2024/early 2025 = ChatGPT wrappers. Now it’s Claude Code wrappers. Everyone’s playbook is how fast you can spin up a wrapper with the Claude Agent SDK and launch it. Claude Cowork is a prime example of that cycle in motion. 🔁

Question for Boris Cherny: In Claude Code, how do you decide between • background sub-agents (agentic async) • background bash tool calls (process-level async) Do you think of them as different layers in the same system, or solving fundamentally different problems?

🤔 Deep dive into Claude Code plugins architecture & multi-agent workflows: I love Claude Code plugins for packaging skills, commands, hooks & MCP servers together - it's elegant & powerful. BUT there's a fundamental limitation I've been wrestling with 🧵 The Problem:

🔥 Just discovered skills.sh and it's giving pip/npm energy for AI agents One command to install skills into Claude Code: npx skills add owner/repo --skill my-skill Why it's a game changer: ⚡ 50-60% fewer tokens vs manual install 🎯 CLI does all the

🧵 If you're tackling big workflows with Claude Code, here's a good orchestration pattern: Break your project into artifacts where each can be completed in ONE session. 📁 The setup: → Define the sequence of artifacts upfront → Each artifact lives in the same project folder

Imagine: Claude Code encounters a task that CLI and MCP servers can't handle. It uses browser automation to complete it, then generates a reusable Playwright script and saves it as a skill. Next time? No AI tokens needed. Just runs the script. One-time learning →

Why software platforms will increasingly prioritize 🤖 agent access 👇 🔍 Distribution AI agents become the new recommendation engine. “Use Notion” vs “Use Asana” will be decided by what the agent can actually operate. 🧲 Stickiness Automated workflows = massive switching

Feature request for Anthropic Boris Cherny The Claude Code workflow I want: 1. LOCAL DEV (Claude Code) Build an agent/workflow in my terminal Test it, iterate, get it working perfectly 2. EXPORT TO PRODUCTION (Claude Agent SDK) (one command)

the Vercel skills.sh CLI is genius because it can own discovery. If it indexes skills across GitHub/GitLab (and eventually the open web via SKILL.md + metadata), “how to do X” stops being blog search and becomes “run find; install; agent does it.”

We need a universal package manager for MCP servers like what we have now for Skills with Vercel npx skills.

🧵 We're entering the age of "skill folderification" of documentation. The primary reader of your docs is no longer a human browsing a website. It's an AI agent grepping through files. Notion, Confluence, - all optimized for human browsing. Rich formatting, sidebars, search

Hey Guillermo Rauch - every Next.js page already exports metadata: title, description, canonical URL, OG tags. That's literally YAML frontmatter waiting to happen. Right now when AI agents need website content, they all do the same wasteful dance: 🔄 Hit the URL → Firecrawl

💡 Pro tip for AI agent power users: Store your API keys as OS-level environment variables, then write a shared skill script that reads from them. Result? One setup, works across: → Claude Code → Gemini CLI → Cursor → Codex → Any future agent No re-entering keys per tool.

Interesting move from OpenAI Claude Agent SDK = "Here's a powerful agentic loop library. Host it yourself, sandbox it yourself, expose it yourself." OpenAI Responses API = "Here's a powerful agentic loop API endpoint. We host it, sandbox it, and expose it. Just call us."

Every knowledge worker will have this. The "PA" is the new "PC." Just like everyone got a personal computer in the 90s, everyone gets a personal assistant from this year onwards 🤞

Managed Agents feature request — the hosting infra is exactly what I needed. Sessions, containers, async execution, file persistence. But here's the gap: I've built a production agent locally in Claude Code using skills with context: fork, allowed-tools, hooks, subagents with