Rami's Readings #146 - Claude Mythos
The latest on AI, LLMs, Claude Mythos, Teams of Agents, Harness Engineering, Voxtral, Japan, Sandboxing Agents, and more.
Welcome to Rami’s Readings #146 - a weekly digest of interesting articles, papers, videos, and X threads from my various sources across the Internet. Expect a list of reads covering AI, technology, business, culture, fashion, travel, and more. Learn about what I do at ramisayar.com/about.
👋🏼 Welcome New Subscribers
Hello! A hearty thank you for subscribing to Rami's Readings! There are quite a few new subscribers this week, thanks to recommendations from The AI Ethics Brief, Product Byte, and The VC Corner. I am thrilled to have you on board! In this newsletter, I curate the best papers, tweets, and articles I have read during the week focusing on LLMs, AI, economics, business, and technology news. You can learn more about me on my website.
🤖 AI Reads
Project Glasswing a.k.a. Claude Mythos
Notes: Rightfully caused a panic in the markets with respect to cybersecurity. Ignore the hype and criticism on all sides. Given the quality level of coding agent automation driven by frontier LLMs since November 2025, it’s a no-brainer that we’re entering a new cybersecurity era. Also read Wired.
multica-ai / multica: Open-Source Managed Agents Platform
Notes: Competing open source project to Claude Managed Agents.
getcompanion-ai / feynman: Four Bundled Research Agents, Dispatched Automatically
Notes: Kind of similar to Model Council.
Meta-Harness: End-to-End Optimization of Model Harnesses
Notes: Autoresearch for your autoresearch.
garrytan / gbrain: Garry's Opinionated OpenClaw / Hermes Agent Brain
Notes: Garry Tan’s opinionated OpenClaw setup.
The Art of Building Verifiers for Computer Use Agents
Notes: This is a really interesting paper for Microsoft Research (not my department).
LFM2.5-VL-450M: Structured Visual Intelligence, Edge to Cloud
Notes: New model from MIT’s Liquid AI.
Mistral Launches Voxtral 4B TTS 2603
Notes: Outperforms ElevenLabs’s closed source model.
💼 Business Reads
Wall Street Banks Try Out Anthropic’s Mythos & Bank of Canada, Major Lenders Met on Anthropic AI Cyber Risk
Notes: I hope there will be a renewed pushed for improved cybersecurity everywhere, not just banks.
Japan Bets $16 Billion to Propel Rapidus in Global AI Chip Race
Notes: More competition for Nvidia is good.
Hermès Sticks With Plan to Raise Output
Notes: This is honestly a surprise to me, but I guess demand is so high that bad vibes wouldn’t lower demand for several years. P.S. Hermès’ flagship store in Japan is something else.
🔀 Other Reads
rtk-ai / rtk: CLI Proxy That Reduces LLM Token Consumption by 60–90% on Common Dev Commands
Notes: Hooks into coding agents like Claude, OpenCode, Codex, etc. to remove unnecessary tokens from common commands like ls, cat, grep, etc. One of my favorite things about the coding agent revolution is the revival of interest in terminal-based development and the explosive growth of CLI tools, (e.g. ripgrep, bat, exa, etc. all new-to-me).
coast-guard / coasts: Localhost Service Isolation and Orchestration for Git Worktrees
Notes: Another approach for coding agent isolation.
Freestyle: Sandboxes for Coding Agents
Notes: YC startup focused on coding agent isolation and sandboxing for enterprise.
Signing off from Redmond, WA.