{{first_name | Leader}}, welcome back.
AI is entering a phase where the details matter more than the headlines. How it runs, who it is built for, and how it scales are becoming the real story.
These are today’s updates.
Anthropic launches managed agents for production
OpenAI pushes power users to higher plans
Anthropic builds AI-driven security infrastructure
Tools, resources, and a prompt to see what should be decided only once. ⬇️
Anthropic just introduced Claude Managed Agents, a set of APIs that help developers build and run cloud-based AI agents without starting from scratch.
The focus is on real-world use. It includes built-in sandboxing, authentication, tool use, and long-running sessions that continue even when connections drop. Agents can also work together, check their own results against goals, and keep improving while staying within set permissions and tracking systems.
It’s still in research preview, but early tests show better task success rates, especially in workflows like generating files. Pricing follows standard Claude token rates, with an extra cost based on sessions. Early partners include Notion, Asana, Rakuten, Sentry, and Atlassian.
The new ChatGPT Pro plan targets power users, especially developers. At $100/month, it offers 5x more Codex capacity than Plus, with a $200 tier going up to 20x for a limited time.
Codex alone has over 3 million weekly users, and overall usage is growing fast. It also puts OpenAI in closer competition with Anthropic, especially for coding. Pricing is starting to split clearly between casual users and heavy users.
Anthropic has teamed up with 40+ companies, including Apple, Google, Microsoft, and JPMorgan, to use AI to secure core software.
At the center is Claude Mythos Preview, shared only with trusted partners to find and fix vulnerabilities in systems and infrastructure. Early results look strong, like a 93.9% SWE-bench score and thousands of serious bugs found, even a 16-year-old FFmpeg issue.
But Project Glasswing is more like a coordination effort. The same AI can also chain exploits, which is why access is tightly controlled. The $100M in compute credits and open-source funding shows this is about setting standards early before misuse catches up.
Productivity Tools
📽️ DepthFlow - Turn static 2D images into dynamic 3D motion videos.
🗣️ Fluents - Build human-like voice agents for enterprise-scale interactions.
🖼️ Adsturbo - Create UGC-style video ads instantly from any product image.
📹️ Prism - Generate, organize, and edit short-form videos using multiple AI models.
Get featured tomorrow: How do you use AI for business/personally? Interesting stories will be shared with 100K curious readers.
Useful Resources
Why would a large enterprise put $100 million into an external AI research fund?
💰 Funding
Luminai raised $38M to scale its healthcare AI automation platform.
Rork raised $15M to build an AI-powered mobile app creation platform.
💼 Roles in AI
🐦 Scarcity creates Value
What Should Only Be Decided Once
When to use this?
When the same decisions keep resurfacing and draining leadership time.
You are my Chief of Staff.
Based on the update below, identify:
Decisions we keep revisiting that should be locked
Why they keep coming back (unclear criteria, missing owner, risk avoidance)
The single best decision to lock before year end
How to document and communicate it so it stays closed in Q1
Keep it practical and under 150 words.
Update: [paste recurring debates, approvals, or leadership discussions]P.S. Get more such prompts in the Prompting Playbook (free for you)
Stay curious, {{first_name | leaders}}
PS. If you missed yesterday’s issue, you can find it here.