<h2>Must know</h2>
<ul>
<li><strong>Cursor launches Composer 2 with self-summarization</strong>: Cursor has released Composer 2, an in-house coding model that uses a reinforcement learning technique called self-summarization to compress its context window to 1,000 tokens when limits are hit. The company claims this reduces compaction errors by 50% compared to traditional sliding windows and outperforms Claude 4.6 on terminal-based benchmarks. Most notably, it is priced significantly lower than Anthropic’s flagship models at $0.50 per million input tokens. <a href="https://app.daily.dev/posts/mIKxoNmwG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Interlock ransomware exploits Cisco zero-day</strong>: A critical zero-day vulnerability in Cisco Secure Firewall Management Center (CVE-2026-20131) was exploited by the Interlock ransomware group for over a month before a patch was available. Amazon’s MadPot honeypot uncovered the attack chain, which allows remote code execution via deserialization and carries a maximum CVSS score of 10. Security teams are urged to patch immediately as the group is actively targeting healthcare and government sectors. <a href="https://app.daily.dev/posts/kS1FWAP6y" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>FBI seizes Handala domains after Stryker wipeout</strong>: The FBI and DOJ have seized the websites of the pro-Iranian group Handala following a devastating cyberattack on medical tech giant Stryker. The attackers reportedly compromised a Microsoft Intune admin account to remotely wipe tens of thousands of employee devices, bypassing traditional ransomware in favor of pure disruption. CISA has responded by urging organizations to implement dual-administrator approval for high-impact Intune actions like device wipes. <a href="https://app.daily.dev/posts/N8kFe3mft" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Google AI Studio adds vibe coding for full-stack app generation</strong>: A new Build Mode generates frontend, backend, and Firebase integrations from natural language prompts, deploying directly to Cloud Run. <a href="https://app.daily.dev/posts/vDejXlEPp" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>LangSmith rebrands Agent Builder to Fleet for enterprise scale</strong>: The platform now features tiered permissions, Slack bot identities, and a centralized ‘Agent Inbox’ for human-in-the-loop oversight. <a href="https://app.daily.dev/posts/fnry81udE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Neo4j Aura Agent simplifies GraphRAG deployment</strong>: The low-code platform auto-generates agent configurations from graph schemas and supports Text2Cypher for complex data retrieval. <a href="https://app.daily.dev/posts/mc7JPTHNY" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Heroku doubles slug size to 1GB for AI frameworks</strong>: The platform update also includes longer build timeouts and a CLI migration to ESM to better support heavy ML dependencies. <a href="https://app.daily.dev/posts/vu5ItVeI2" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Checkmarx One adds AI orchestration for vulnerability triage</strong>: The framework manages five specialized agents to automate discovery and remediation across the DevSecOps pipeline. <a href="https://app.daily.dev/posts/PE8tBh7Qo" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AWS SDK for Ruby adds bulk S3 directory transfers</strong>: New methods in aws-sdk-s3 automate multipart handling and parallelism for recursive directory uploads and downloads. <a href="https://app.daily.dev/posts/tlRI3NVwm" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Databricks launches serverless GPU training runtime</strong>: The AI Runtime (AIR) provides on-demand NVIDIA H100 clusters with pre-installed PyTorch and Ray for distributed fine-tuning. <a href="https://app.daily.dev/posts/ofgaFNNzJ" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The vibe coding marketing offensive</strong>: Google is unironically adopting the term ‘vibe coding’ to describe what is essentially just a wrapper for Gemini generating boilerplate code. While they pitch it as a ‘shift toward AI-native development,’ it’s mostly a way to lock developers into the Google Cloud Run and Firebase ecosystem before they realize the ‘vibe’ doesn’t include debugging complex state issues. It’s a classic case of rebranding prompt engineering as a revolutionary architectural paradigm. <a href="https://app.daily.dev/posts/vDejXlEPp" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Semantic models as the new prompt engineering</strong>: Snowflake is pitching ‘Semantic Models’ as the cure for AI agents that can’t understand basic business logic, essentially asking humans to do the hard work of data modeling so the AI can take the credit. Instead of fixing the underlying reasoning flaws in LLMs, we’re now expected to write exhaustive metadata files just so the agent doesn’t hallucinate a ‘revenue’ column that doesn’t exist. It’s just another layer of manual labor disguised as an ‘AI-first’ strategy. <a href="https://app.daily.dev/posts/6wWXcyCzZ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The 6-engineer unicorn migration story</strong>: Vercel is touting a story about a team of six engineers serving 3 million customers by migrating to their platform, conveniently ignoring the massive architectural debt usually incurred by ‘coding agents’ doing a full product rewrite. The claim that they ‘eliminated the need for DevOps’ is the standard serverless siren song that usually ends in a massive bill and a complete lack of observability when things actually break. It’s a perfect example of ‘survivorship bias’ used as a marketing case study. <a href="https://app.daily.dev/posts/zNuNcsVh2" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
