Anthropic’s Building Effective Agents appendix highlights two proven patterns—support agents that pull data and take actions like refunds, and coding agents that iterate against tests—with clear success metrics and human review.
Official n8n AI example: a chat agent answers with GPT-4; if unsure it calls a sub-workflow that asks for the user's email and posts the unanswered question to a Slack support channel.
Indie Hackers guide: Jotform intake → Zapier → Claude/ChatGPT triages priority and drafts a reply → a control form lets you edit → a second Zap finishes CRM/notify actions—only high-priority items pause for review.