n8n AI Agent Templates: Which One to Start With

The short answer

Start from a template when the job is a known loop: a trigger, one LLM call with strict JSON, a human check, and a write to Gmail, a sheet, or Slack. Build from scratch when the steps are unusual or you are still deciding what the agent should do. A template saves the wiring. It does not save you from editing the rules.

What you need before you import

An n8n instance, cloud or self-hosted, and an LLM API key. Most of these also need one OAuth credential, such as Gmail. Import the workflow, run the sample, then point it at real data. Leave anything that sends, posts, or spends behind an approval step until you have watched a few runs.

Inbox, leads, and content

Inbox triage labels new Gmail and drafts replies. Lead research takes domains and returns a sheet with a first line you can edit. A content repurposer turns one post into X and LinkedIn drafts and waits for approval. Those three cover the loops most solo businesses run every week.

• Inbox: label, draft, daily digest. Nothing sends itself.
• Leads: public company pages in, a sheet out.
• Content: one source post, several drafts, you approve.

Support, SEO, invoices, meetings, social

Customer support drafts a reply from your docs and sends refunds and low-confidence answers to a person. An SEO agent turns a keyword into a brief and a Markdown draft, and marks claims that need a source. Invoice extraction reads PDFs and photos into a CSV and flags bad maths. Meeting notes pull action items that quote the transcript. A social scheduler plans a week and publishes only what you approved.

Template or blank workflow

Use a template when you want the trigger, the JSON shape, the approval stop, and a sample run already connected. Build your own when no template matches the apps you use, or when the workflow is two nodes and a prompt. Copying a template and deleting half of it is still faster than a blank canvas if the hard part is the approval path.

Check the cost notes

A small model on these jobs is usually cents a day, and scraping or a vision model can cost more than the LLM. Any honest template prints token usage and says when a number is an estimate. Check that page before you leave a workflow on a schedule.

n8n, or the Python file

Each of these ships as an n8n workflow and as a Python version with an AGENTS.md file, so you can run it in n8n or hand it to Claude Code or Cursor. Pick n8n if you want the visual editor. Pick the Python file if you want it in the repo with tests.

Where to get them

ShipReadyAgents sells these as individual templates: inbox triage, lead research, content repurposer, customer support, SEO content, invoice extraction, social scheduling, and meeting notes. During launch week, code LAUNCH20 takes 20% off through Oct 21, 2026.