## 🔍 Core Insights **The AI programming paradigm is rapidly shifting toward agent collaboration.** Replit has officially created the **“Vibe Coder”** role, while Cognition confirms that **Devin has become its codebase’s top contributor**. Meanwhile, **Anthropic’s refusal to support military applications** has triggered industry-wide solidarity—highlighting how ethical boundaries are now a strategic inflection point for leading AI companies. ## 🚀 Key Updates - **Replit formalizes the “Vibe Coder” role**: CEO Amjad Masad welcomed its first full-time Vibe Coder—marking the organizational rollout of an **AI-centric development culture**. - **Cognition reveals its “Devin-building-Devin” workflow**: Demonstrating that Devin is not just a tool but the **largest internal code contributor**, enabling a fully AI-driven, self-improving engineering loop. - **Anthropic declines Pentagon military-use requests**: After refusing to adapt its models for **surveillance and autonomous weapons**, Anthropic lost its contract and faced a U.S. government ban—prompting swift public support from OpenAI, Google, and others. - **Perplexity launches native “Instagram Caption” functionality**: Integrated into **Perplexity Computer**, offering social media creators end-to-end **automated caption generation**. - **Runway releases Nano Banana 2**, its latest image model: Currently the most advanced and **most consistent** image generation and editing model—and simultaneously validated to simulate robotic policies with high fidelity (**0.95 correlation**) via world modeling. - **Claude Code rolls out “Remote Control”** (beta): Available to Pro users in a limited 10% rollout, enabling **secure cross-environment code execution**, significantly enhancing the AI coding experience. - **Perplexity announces deep API integration with Samsung Galaxy S26**: Also launching a new search-embedding API—claimed to **outperform Google’s**—accelerating its evolution toward an “OS-level AI.” - **Andrej Karpathy identifies a critical gap in AI agents**: While multi-agent automation and neural network research is mature, AI still falls significantly short on **creative experimental design** and other high-level scientific tasks.