER functionality to reclaim disk space from bloat without a full rewrite. <a href="https://app.daily.dev/posts/NMLmBVAXs" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Wasm emerges as the sandbox of choice for AI agents</strong>: WebAssembly is being adopted by tools like Helm 4 and Boxer to provide ‘deny-by-default’ security for executing untrusted agent code. <a href="https://app.daily.dev/feed-by-ids?id=uqge8CcV2&amp;id=cDNCkYgFI" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The 19 percent AI productivity tax is real</strong>: While marketing departments scream about 10x gains, a METR study cited at RSAC 2026 found experienced developers were actually 19% slower when using AI tools. The ‘velocity’ is an illusion created by generating boilerplate faster, which is immediately offset by a 91% jump in PR review time and a 45% security flaw rate in generated code. We’re essentially paying for the privilege of debugging more expensive, machine-generated technical debt. <a href="https://app.daily.dev/feed-by-ids?id=6DCcfVtqG&amp;id=Lb19kFYlj&amp;id=n30r6zE36" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Nvidia DLSS 5 is just Instagram filters for games</strong>: Nvidia is framing DLSS 5 as a ‘neural rendering’ breakthrough, but gamers and developers are rightfully panning it as an over-processed AI aesthetic that undermines artistic intent. The demo required two RTX 590s just to smear AI-generated lighting over washed-out textures, proving Nvidia is now an AI infrastructure company that views gaming as a secondary test lab for its ‘content-control’ models. It’s motion-smoothing for the next generation, and it looks just as bad. <a href="https://app.daily.dev/feed-by-ids?id=BsIJ2t6dP&amp;id=DLHx86Ehh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The data center construction mirage</strong>: Industry reports indicate that while 241GW of AI data center capacity has been ‘announced’ in the US, only 33% is actually under development. The gap between Nvidia GPU sales and actual operationalized IT load suggests we are in a massive capital misallocation bubble propped up by opaque industry reporting. Most of these ‘AI Factories’ exist only in press releases designed to pump stock prices before the power grid reality check hits. <a href="https://app.daily.dev/feed-by-ids?id=X3my4Sf8s&amp;id=5ZFSDwxdc" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
