Jason Says
Today's most telling signal: Zuckerberg quietly admits agents aren't working as planned, while SiliconFlow marches toward an IPO with a -24% gross margin — the two hottest AI tracks (agent deployment and inference infra) are both showing their cracks at the same time. This is the reality the industry needs to stop glossing over.
An open-source Claude Code skill/plugin that makes your AI agent 'talk like caveman' — stripping verbose filler while keeping full technical accuracy. Cuts ~65-75% of output tokens, slashing API costs for heavy coding agent users. Compatible with Cursor, Codex, Gemini, Copilot, and 30+ other tools.
A developer fed up with manual job applications built a complete AI-powered job search system on Claude Code with 14 skill modes, a Go dashboard, PDF generation, and batch processing. Now open-source. The core insight: companies use AI to filter candidates — candidates should use AI to filter companies.
Strix is an open-source AI pentesting tool with autonomous agents that find and fix app vulnerabilities. It integrates natively with GitHub Actions and CI/CD pipelines, automatically scanning every pull request to block insecure code before production — giving solo devs enterprise-grade security without the security engineer.
Vercel's Chief of Software Andrew Qu breaks down how their agent framework 'eve' was built, arguing that skills layers, execution sandboxes, and agent-readable websites are the new infrastructure primitives. Key takeaway: agents aren't just smarter function calls — they represent a fundamentally new software paradigm.
Adobe is testing 'agentic sites' where web pages are no longer static HTML but are assembled in real-time by AI around each visitor's individual intent. Shared at the AI Engineer World's Fair, this concept could fundamentally reshape frontend development and content ops — opening new product opportunities for indie developers.
Mark Zuckerberg told staff in an internal meeting that Meta's AI agent development is not moving as fast as he had hoped. This rare admission from the CEO of the world's largest social platform signals that even top-tier AI labs are hitting real walls with agent deployment — and scrutinizes his earlier bold claims about AI replacing engineers.
AutoMem brings cognitive science's 'metamemory' concept to LLM training — letting models learn when to encode, retrieve, and organize information as a first-class trainable skill. Why it matters for developers: future agents could manage their own long-term memory without you hardcoding memory scheduling logic into your prompts or pipelines.
Chinese AI inference platform SiliconFlow has filed for a Hong Kong IPO: 10M registered users, 292x paying customer growth in one year, $1.1B valuation — but buried in the 347-page prospectus is a cold truth: a -24% gross margin in 2025. It's the most transparent data point yet on the 'grow now, profit later' bet in AI infrastructure.
伦敦VC基金，押注欧洲连续创业者，预期AI退出浪潮将催生下一代有经验的创始人——间接加速欧洲AI生态资本循环。
Subscribe for daily AI updates + free playbook
📘 Subscribe Free