Trigger
n8n starts the run from a form, schedule, video upload, Search Console export, webhook, or status change.
Ship Lean Workflow Human approval
n8n AI agent workflow.
Use n8n as the trigger and routing layer around one scoped AI judgment step.
Keep the agent focused. Keep approval where consequences exist. Let n8n own the repeatable run.
workflow/safe-agent
$ agent job
score query, draft response, classify lead
$ approval gate
ok no public action without human review
Workflow shape
Five steps. One judgment call.
Context
n8n gathers the transcript, CSV, task, URL, repo link, customer note, or brief the agent needs.
Agent task
Claude, OpenAI, Anthropic, or another model handles one judgment-heavy job: draft, classify, score, plan, or review.
Approval
A human approves anything public, expensive, customer-facing, production-facing, or brand-sensitive.
Route
n8n saves the result to Notion, GitHub, email, Slack, a CMS, or a publishing queue.
When this workflow makes sense
Use this pattern when the work repeats, needs multiple app handoffs, and has one judgment-heavy step. Good examples: content brief creation, lead triage, Search Console opportunity scoring, support summary routing, and workflow planning.
What not to do
Do not put the whole business process inside one giant agent prompt. n8n should own the run history, retries, credentials, logs, and routing. The agent should own one reasoning step.
Starter file
Download the n8n human approval workflow JSON and replace the decision stub with your AI Agent, Claude, or OpenAI node.
The same proof asset is also published on GitHub as an importable template: n8n AI lead qualification workflow with human approval.
Real example source
This public starter is based on a real lead-routing pattern: webhook intake, lead enrichment, AI qualification, structured parsing, and routing into follow-up paths. The private workflow stays private; the public file keeps the reusable shape without credentials, personal data, or business-specific routing.
Related assets