Viewpoint
What 250 production workflows taught me about n8n.
n8n is easy to start and easy to abandon. The workflows that lasted all shared the same unglamorous habits.
Notes from legal, finance, SMB operations, and enterprise IT deployments.
02 / What I build
Workflows that earn their place.
Lead and sales ops
Inbound enrichment, qualification, CRM sync, slack alerts, multi-channel follow-up with AI drafting.
Support and ticketing
Triage with Claude, auto-tag, suggested replies, escalation rules with human approval.
Document workflows
PDF and contract extraction, redlining, signature flows, archival to your data lake.
Reporting and finance
Multi-system reconciliation, recurring reports, AI summaries, anomaly alerts.
AI agent orchestration
n8n as the control plane for Claude and OpenAI agents with tool use, retries, and approval gates.
Migrations
Zapier and Make to n8n. Same outcomes, lower bill, more control, no rate limits.
03 / Production discipline
The boring parts are the point.
Idempotency keys on every external call. Dead-letter queues on every webhook. Retries with exponential backoff. Execution history retained. Alerts wired to Slack and PagerDuty. Cost dashboards per workflow. You see what runs, what fails, and what it costs.
04 / AI in the loop
Claude does the thinking. n8n does the orchestration.
Claude (Anthropic) for reasoning, structured extraction, and drafting. OpenAI when needed. MCP for tool exposure. pgvector for retrieval. Human approval steps for anything that touches money, customers, or compliance.
05 / Proof
Years of n8n workflows running real operations.
I have shipped n8n inside legal, finance, SMB ops, and enterprise IT. Completed 9 Anthropic Academy courses. Lovable Champion and community leader.
FAQ
Common questions.
- Why n8n and not Zapier or Make?
- n8n is self-hostable, code-friendly, has proper error handling, and does not bill you per step. For production AI workflows with retries and queues, it wins.
- Do you build with AI inside n8n?
- Yes. Claude and OpenAI nodes, MCP tool use, RAG with pgvector, and human-in-the-loop approval steps. The boring infrastructure is what makes the AI behave.
- Self-hosted or n8n Cloud?
- Both. I deploy to Fly, Railway, or your own infra when data sensitivity demands it. n8n Cloud when speed matters more than control.
- Can you migrate from Zapier or Make?
- Yes. Most migrations cut cost by 60 to 80 percent and remove rate limits that block scale.
- How do you handle failures?
- Retries with backoff, dead-letter queues, idempotency keys, alerting to Slack or PagerDuty, and full execution history. Workflows that fail loudly, never silently.
- How does handover work?
- A runbook, recorded walkthroughs, and a week of pairing while the team runs it. If nobody can operate it without me, it is not finished.
- Where does the paid work live?
- On TechTide AI. This page is the field notes, not an offer.
Automations only fix so much.
When the workflows are fine but the data underneath them lives in Airtable bases, the fix is the Airtable Rescue Decision Sprint at TechTide AI. When the AI layer is the fragile part, it is the Systems Decision Audit.