<h2>Must know</h2>
<ul>
<li><strong>AI agents face production reality check with file corruption and security leaks</strong>: Recent post-mortems from daily.dev and independent developers reveal that AI coding agents are corrupting .NET project files and leaking GitHub tokens between user sessions. These incidents, reported on March 29, 2026, highlight that ‘vibe coding’ lacks the necessary guardrails for destructive operations like git cleans or complex file encoding. Engineers are now implementing ‘Rules of Engagement’ via <a href="http://CLAUDE.md" target="_blank" rel="noopener nofollow">CLAUDE.md</a> and cgroup limits to prevent memory exhaustion from long LLM conversations. <a href="https://app.daily.dev/feed-by-ids?id=9HucugjHw&amp;id=tdNgi1Q4D" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Critical RCE vulnerability in Langflow framework</strong>: CISA has issued a warning for CVE-2026-33017, a critical remote code execution flaw in the Langflow AI agent framework with a CVSS score of 9.3. According to AI Cyber Insights, the vulnerability allows unauthenticated attackers to execute arbitrary Python code via a crafted HTTP request to a public API endpoint. Users are urged to upgrade to version 1.9.0 immediately to prevent the exposure of cloud API keys and potential lateral movement within CI/CD pipelines. <a href="https://app.daily.dev/posts/vNn8UUgk6" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Rust rewrite of vLLM delivers massive performance gains</strong>: A new project called rvLLM has successfully rewritten the vLLM inference engine in Rust, achieving 3,467 tokens per second on an A100. The binary is significantly leaner at 16MB compared to the 500MB Python original, and it cuts startup time from 120 seconds down to just 6. This drop-in OpenAI-compatible API replacement demonstrates the growing trend of moving AI infrastructure to memory-safe, high-performance languages to reduce cloud overhead. <a href="https://app.daily.dev/posts/u7fISleFK" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Liter-llm provides a Rust-core universal API client with 11 language bindings</strong>: A supply-chain-safe alternative to litellm that uses TOML configuration and OpenTelemetry for observability across 142 providers. <a href="https://app.daily.dev/posts/2yHoir3TW" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Virtual clusters cut Kubernetes control plane costs by 90 percent</strong>: Tools like vCluster and Kamaji allow platform teams to run tenant control planes as pods, eliminating the $876/year per-cluster ‘tax’ on EKS. <a href="https://app.daily.dev/posts/LZn0volzl" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>C++26 work completes with reflection and memory safety features</strong>: The ISO committee finalized C++26, introducing compile-time reflection and hardened bounds safety already being deployed by Google and Apple. <a href="https://app.daily.dev/posts/qGBXVZMU3" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Pretext library enables DOM-free text measurement in JavaScript</strong>: This library uses Canvas to measure multiline text layout, avoiding expensive DOM reflows for virtualization and masonry layouts. <a href="https://app.daily.dev/posts/HTzI8mpcv" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>DeepSeek R2 projected to undercut GPT-4o costs by fourfold</strong>: Early benchmarks suggest DeepSeek’s upcoming model offers significantly lower token costs for batch architectural tasks despite GPT-4o’s latency lead. <a href="https://app.daily.dev/posts/FiMuYZXRT" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Ruby Central addresses governance crisis following infrastructure security audit</strong>: The board promised a full incident report after a breakdown in stewardship led to concerns over <a href="http://RubyGems.org" target="_blank" rel="noopener nofollow">RubyGems.org</a> access and community trust. <a href="https://app.daily.dev/posts/M2k1JeSOT" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Miasma traps AI web scrapers in endless loops of poisoned data</strong>: A Rust-based tool that feeds garbage content to aggressive bots to protect site resources and training data integrity. <a href="https://app.daily.dev/posts/QZnDQHaJE" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The agentic passive voice is the new dog ate my homework</strong>: Engineers are now using ‘the model decided’ as a linguistic shield to deflect responsibility for shipping broken code, according to recent industry critiques. This ‘agentic passive voice’ treats LLMs like autonomous deities rather than the fancy autocomplete tools they actually are. If your AI agent nukes the production database, you still typed the prompt; stop pretending the software has its own legal department. <a href="https://app.daily.dev/posts/a4WjYQvt2" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI slop is the new open source DDoS attack</strong>: Maintainers are being buried under ‘AI slop’—low-quality, hallucinated pull requests that take 12 times longer to review than they did to generate. This flood of plausible-looking garbage is forcing projects to shut down or implement draconian contributor policies just to survive the ‘productivity’ boost AI promised us. We’ve successfully automated the process of wasting a volunteer’s Saturday afternoon. <a href="https://app.daily.dev/posts/zIHU4vVQv" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Azure Copilot Migration Agent is just a fancy VMware scanner</strong>: Microsoft’s new ‘Migration Agent’ is being hyped as an AI-driven cloud transition tool, but it currently can’t actually execute a migration. It’s essentially a public preview of a dependency mapper that gives you a Bicep template and tells you to do the hard work of replication and cutover manually. It’s a classic case of rebranding existing assessment tools with an ‘AI’ sticker to compete with AWS. <a href="https://app.daily.dev/posts/25eJWn6ov" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
