Jason Says
OpenWork is the sleeper hit today — a single MCP that lets you share Skills and connected services across Claude Code, Codex, and Cursor is exactly the interoperability layer the agentic dev ecosystem has been missing, and it's open source right now.
OpenAI publicly disclosed it disrupted a Cambodia-based criminal operation using ChatGPT to power investment, romance, gambling, and impersonation scams. This marks OpenAI's first named takedown of a specific criminal operation, signaling that its abuse-monitoring capabilities are now operationally mature.
OpenWork is a free, open-source desktop app that lets developers share AI workflows across Claude Code, Codex, and Cursor via a single MCP. Skills, connected services, and agent configurations created once can be reused across tools, teammates, and machines — positioning it as an open alternative to Claude Cowork.
Google launched and then swiftly pulled an Earth AI feature that let users superimpose AI-generated imagery over real maps, after backlash that it could spread geographic misinformation. The 24-hour reversal underscores how quickly AI tools can become liabilities when misuse vectors aren't anticipated at launch.
Apple CEO Tim Cook hinted that advanced Siri AI capabilities could be unlocked via expanded iCloud+ subscriptions, letting users buy more compute. Rather than a standalone AI subscription, Apple may monetize AI through its existing storage ecosystem — a notably different GTM approach from rivals.
Snapchat updated its recommendation systems to exclude fully AI-generated videos from Spotlight, restricting distribution to real human-made content. It's one of the first major platforms to formally penalize AI slop with algorithmic demotion — a significant signal for AI content tool builders and creators.
Safe Superintelligence, Ilya Sutskever's safety-focused AI lab, reportedly closed a $5 billion round led by Nvidia — the week's largest deal and a record for foundational AI research. The investment signals Nvidia's strategic push beyond hardware into backing the labs that will define next-generation AI architectures.
OmniScope is a training-free token compression framework for omnimodal LLMs that uses the query to independently guide compression per modality, instead of letting one modality dictate what to keep in another. For developers building multi-modal agents, this means dramatically lower token costs without sacrificing answer-critical audio or video cues.
Σ-Mem introduces an online reliability memory for multi-agent LLM systems that tracks not just interaction content but the historical competence and trustworthiness of individual peer agents. For developers building multi-agent pipelines, this means agents can make smarter trust decisions rather than blindly accepting potentially flawed peer outputs.
构建超低延迟、接近真人效果的语音 AI 模型，目标让 AI 电话通过图灵测试，是 AI 客服与语音 Agent 赛道的有力竞争者。
专为私募信贷基金经理打造的 AI 工作流工具，由连续创业者 Ryan Williams 创立，金融 AI 垂直赛道新入局者。
Subscribe for daily AI updates + free playbook
📘 Subscribe Free