Jason Says
Today's top thing to watch is Anthropic's official plugin directory — this isn't just a GitHub repo, it's the starting point for Claude Code's ecosystem transitioning from wild growth to structured governance. Whoever secures a spot in the directory first gains early access to official traffic.
Anthropic has opened a curated, officially maintained plugin directory for Claude Code, listing vetted high-quality plugins. This marks a shift from a fragmented third-party MCP server landscape toward an officially endorsed plugin marketplace. Users are still advised to independently assess trust before installing, as Anthropic does not guarantee plugin content.
This open-source tool converts any codebase, knowledge base, or documentation into an explorable, searchable, interactive knowledge graph. It natively supports Claude Code, Codex, Cursor, Copilot, and Gemini CLI. It directly improves agent context comprehension on large projects and offers an alternative approach to tools like CodeGraph.
A Firefox-forked browser purpose-built for AI agents offers deep optimization for automated operations. Unlike Chrome DevTools MCP solutions, this is a standalone agent-specific runtime giving agents a more controlled, isolated environment for web tasks — without interfering with the user's regular browsing sessions.
Abacus AI CEO Bindu Reddy predicts June will see a wave of flagship model launches, including Gemini 3.5 Pro, GPT 5.6/5.6 Pro, Sonnet 4.7, and Mythos. She expects Google to reclaim competitive advantage through competitive pricing — if Gemini 3.5 Pro delivers Opus-level performance at a lower cost, a price war could follow.
DeepSeek's new Flash model is priced extremely low, targeting small tasks and high-volume data operations. For developers making frequent API calls, it's a serious cost-optimization option worth evaluating — particularly well-suited for high-throughput scenarios like data cleaning, batch labeling, and log analysis.
Researchers introduce CUSP, the first standardized benchmark for evaluating AI systems' ability to predict scientific progress, covering feasibility assessment and milestone forecasting. It provides the first quantitative framework for measuring AI "foresight," offering a baseline for building research-assistance products and defining future scientific agent capabilities.
现有文生图奖励模型依赖大规模人工偏好数据，成本高且评分标准不透明。AutoRubric-T2I 用 VLM 自动生成评分规则，无需人工标注即可对图像质量和提示词对齐度打分。对开发者意义：可低成本搭建自定义图像生成质量评估流水线，适合独立开发者做产品质控。
Subscribe for daily AI updates + free playbook
📘 Subscribe Free