Jason Says
The most interesting signal today: Anthropic and Cloudflare both dropped production-grade skills on the same day — finance and security, two of enterprise's hardest domains. The Skills ecosystem is rapidly moving from toy demos to real infrastructure, and indie developers now have the best starting point they've ever had.
Anthropic released an open-source financial services repo with reference agents, skills, and data connectors for investment banking, equity research, PE, and wealth management. Works as a Claude Cowork plugin or via the Managed Agents API — same system prompt, two deployment paths. The most production-ready official skills reference for enterprise finance workflows to date.
Cloudflare open-sourced a coding-agent skill that orchestrates isolated agents through six phases of security auditing — from recon to neutral reporting with machine-readable, independently verified findings. This is the seed skill behind Cloudflare's own vulnerability discovery harness, offering developers a reusable template for automated security auditing.
Builder.io open-sourced Agent-Native, a TypeScript framework where each capability is defined once as an action — the agent uses it as a tool, and the UI calls it from code. This eliminates the split between agent logic and frontend logic, making it easier to ship agentic apps with purpose-built interfaces without duplicating implementation.
Meta's AI agent Muse has surpassed ChatGPT's early mobile launch in US and Canada downloads and DAUs, per Appfigures data. However, Amazon has already blocked Muse from accessing Amazon.com, citing its own AI ecosystem interests. Strong growth metrics, but the platform gatekeeping battle has begun.
OpenAI announced its AI has resolved more than 100 open mathematical problems, while forming an independent advisory group to review and communicate findings. Notably, the group has no authority to slow or redirect OpenAI's math research — positioning it more as a credibility layer than genuine governance.
SiliconBench evaluates nine Apple Silicon LLM serving engines across speed, memory, and fidelity — dimensions that speed-only rankings miss. Tested on Qwen3, Qwen3.5, and Gemma 4, it's the first benchmark designed for developers running local models on Mac who need to balance performance without quality regression.
APort Vault replays 4,371 real human-written attacks from a public CTF event against a live payment agent, testing 14 models from 8 labs across five policy configs — totaling 225,964 evaluations. The most realistic benchmark yet for AI agent payment security, essential reading for any developer building agents that touch financial transactions.
BI-Agent automates the full BI workflow — table identification, data transformation, join relationships, and question answering — that previously required manual analyst work in tools like Power BI or Tableau. BI-Bench provides standardized evaluation. For developers building enterprise data products, this validates the technical feasibility of fully automated natural-language-to-insight pipelines.
智能健康戒指公司 Oura 启动 IPO，此次上市主要是为现有股东套现，Forerunner Ventures 计划出售全部持仓，最高套现 12.6 亿美元。
Subscribe for daily AI updates + free playbook
📘 Subscribe Free