Michael Cohen

Working on APIs for Claude Managed Agents @AnthropicAI. Probably wrong a bunch. Views are my own.

[Michael Cohen @_hi_mc]

@trq212 Suzanne's on of my favorite people at Anthropic (besides @trq212)

[Michael Cohen @_hi_mc]

ant CLI unblocks so many amazing things. put it in your CI pipeline to create or update stateful resources like Agents and Environments and Skills as part of your regular deploy process.

hackable, modular, and Claude-friendly. a winning trio

[ClaudeDevs @ClaudeDevs]

We’ve added a CLI for Claude Platform to make every API endpoint runnable from your terminal.

Call the Messages API, stand up Claude Managed Agents, pipe results straight into your shell.

The ant CLI is well understood by coding agents (Claude Code) using the claude-api skill.

[Katelyn Lesse @katelyn_lesse]

there's a ton of demand right now for excellent infra talent, definitely outpacing supply

im hiring engineers who have tackled megascale, multicloud product infra to work on the Claude Platform. help me find them 🙂

[Michael Cohen @_hi_mc]

i've been using workflows in claude code for a bit now and it's completely transformed how i claude. it's an extremely amazing new capability in Claude Code and i highly encourage folks to go try it out.

the outputs that come out on the other side are amazing. happy clauding friends

[ClaudeDevs @ClaudeDevs]

New in Claude Code (research preview): dynamic workflows.

Claude writes an orchestration script on the fly, then spins up a large fleet of coordinated subagents in parallel to take on your most complex tasks.

Use the word "workflow" in a prompt to get started.

[Rod Howarth @rodhowarth]

“Next, we will work with critical partners—including US and allied governments—to expand Project Glasswing to additional partners. And in the near future, once we’ve developed the far stronger safeguards we need, we look forward to making Mythos-class models available through a general release.”

[Michael Cohen @_hi_mc]

@yusuke_m_MU we can get you model switching mid-session too! @gcemaj what do you think?

[Michael Cohen @_hi_mc]

@rodhowarth i'm ngl for a second i thought you were talking about the chair

[Ivan Burazin @ivanburazin]

Perfect product brief for what sandboxes solve. You shouldn't have to babysit a computer to keep an agent alive. It should run remotely so that you can pick up where you left off on your phone.
https://t.co/RHz4d4Sw6u

[Michael Cohen @_hi_mc]

oh and shoutout to the insanely talented @gcemaj for shipping both and being more obsessed with devx than i am

[Michael Cohen @_hi_mc]

two nice quality of life improvements we shipped yesterday. these didn't get as much attention as self hosted sandboxes and MCP tunnels but regardless make me happy:

1. you can now update the list of tools and MCP servers that a session has in the middle of a session. helps a ton with giving Claude elevated capabilities in the middle of a session without having to start from scratch
2. large tool results now get 'offloaded' into a file in the container in order to prevent context windows from blowing up

we want Claude Managed Agents to be the best possible platform to build on top of, and small improvements like this are gonna keep coming over the next few days, weeks, months, years.

more info in our docs: https://t.co/ATZjGFPuA2

[Michael Cohen @_hi_mc]

I want to uplevel the level of demos and cookbooks we have to offer for Claude Managed Agents. Managed Agents work everywhere, in any piece of software that can make API calls. put them in your CI/CD pipelines, as code reviewers on your PRs, as always on chiefs of staff, or as nodes in your complex workflows.

the lil chatbot app demos I've been presenting are cute but are the most basic "hello world" examples out there. what kinds of really out there demos should we show off next?

[Yusuke Murata | AI Agent: Shirabe | Claude Code @yusuke_m_MU]

MCP Tunnel、病院向けAIエージェント導入でかなり重要だと思う

✅従来のMCP
通常、ClaudeなどのAIエージェントから院内のリモートMCPサーバーを使うには、院内側に「外部からアクセスできる入口」を作る必要がある

たとえば、公開URL、VPN、IP制限、Cloudflare Tunnelなど

この入口設計が面倒くさい

院内DB、文書サーバー、部門システムにつながるMCPサーバーを、どう安全に外部から使わせるかが問題になる


✅MCP Tunnel
院内MCPサーバー側から、.tunnel.anthropic.com に向けて安全な通信路を張る

Claudeエージェントは、その通信路を通って院内MCPサーバーにアクセスする

つまり、院内ネットワーク側に外部公開用の入口を開けなくてよくなる

楽やん

院内サーバーをパブリックネットワークに晒さず、既存の境界防御を保ったままAIエージェント連携できる

MCP Tunnelを使えば、組織限定のMCPを作る際に、セキュリティレイヤーが一段上がる

[Michael Cohen @_hi_mc]

so excited for more opportunities like this for platforms to interplay more seamlessly.

go try out self hosted sandboxes from Vercel while using Claude Managed Agents!

[Vercel Developers @vercel_dev]

You can now run Claude Managed Agents with Vercel Sandbox.

Sessions run on the same Firecracker microVM architecture that powers 1B+ Vercel deployments, with millisecond start times and credentials brokered at the firewall. https://t.co/HXTaOAKbo1

[Michael Cohen @_hi_mc]

@ivanburazin mc (and anthropic) ❤️ daytona

[Michael Cohen @_hi_mc]

@oikon48

[Michael Cohen @_hi_mc]

oh also massive shoutout to @MikeNomitch_CF for this incredible writeup on how CMA hooks naturally into @Cloudflare

https://t.co/qizrtrpTgI

[Guillermo Rauch @rauchg]

Claude Managed Agents 🤝 Vercel Sandbox
https://t.co/pW6iGH63MG

[Michael Cohen @_hi_mc]

@ivanburazin see my tweet below the first one!

[Michael Cohen @_hi_mc]

self hosting sandboxes has been our #1 ask from users since we launched Claude Managed Agents in April. there's a 1001 reasons why you may want to host your own sandboxing infra, so this was one of those things that was super fun to build. the team worked super hard to make sure we got things working in a way that both helps you plug-n-play your existing infra while also making sure the dev x "just makes sense".

super super excited to see what everyone cooks up with their own sandboxing infra, and please continue giving us feedback on what's missing so we can go and build it for you.

massive shoutout to @CloudflareDev , @daytonaio, @modal, and @vercel for being amazing partners on this project. thank you for building such amazing platforms.

[ClaudeDevs @ClaudeDevs]

We’ve added two security improvements to Claude Managed Agents.

Self-hosted sandboxes keep the agent’s execution environment in your infrastructure or with a managed sandbox provider.

MCP tunnels let the agent connect to services inside your security perimeter.

[Cloudflare Developers @CloudflareDev]

Cloudflare Tunnels 🤝 @AnthropicAI Claude Managed Agents

[Gabe @gcemaj]

The first thing folks asked when we released CMA was when can I connect this to my Modal/Daytona/Cloudflare/Vercel sandboxes? The answer is today!