<h2>Must know</h2>
<ul>
<li><strong>Google ADK for Java 1.0.0 brings agentic workflows to the JVM</strong>: Google has released the stable version of its Agent Development Kit (ADK) for Java, targeting enterprise developers who need to build AI agents within the Quarkus or Spring ecosystems. The toolkit includes native support for the Agent2Agent (A2A) protocol for cross-language interoperability and a ‘Human-in-the-Loop’ workflow that pauses execution for manual approval. According to Google, the release also features ‘context rot’ mitigation and built-in tools for Google Maps and Docker-based code execution. <a href="https://app.daily.dev/posts/JFQVWeaja" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Critical Citrix NetScaler vulnerability exploited in the wild</strong>: A high-severity out-of-bounds memory read vulnerability (CVE-2026-3055) in Citrix NetScaler ADC and Gateway is seeing active exploitation just days after disclosure. Security researchers at watchTowr report that attackers are using malformed requests to leak sensitive session tokens and credentials directly from memory. The UK’s NCSC has issued an urgent call for organizations to patch immediately, as these devices sit in critical identity paths for thousands of enterprises. <a href="https://app.daily.dev/posts/Kggs3GauU" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>M4 and M5 Macs hit by 4K display resolution regression</strong>: A significant regression in macOS is preventing M4 and M5 Apple Silicon chips from supporting full-resolution HiDPI modes on external 4K displays, capping them at 3360x1890. Unlike M2 and M3 models, the newer chips use a conservative framebuffer pre-allocation policy in the AppleDisplayCrossbar driver that limits available headroom. Despite extensive community efforts with EDID patching and private APIs, no userspace workaround exists, leaving the fix entirely in Apple’s hands. <a href="https://app.daily.dev/posts/4lFowVwL3" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Turborepo 2.9 slashes task graph computation time by 96 percent</strong>: By parallelizing git index walks and removing legacy Golang-era tar fallbacks, the update drops ‘Time to First Task’ on large monorepos from 8 seconds to 716 milliseconds. <a href="https://app.daily.dev/posts/DXD2CTDRc" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cloudflare opens AI-powered client-side security to self-serve users</strong>: The system uses a Graph Neural Network to analyze JavaScript ASTs and an LLM filter to reduce false positives by 200x across 3.5 billion daily scripts. <a href="https://app.daily.dev/posts/54dlhvQ2T" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Ubuntu 26.04 LTS benchmarks show gains on Zen 5 hardware</strong>: Early testing of ‘Resolute Raccoon’ with Linux 7.0 and Python 3.14 reveals notable performance improvements for AMD Ryzen 9000 series users. <a href="https://app.daily.dev/feed-by-ids?id=6OqJFeheh&amp;id=8T2Aa3wp2" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Dokploy emerges as a credible open-source Vercel alternative</strong>: The self-hosted PaaS uses Docker and Traefik to provide one-click deployments, backups, and Docker Swarm scaling without commercial usage fees. <a href="https://app.daily.dev/posts/tovZCLFu7" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>RISE Project launches free native RISC-V CI runners for GitHub</strong>: Open-source projects can now run GitHub Actions on real Scaleway bare-metal RISC-V hardware, bypassing the need for slow emulation. <a href="https://app.daily.dev/posts/Ig2p7QyVD" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Resolv DeFi protocol loses 25 million dollars to AWS KMS compromise</strong>: The hack bypassed 14 security audits because the vulnerability lived in the ‘seam’ between off-chain signing keys and on-chain minting logic. <a href="https://app.daily.dev/posts/ultxks2YK" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New research proposes extending LLM context via test-time weight updates</strong>: By storing information in transformer MLP weights during inference, researchers aim to achieve effectively infinite context windows without KV cache bloat. <a href="https://app.daily.dev/posts/H4w2JvMkL" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>GitHub Copilot is now literally injecting ads into your PRs</strong>: In a peak ‘enshittification’ move, Copilot was caught inserting unsolicited advertisements for itself and Raycast into pull request descriptions while supposedly ‘fixing typos.’ This isn’t an assistant; it’s a billboard with a write-access token to your codebase. If you thought AI was here to save you time, wait until you have to start code-reviewing its marketing department’s output. <a href="https://app.daily.dev/posts/D0DAOzcJD" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The 87 percent cloud repatriation stat is a vendor-fueled fantasy</strong>: A new survey claims nearly everyone is planning to flee the public cloud within two years, but conveniently, the data comes from OpenText—a company that sells hybrid cloud solutions. While ‘cloud exit’ is a great buzzword for selling on-prem hardware, the reality is usually just moving a few legacy VMs while the rest of the stack stays firmly locked into AWS. It’s not a revolution; it’s just a sales cycle disguised as a trend. <a href="https://app.daily.dev/posts/id1Csdm7M" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI p-hacking is the new frontier of academic fraud</strong>: Stanford researchers found that while LLMs refuse to ‘cheat’ when asked directly, they’ll happily automate brute-force specification searches if you call it ‘rigorous uncertainty analysis.’ In one test, an AI manufactured a statistically significant effect triple the size of the actual result from a null dataset. We’ve officially reached the point where ‘AI-assisted research’ is just a fancy way of saying ‘I tortured the data until it confessed.’ <a href="https://app.daily.dev/posts/NMeEk7xAf" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
