35 n8n Workflow Ideas and Examples for 2026 (Including AI Agent Workflows)

18 classic n8n automations, 10 AI agent workflows you can build inside n8n, and 7 jobs that run better on a dedicated AI agent. With the nodes to use, a decision framework, and current n8n pricing. Updated November 2026.

I was 47 nodes deep in an n8n workflow when I realized something was wrong.

The workflow was supposed to triage customer support emails. Read the email. Classify the priority. Route it to the right team. Auto-respond if it's a known question. Simple enough on paper.

But "classify the priority" meant understanding context. Is "our system is down" a P1 or a P3? Depends which system. Depends on the customer. Depends on whether it's actually down or they forgot their password. I'd built that judgment out of IF nodes and keyword matches, and every week a new edge case broke it.

Here's what's changed since then: n8n can now do that judgment step itself. Its AI Agent node drops a model with tools into any workflow, and its new Agents feature builds standalone agents with memory and channels. So the question isn't "can n8n do AI?" any more. It's where each kind of work belongs.

That's what this list is for. n8n has 13,820 community templates (9,574 of them AI templates), and companies like Delivery Hero have saved 200 hours a month with a single IT ops workflow. The platform works. Below are 35 ideas, split three ways:

• 18 classic workflows that need no AI at all
• 10 AI agent workflows you can build inside n8n
• 7 jobs that run better on a dedicated, always-on AI agent

The 18 classic n8n workflow examples that need no AI

These share three traits: predictable inputs, fixed rules and the same output every time. If you can draw it as a flowchart with no "it depends" branches, n8n handles it cheaply and reliably, without paying for a single model token.

Data movement and sync

1. New Stripe payment → Slack message + Google Sheets row. Stripe Trigger, Set node to format, Slack and Google Sheets nodes. Five nodes. Done.
2. GitHub PR merged → deploy to staging + notify the team. GitHub Trigger, HTTP Request to your deploy API, Slack or Discord node.
3. Form submission → CRM contact + welcome email. Typeform or Jotform trigger, HubSpot node to create the contact, SendGrid or Gmail node for the welcome.
4. Inventory below threshold → purchase order + supplier email. Schedule Trigger, database query, IF node on stock level, generate the PO, email the supplier.
5. New blog post → social cross-posting. RSS Feed Trigger, format per network, post to LinkedIn, X and Facebook.

Notifications and monitoring

6. Server health check → PagerDuty alert. Schedule Trigger, HTTP Request to the health endpoint, IF non-200, PagerDuty node.
7. New Google or Yelp review → Slack notification. Poll the reviews API, compare with the last run, post anything new.
8. SSL certificate expiring → email warning 30 days out. Schedule Trigger, check the certificate date, compare with today, send if inside the window.
9. Failed CI/CD run → Slack message with the error log. GitHub Actions webhook, extract the failing step and log, post a formatted message.
10. Competitor pricing change → spreadsheet log. Scrape the pricing page, diff against the last check, log changes and alert past a threshold. (If you want someone to explain what changed, that's idea #31.)

Data processing

11. CSV upload → cleaned data → database insert. Watch a folder, parse, validate fields, transform, insert into PostgreSQL.
12. Daily report → PDF → email to stakeholders. Schedule Trigger, query several sources, fill a template, generate the PDF, send.
13. Invoice PDF → OCR → accounting system. Email trigger, extract fields, map them, push to QuickBooks or Xero. (For messy invoices, idea #27 swaps OCR rules for a model.)

Internal operations

14. Employee onboarding → accounts + access. HR system trigger, create accounts in five tools, assign permissions, send the welcome pack.
15. Meeting booked → Zoom link → attendees notified. Calendar trigger, create the Zoom meeting, update the invite, notify.
16. Expense receipt → amount extracted → Slack approval. Upload trigger, extract the amount, categorise by rules, approve with a Slack button.
17. Weekly backup → compress → S3 → verify. Schedule Trigger, dump the database, compress, upload, verify the checksum, report success or failure.
18. Document approval → route by department → track status. Form trigger, Switch node on department, track approvals, notify on completion.

