Tuesday, July 21, 2026

1. Fireside Chat with Anthropic's Claude Code Team: How AI Coding Agents Reshape Software Development
   T2Simon Willison 博客（Atom）

   Cat Wu and Thariq Shihipar from Anthropic's Claude Code team discussed the latest on Claude Code, Claude Tag, and Fable at the AI Engineer World's Fair. Claude Tag now handles 65% of product engineering PRs, and the system prompt was reduced by 80%. They highlighted how model improvements free developers from micro-monitoring to focus on higher-level creativity and product decisions. Traditional software engineering norms are overturned: rewrites become beneficial, and product taste and business acumen become more valuable. They also shared their internal 'ant fooding' culture and Claude Tag's proactive collaborative features.

2. PPL-Factory: Task-Aware and Budget-Aware Data Selection from Language Modeling to Reasoning
   T2arXiv AI/CL/LG（论文）

   PPL-Factory proposes a data selection framework combining task-aware perplexity scores with data budget-aware criteria for LLM fine-tuning. Experiments show it outperforms existing methods with only 1% training data, and exceeds full-data fine-tuning by 0.9 on GSM8K and 4.8 on MATH with 10% data. The method is simple, interpretable, and effectively reduces fine-tuning computational cost.

—

1. X Relaunches Rebuilt Android App After Year-Long Effort
   T2TechCrunch：AI（RSS）

   X has announced the global release of its fully rebuilt Android app. The new version, developed over the course of a year, aims to improve performance, stability, and user experience while integrating the platform's latest features. This launch marks a significant upgrade to X's mobile tech architecture, with deep optimizations for the Android ecosystem.

—

—