Claude Code subagents explained #claude #chatgpt #aiagents #anthropic #ai #openai #aiexplained

4 Просмотры

  • Спасибо! Поделитесь с друзьями!

    Вам не понравилось видео. Спасибо за то что поделились своим мнением!

aii
Издатель
A head chef doesn't chop every onion herself, and neither should your AI agent. Here's what Claude Code subagents actually are, why they exist, how to build one in a single markdown file, and the surprising part: Anthropic's own research says they're not the right tool for most coding tasks.

Every claim is sourced on screen:
→ Claude Code docs: "Create custom subagents"
→ Claude Agent SDK docs: "Subagents in the SDK"
→ Anthropic Engineering: "Effective context engineering for AI agents" (Sep 2025)
→ Anthropic Engineering: "How we built our multi-agent research system" (Jun 2025)
→ Zach Wills, zachwills.net + the Hacker News discussion of his post

⏱ Chapters
0:00 The head chef analogy
0:23 What we'll cover
0:36 What a subagent actually is
0:58 The problem: context pollution
1:18 The trick: compress before returning
1:38 What a subagent can (and can't) see
2:01 Building one: a single markdown file
2:23 Anthropic's example: code-reviewer + test-runner
2:42 Running subagents in parallel
2:58 How Claude picks a subagent (Task → Agent)
3:21 The twist: 90.2% better, ~15× the tokens
3:46 Mistakes developers report
4:03 3 rules for when to use one
4:22 Back to the kitchen

More in the AI Careers series:
Harness Engineering: [link]
Loop Engineering: [link]
MCP Explained: [link]

Категория
ChatGPT и AI Чаты
Выполните вход или зарегистрируйтесь для отправки комментария.
Комментариев нет.