For more on choosing a visual builder, our n8n vs Make for AI agents and n8n vs Make vs Zapier guides weigh them up.

10 AI agent workflows you can build inside n8n

Now the judgment step. n8n gives you two ways to add it, and both are worth knowing before you pick an idea from this section.

A couple of cost notes: on n8n Cloud, one turn with an agent counts as one execution, and tool calls to your workflows don't count separately. Model tokens go to your own provider key, not to n8n.

Here are 10 workflows that put the model where it earns its keep.

19. Support email triage. Gmail Trigger → AI Agent node with a "look up customer" tool (your CRM) and a "search help docs" tool → Switch node routes P1 to PagerDuty, P2 to the support queue, known questions to an auto-reply for human review. The model classifies; the workflow keeps control of what happens next. Ready-made version: Email Triage.
20. Meeting notes → decisions and action items. Zoom or Google Meet transcript webhook → AI Agent node with a structured output parser (decisions, owners, due dates) → create tasks in Linear or Asana → post the summary to Slack. See also Meeting Notes.
21. Inbound lead enrichment and scoring. Form trigger → HTTP Request to an enrichment API → AI Agent node scores fit against your ideal customer profile with a short reason → HubSpot update → Slack ping for hot leads. Compare with Lead Qualification.
22. Docs Q&A bot for your team. Slack or Telegram trigger → AI Agent node with a vector store tool over your docs and Simple Memory (or Postgres memory) for follow-ups → reply in thread. Add a "not sure" path that tags a human.
23. Alert triage for engineering. Error webhook (Sentry, PostHog or cloud logs) → AI Agent node groups duplicates, guesses the likely cause from recent deploys and checks past tickets → opens one Linear issue instead of fifty. Ready-made: PostHog errors to Linear and GCP error log triage.
24. AI search visibility tracker. Weekly Schedule Trigger → ask several AI assistants the questions your buyers ask ("best tool for X") through their APIs → AI Agent node checks whether your brand is mentioned, how it's described and which competitors appear → log to Google Sheets and post the week's changes to Slack. A good fit for marketing teams watching AI Overviews and chat answers. Related: Brand Mention Monitor.
25. Ecommerce order-status agent. Website chat or Telegram trigger → AI Agent node with tools for "look up order" (Shopify), "check shipment" (carrier API) and "start return" (behind an approval step) → answer in plain language. Keep refunds and cancellations as fixed workflow steps, not model decisions.
26. Client onboarding assistant. New client in your CRM → AI Agent node drafts a personalised welcome email and kickoff agenda from the signed proposal → creates the project in your PM tool → schedules a check-in. A human approves the email before it sends.
27. Messy invoice extraction. Email with attachment → AI Agent node with native PDF passthrough (supported since n8n 2.28) pulls supplier, line items and totals into a fixed JSON schema → validate totals with a Code node → push to Xero. Rules handle the clean invoices; the model handles the ugly ones. See Invoice & Expense.
28. Personal task bot (GTD style). Telegram trigger → AI Agent node turns "remind me to send the deck to Priya after Thursday's call" into a task with a due date and context → Todoist or Notion → daily digest each morning.

How to build your first AI workflow in n8n (6 steps):

1. Add a trigger (Gmail, Webhook, Slack, Schedule, whatever starts the job).
2. Add an AI Agent node and attach a chat model sub-node (OpenAI, Anthropic, Gemini, Mistral, Groq and others are available; n8n also added MiniMax, Kimi and Vertex AI credentials this year).
3. Attach at least one tool: another n8n node, a sub-workflow, an HTTP request or an MCP client.
4. Add memory if the agent needs to remember the conversation (Simple Memory to start, Postgres or Redis in production).
5. Use a structured output parser when the next node needs fixed fields, and route on those fields with an IF or Switch node.
6. Put a human approval step (Slack or email button) in front of anything that sends, deletes or spends.

If you'd rather describe it than wire it, n8n's AI Assistant can draft the workflow or agent for you, and you refine it from there. If you're exposing n8n tools to Claude through MCP, our MCP Server Trigger fix covers the most common error.

n8n use cases by team

If you landed here searching "n8n use cases", this is the same list cut by department.

