<h2>Must know</h2>
<ul>
<li><strong>Quantum threat to Bitcoin and Ethereum is closer than expected</strong>: New research from Google and Caltech indicates that the elliptic curve cryptography used by major blockchains is significantly easier to break with quantum computers than RSA. The Caltech paper introduces error correction techniques that could reduce the physical qubit overhead for Shor’s algorithm to just 10,000, a threshold nearing the 6,000 qubits already demonstrated last year. Cryptography engineer Filippo Valsorda urges immediate migration to ML-DSA signatures, warning that non-post-quantum key exchanges should now be treated as potential active compromises. <a href="https://app.daily.dev/feed-by-ids?id=cr4dvz28B&amp;id=GJZmZyZhK" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cursor releases Composer 2 model to challenge general frontier AI</strong>: Cursor has launched Composer 2, a 1.04 trillion parameter Mixture-of-Experts model specifically trained on real-world engineering tasks rather than contaminated benchmarks like SWE-bench. In internal CursorBench evaluations, which feature median changes of 181 lines, the model outperformed Claude Opus 4.6 while maintaining lower inference costs. This development, reported by Robert Youssef, suggests that domain-specialized training on specific developer harnesses can beat the ‘bigger is better’ general intelligence strategy. <a href="https://app.daily.dev/posts/osQspcxBm" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic accidentally leaks Claude Code source code via npm</strong>: Anthropic exposed over 500,000 lines of proprietary TypeScript code for its Claude Code CLI due to a misconfigured source map in a public npm release. The leak, attributed to a setting in the Bun build tool, revealed unreleased features like ‘KAIROS’ autonomous background agents and internal model codenames. While no customer data was compromised, security researchers at Adversa and Socket warn that the exposure allows attackers to deterministically exploit the agent’s internal context management and tool-calling logic. <a href="https://app.daily.dev/feed-by-ids?id=OULS6HmRL&amp;id=NF4kLGa2o&amp;id=nOFJ29Zgp" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>AWS releases Smithy-Java framework for type-safe clients using Java 21 virtual threads</strong>: The framework generates blocking-style APIs that simplify development while matching the performance of complex async alternatives. <a href="https://app.daily.dev/posts/08k8pNrZM" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kubernetes community launches Certified AI Conformance program</strong>: The program standardizes Dynamic Resource Allocation for GPUs and all-or-nothing scheduling via Kueue to prevent training deadlocks. <a href="https://app.daily.dev/posts/kcqVDySrK" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Dragonfly P2P mesh reduces AI model distribution traffic by 99 percent</strong>: The CNCF project now natively supports hf:// protocols, allowing multi-node GPU clusters to fetch large models via a peer-to-peer mesh. <a href="https://app.daily.dev/posts/8ZoQ9LlXc" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Discord open sources Osprey safety engine processing 2 million rules per second</strong>: The polyglot architecture uses a Rust coordinator and Python workers to execute safety rules written in a custom DSL. <a href="https://app.daily.dev/posts/c0Z9GNgQz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Coupang replaces database sequences at scale using DynamoDB and client-side caching</strong>: The system serves 50,000 sequences per second with minimal writes by accepting gaps and using a sliding window refill algorithm. <a href="https://app.daily.dev/posts/TEv7fV07V" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Tree-sitter grammar for R significantly improves programmatic code analysis and linting</strong>: The new grammar powers a Rust-based ecosystem of formatters, mutation testing tools, and structural diffing for the R language. <a href="https://app.daily.dev/posts/xQoTxzSey" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Postman fellow argues APIs are the true primitive over serverless functions</strong>: Lessons from Postman’s infrastructure show that mandating interface-oriented design over function-oriented design fixes observability and architectural chaos. <a href="https://app.daily.dev/posts/7tqnVEN9g" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The Agent Skills hype is just <a href="http://AGENTS.md" target="_blank" rel="noopener nofollow">AGENTS.md</a> with a marketing budget</strong>: Industry voices are pushing ‘Agent Skills’ as a revolutionary procedural knowledge bundle, but as critics point out, a well-maintained markdown file in your repo covers 80% of the use cases. Unless you enjoy adding more CLI dependencies and complex permission models to your stack, stick to version-controlled documentation that AI can already read. <a href="https://app.daily.dev/posts/Lb3Z9DESw" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft Copilot is officially for entertainment purposes only</strong>: Despite the aggressive enterprise marketing, Microsoft’s legal team has buried a disclaimer in the terms of use stating the tool is for ‘entertainment purposes only.’ It’s a classic case of corporate double-speak: the marketing department calls it a ‘dependable workmate’ while the lawyers treat it like a digital Magic 8-Ball to dodge liability for hallucinations. <a href="https://app.daily.dev/feed-by-ids?id=QPAxbn9Aj&amp;id=chPbgLDu0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Nvidia’s SchedMD acquisition is a slow-motion train wreck for open scheduling</strong>: Nvidia claims acquiring the company behind Slurm is about ‘innovation,’ but the industry knows it’s about hardware lock-in. By employing the core maintainers of the scheduler used by 60% of supercomputers, Nvidia can subtly deprioritize AMD and Intel support until Slurm becomes just another proprietary driver for InfiniBand. <a href="https://app.daily.dev/posts/RZr93bEOr" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
