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.
Make’s Feb 2026 AI Agent docs list concrete agent scenarios—support ticket triage, sales outreach, content marketing, market research, candidate screening, SEO reports—plus when to use agents vs fixed scenarios.
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.
Claire Vo moved ChatPRD's Holly Helpdesk to Grok Bot: it sweeps Intercom and email on an hourly routine from a playbook, drafts replies, and sends refund requests as human approval cards that open a single Stripe action—never auto-refunds.
User's flight was cancelled by Muse; refund is stuck at Duffel (Muse's booking provider). Muse says it's with Duffel; Duffel has no customer contact—highlights agent travel cancel/refund handoff risk.