More by role on our use cases page.

7 jobs that run better on a dedicated AI agent

Everything above runs because something triggers it. These seven are different: they need an agent that's always there, remembers context across days and conversations, and lives where people already talk to it. You can build them in n8n, especially with the new Agents feature, but you'll be designing, hosting and maintaining an agent rather than a workflow. That's where a dedicated agent platform saves real time.

29. Customer support conversations, not just triage. Triage (#19) sorts the inbox. A support agent holds the whole conversation: checks the customer's history, asks a follow-up question, remembers what they said last week, and escalates with a summary. Start with Customer Support Triage.
30. Follow-ups that know the relationship. "They opened the proposal twice, mentioned a competitor in a support ticket and went quiet." An agent connects those signals over days and drafts the nudge at the right moment. See Follow-Up Reminder and Gmail follow-up automation.
31. Competitor monitoring that explains itself. Idea #10 logs that a price changed. An agent reads the new page, the changelog and the launch post, and tells you in Slack what changed and why it matters. See Competitor Monitor.
32. A morning briefing that adapts. Calendar, inbox, Slack mentions and yesterday's metrics, summarised and prioritised every morning, with the agent remembering what you told it to ignore. See Morning Briefing.
33. Content moderation with nuance. "I'm dying" might be a medical emergency or "I'm dying laughing". An agent that sees the thread, the user's history and your community rules makes a better call, and keeps learning from your overrides.
34. Sales call analysis into the CRM. The agent reads the call, pulls out buying signals, objections and next steps, updates CRM fields and reminds the rep before the next meeting, with memory of earlier calls on the same deal.
35. Email drafting in your voice. Reads the thread, checks the last few messages with that person, matches your tone and leaves a draft for you to send. Our guide to AI agent email automation goes deeper.

The decision framework (when to use which)

The rule: if you can draw it with no "it depends" branches, it's a workflow. If one step depends on understanding context, tone or intent, add an AI Agent node. If the whole job is a conversation or an ongoing responsibility, it's an agent.

The hybrid that power users run: n8n handles triggers and data movement. When a step needs judgment, n8n calls the agent through a webhook or API, and the agent does the thinking. It works the other way too: with BetterClaw's n8n workflow skill, an agent can trigger and monitor your existing n8n workflows from a chat message. Our BetterClaw vs n8n breakdown covers the feature-by-feature differences.

Why the approval step matters. In February 2026, Meta AI alignment director Summer Yue asked an OpenClaw agent to suggest which emails to archive and told it not to act without her say-so. Her real inbox was so large that the agent compacted its context, lost that instruction, and deleted more than 200 emails while she typed "stop" from her phone. The fix isn't "don't use agents". It's making approval part of the system, not a sentence in a prompt.

The honest take

n8n isn't competing with AI agents any more; it's become one of the ways to build them. That's good news. For anything triggered and structured, n8n is still one of the best and cheapest tools around, and the AI Agent node makes the judgment steps far less brittle than my 47-node keyword maze.

The trade-off is ownership. In n8n you design the agent, pick and pay for the model, host it (or pay for Cloud), wire up its channels and maintain it as the platform changes. n8n's own Agents feature is still in Preview, and n8n says itself that agents watching an event stream continuously and full multi-agent orchestration aren't there yet.

If that's a fun project, build it in n8n. If you'd rather have the agent side ready on day one, that's what BetterClaw is for.

Where BetterClaw fits

BetterClaw is a no-code AI agent builder for the always-on half of the list:

• Agents that live where your team talks: Telegram, Slack, Discord and Gmail, plus webhooks so n8n can hand off to them.
• 500+ connectors and ready-made workflows you switch on instead of wiring.
• Memory across conversations, so follow-ups and briefings get better over time.
• Your models, your key: bring Claude, GPT, Gemini, OpenRouter or local models, with zero markup on inference.
• Projects: several agents working together alongside your team.
• Guardrails built in: approval levels for risky actions, an instant kill switch, sandboxed execution, AES-256 encryption and secrets purged from agent memory after 5 minutes.

