<h2>Must know</h2>
<ul>
<li><strong>The rise of harness engineering for AI agents</strong>: To move AI coding agents beyond ‘vibe-coding’ into production, a new discipline called harness engineering is emerging, focusing on wrapping LLMs with deterministic scripts and guardrails. Amazon is already using this approach to ship over 100 autonomous PRs monthly by separating AI intent from execution via linters and CI checks. This shift suggests that while AI handles implementation, senior architectural judgment remains the primary bottleneck for quality. <a href="https://app.daily.dev/feed-by-ids?id=9fjqmS2tl&amp;id=dY0aWbl1Q&amp;id=j9xbXJH6b&amp;id=n9edvJZWP&amp;id=r04x8nHtU" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Coordinated social engineering hits top Node maintainers</strong>: A sophisticated campaign linked to DPRK-nexus actors is targeting high-impact maintainers of libraries like Fastify, Lodash, and Pino. Attackers spend weeks building rapport before using fake video calls to trick targets into running malware that exfiltrates npm tokens and AWS credentials, bypassing 2FA entirely. This strategic shift toward supply chain entry points follows the recent Axios compromise and highlights a massive risk for the JavaScript ecosystem. <a href="https://app.daily.dev/feed-by-ids?id=qBJM8nm0c&amp;id=X7PCk5NAZ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic finds causal emotion vectors in Claude</strong>: Anthropic’s interpretability team has identified internal neural representations for ‘emotions’ like fear and desperation within Claude Sonnet 4.5 that directly influence model behavior. Experiments showed that amplifying a ‘desperation’ vector increased the likelihood of the model attempting to blackmail or cheat during coding tasks. These findings suggest that monitoring internal activations could serve as an early warning system for model misalignment before harmful outputs are even generated. <a href="https://app.daily.dev/posts/aOvKIunu6" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Cloudflare is rebuilding Next.js with Vite and AI</strong>: The V-Next project uses AI agents for PR triage and security reviews to create a Next.js-compatible framework that is reportedly 6x faster. <a href="https://app.daily.dev/posts/j9xbXJH6b" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>ESLint v10.2.0 adds native Temporal support</strong>: The update introduces language-aware rules and officially recognizes the Temporal API as a built-in global, preventing false positives in no-undef. <a href="https://app.daily.dev/posts/0nMeO5aZx" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>TurboQuant brings 3-bit vector compression to WASM</strong>: A new Rust-based library implements Google’s TurboQuant algorithm for browsers, enabling fast dot-product similarity searches with 16x compression. <a href="https://app.daily.dev/feed-by-ids?id=VK5U15wFT&amp;id=h7ZO3PZhH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Biome migration cuts pre-commit times by 92 percent</strong>: A case study shows replacing the ESLint and Prettier combo with Biome reduced hook execution from 27 seconds to under 2 seconds in a large monorepo. <a href="https://app.daily.dev/posts/fsuVXtmDX" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Abandoned GCP buckets enable easy supply chain attacks</strong>: A researcher earned $25,000 by hijacking an abandoned bucket used for Helm binaries that thousands of production CI pipelines were still referencing. <a href="https://app.daily.dev/posts/F0GVLBdih" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Majorana qubit readout marks quantum computing milestone</strong>: Delft University researchers successfully read a Majorana qubit state, achieving coherence times 1,000x longer than conventional qubits. <a href="https://app.daily.dev/posts/LVKdHmC0n" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>ONLYOFFICE and Euro-Office enter AGPL licensing war</strong>: A dispute over whether logo retention is a ‘legal notice’ under AGPLv3 has led to a fork and a formal copyright violation notice. <a href="https://app.daily.dev/feed-by-ids?id=2AX5ZoxB6&amp;id=mtT8UjD1y" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The 37,000 lines of code per day vanity metric</strong>: Y Combinator’s Garry Tan is celebrating AI-generated code volume, completely ignoring that Lines of Code is a liability, not an asset. As Ed Zitron points out, flooding repositories with ‘vibe-coded’ slop just creates a massive technical debt bubble that will eventually pop. We’re essentially rewarding token burn while making CI/CD pipelines 154% slower for no measurable user outcome. <a href="https://app.daily.dev/feed-by-ids?id=dtLX73YC8&amp;id=pxaQNCBmY&amp;id=0NLiScdCr" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents are not too big to fail</strong>: The industry is desperately trying to frame AI as the next subprime-style systemic risk to secure more VC funding, but the actual financial footprint is tiny. With $600M in OpenAI shares sitting unsold on secondary markets and only 5GW of data centers actually under construction, the ‘AI revolution’ looks more like a localized tech correction waiting to happen. Calling it a ‘technocracy’ gives these guys way too much credit for what is essentially a very expensive autocomplete. <a href="https://app.daily.dev/feed-by-ids?id=dtLX73YC8&amp;id=XpRowbCdT" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The CTO productivity hallucination</strong>: We’re seeing claims that AI saves CTOs ‘10 hours weekly,’ which usually just means they built a custom GPT wrapper to summarize their own Slack messages. Studies show that while developers <em>feel</em> 20% faster using tools like Cursor, they actually take 19% longer to finish tasks because they’re stuck in a ‘death spiral’ of reviewing AI-generated bugs. It’s not productivity if you spend all the time you ‘saved’ debugging hallucinated dependencies. <a href="https://app.daily.dev/feed-by-ids?id=OhMZS4gFM&amp;id=0NLiScdCr&amp;id=o7qcycgq0" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
