Build an AI agent business with n8n + Claude and charge $1,000–$5,000 per project

Before opening n8n, sketch the workflow on paper or in Notion. Every node is a decision: what comes in, what judgment gets made, what goes out? A simple lead qualifier looks like: email arrives → Claude reads and scores it 1–10 → if score > 7, add to Salesforce and ping Slack → if score < 7, send polite auto-reply.

Use Claude to stress-test your logic: *"Here is my proposed workflow: [paste it]. What are the 5 edge cases this won't handle? What data inputs am I missing?"*

A workflow you can explain in 3 sentences is deployable. One that requires 10 minutes of explanation is a maintenance burden. Keep your first build to 5–7 nodes maximum.

Create a free trial at [n8n](https://n8n.io). n8n Cloud gives you a hosted instance in minutes — no servers to manage. The free tier is enough to build and demo your first agent; upgrade to Starter ($20/mo) when you have a paying client covering the cost.

In n8n, add an Anthropic Claude node via the credentials panel. You'll need an API key from [console.anthropic.com](https://console.anthropic.com) — the free tier gives $5 in credits, enough for hundreds of test runs.

Test your Claude connection with a simple one-node workflow first: send 'Hello' and confirm the reply comes back. Spending 20 minutes on a working hello-world before building the full agent saves hours of debugging later.

Build the full workflow in n8n's canvas. Drag nodes from the sidebar, connect them, and configure each. For a lead qualifier: Webhook trigger (receives form data) → Claude node (structured prompt returning JSON: score, pain point, recommended next action) → IF node (routes by score) → Slack node (pings sales team) + Gmail node (sends auto-reply to low-score leads).

Prompt engineering is most of the work. Use Claude to write your own Claude prompt: *"Write a system prompt for an AI that reads sales inquiry emails. Output JSON: lead_score (1–10), primary_pain_point, recommended_action (fast-track/standard/nurture), one_line_reasoning. Be ruthless about score accuracy."*

Test with at least 10 real (or realistic fake) inputs before calling it done. Real edge cases only appear when real data runs through the workflow.

A 2-minute Loom screen recording is your most powerful sales tool. Show the agent working live: paste in a test lead, watch it get scored and routed in real time. Buyers need to see it working before they'll pay for a build.

Use Claude to write your outreach pitch: *"Write a 3-paragraph outreach message for an AI agent consultant reaching out to [niche] businesses. Lead with the time or money the automation saves. End with a specific low-friction CTA. Under 150 words."*

Your pricing anchor: quote the time saved first. "This agent handles what your team currently spends 12 hours/week doing manually — I build it once for $1,800 and it runs forever." The ROI sells the project; the pitch just opens the door.

Find your first 10 prospects on [Upwork](https://www.upwork.com) — search for 'automation consultant', 'Zapier specialist', or 'Make.com developer'. These buyers already understand automation value and have budget allocated. Alternatively, LinkedIn: search '[niche] operations manager' or '[niche] founder' and look for profiles with recent posts about scaling or hiring.

Your message: lead with the specific time sink you solve, not your service name. *"I noticed [company] has a [role] open for handling [task] — I build AI agents that automate this for under $2K. Here's a 2-minute demo of one I built last week. Happy to show you a version for your workflow."*

Send 20 outreach messages over 3 days. Expect 2–5 replies. One qualified demo call closes at 50% if the demo is clean and you've done your homework on their setup.

After your first paid build, do two things before the next project: (1) Write a 1-page 'how it works' doc for the client — what each node does, what to do if X breaks, how to contact you. This is your trust asset and your support surface. (2) Save the n8n workflow as a JSON template. Your second build in the same niche takes 2 hours instead of 4.

The retainer pitch: after delivery, propose a maintenance retainer: *"Happy to monitor the automation, fix anything that breaks as your tools update, and add small improvements each month — $300–$500/mo."* Most clients take it. Automations do break when third-party APIs change, and clients don't want to manage n8n themselves.

Five retainer clients at $400/mo = $2,000/mo baseline before any new projects. Specialize as you build — 'AI automation for real estate agencies' or 'lead qualification agents for B2B SaaS' commands higher rates than 'general AI automation'.