Plans start at $19/month for Starter, with Pro at $49/month for 5 agents ($39/month billed annually) and Business at $149/month for 25 agents. Every plan starts with a 7-day trial.

Frequently Asked Questions

What are the best n8n workflow ideas for 2026?

For classic automation: payment notifications (Stripe → Slack → Sheets), employee onboarding across several tools, data pipelines (CSV → clean → database) and monitoring alerts (health check → PagerDuty). For AI workflows: support email triage, meeting notes into tasks, lead scoring, docs Q&A bots and AI search visibility tracking, all built with n8n's AI Agent node. For ongoing, conversational jobs like follow-ups and support conversations, a dedicated AI agent is usually simpler.

What are some n8n workflow examples for beginners?

Start with something that has one trigger and two or three actions: a new form submission that creates a CRM contact and sends a welcome email, a Stripe payment that posts to Slack, or a weekly schedule that emails a report. Then add an AI Agent node to one step, like classifying incoming emails, once the basic flow works.

Can n8n handle AI-powered workflows?

Yes. The AI Agent node adds a chat model plus tools to any workflow, with optional memory and structured output, and works with OpenAI, Anthropic, Gemini, Mistral, Groq and other providers. Since version 2.32.3, n8n also has Agents (in Preview): standalone agents with memory, knowledge files, MCP tools, sub-agents, Slack/Telegram/Linear channels and schedules. There are 9,574 AI templates in n8n's library.

What's the difference between the n8n AI Agent node and n8n Agents?

The AI Agent node is one decision-making step inside a workflow you design; the workflow decides when the model runs and what happens next. n8n Agents are standalone agents built in their own Agent Builder, reachable by chat, channels, schedules or the Message an Agent node, and they choose their own next step among the tools you attach. Agents are still in Preview.

How many n8n workflow templates are there?

n8n's template library lists 13,820 workflow templates, including 9,574 AI templates. Check the node versions before copying older AI templates: from n8n 1.82.0 the AI Agent node's agent type setting is deprecated and every AI Agent node works as a tools agent.

How does n8n compare to an AI agent like BetterClaw?

They overlap but lean different ways. n8n is a workflow automation platform that now also builds agents; you design, host and maintain them. BetterClaw is a managed AI agent platform: agents that live in Telegram, Slack, Discord and Gmail, remember context, connect to 500+ apps and run around the clock without you hosting anything. Many teams use n8n for structured workflows and hand the judgment steps to an agent.

How much does n8n cost compared to BetterClaw?

n8n's pricing page lists Cloud Starter at €20/month billed annually (2,500 executions), Pro at €50/month billed annually (10,000 executions), Business at €667/month (self-hosted) and custom Enterprise pricing; the self-hosted Community Edition is free if you run your own server. Model costs go to your own provider either way. BetterClaw starts at $19/month for Starter, $49/month for Pro and $149/month for Business, with zero markup on your own model keys.

Is it safe to automate business processes with AI agents?

With guardrails, yes. Keep human approval on anything that sends, deletes or spends, give agents the narrowest tools they need, and make sure you can stop them instantly. The February 2026 incident where an OpenClaw agent deleted a Meta researcher's emails happened because the "ask first" rule lived only in the prompt and was lost during compaction. BetterClaw builds approvals, a kill switch and sandboxed execution into the platform instead.

Can n8n and an AI agent work together?

Yes, and that's often the best setup. n8n triggers the workflow and moves the data; when a step needs judgment, it calls the agent through a webhook or API. In the other direction, BetterClaw's n8n workflow skill lets an agent start and monitor your n8n workflows from a chat message.

Related Reading

• BetterClaw vs n8n: Feature-by-feature differences
• n8n Alternative: Managed AI Agents: The architectural differences in depth
• n8n vs Make vs Zapier for AI Agents: Three automation tools, one question
• Best AI Automation Tools in 2026: The wider landscape
• AI Agent Frameworks: Code-first and no-code options mapped out
• n8n MCP Server Trigger: Tools Rejected by Claude: The fix