AI workflow automation
Automation that ships work, not surprises — with humans on the writing side until evals prove otherwise, and a log for every decision.
#What I build
One workflow, end-to-end
Intake → classify → route → act, in your accounts, on n8n/Make or code-level LangGraph.
Lead capture → qualify → route
Every lead caught, scored, routed, and followed up in minutes — nothing slips.
Data pipelines / ETL
Validation, idempotency, and alerting so the report your business runs on is correct and fresh.
Integrations
Make the CRM, billing, support desk, and spreadsheet talk — with error handling and an audit trail.
#n8n, Make, or code — how I choose
The platform is a means, not a religion. I pick the lightest tool that fits the workflow’s complexity and who has to maintain it:
| Reach for | When | Trade-off |
|---|---|---|
| Make / Zapier | Simple, linear glue between SaaS apps your team already pays for | Fast to build; gets brittle past a few branches |
| n8n (self-host) | Branching logic, your own data, or you want to own the runtime | More power + control; you host and maintain it |
| Code (LangGraph / typed) | AI in the loop, real state, retries, approval gates, tests | Most robust + testable; needs an engineer to change |
#The design principle
The most reliable automation I’ve shipped sends zero AI-written messages — it reads, classifies, routes, and alerts a human. Human approval points are placed where the risk is, and earned down with evidence, never assumed away. That design deletes entire classes of failure.