AI Automation
AI-powered workflows that read, decide, and act — replacing the judgment calls that used to need a human in the loop.
Every business runs on repetitive judgment work: qualifying leads, routing enquiries, personalizing outreach, summarizing information. Done manually, it eats hours every week and doesn't scale past a certain volume.
Classic automation only helped when the rules were simple. The moment a task needs reading, understanding, or deciding, a rigid script breaks — so the work stays manual.
- Businesses drowning in manual processes — outreach, data handling, repetitive operations that follow a pattern but need judgment at each step.
- Founders and teams who want AI as an engine inside their operations — reading, deciding, acting — not a chatbot bolted on at the end.
AI workflows designed around one question: what is the repetitive work costing you? From there, the workflow is mapped — what the AI reads, what it decides, where a human stays in the loop — and then built with the tools that fit: n8n for orchestration, Python for the logic, OpenAI, Claude, or Gemini for the reasoning, depending on the task.
A real example: a personalized outreach pipeline that finds target companies, discovers their contact emails, researches each prospect, writes individually tailored copy with AI, and runs the whole campaign end to end.
- 01
Map the work — identify the repetitive process, what it costs, and which steps need judgment versus pure rules.
- 02
Design the workflow — what the AI reads, what it decides, what it acts on, and the human checkpoints for high-stakes steps.
- 03
Build and integrate — connect your existing tools, handle the edge cases, test against real data.
- 04
Hand over — the workflow runs on autopilot, with clear logs so you can see what it did and why.
- High-stakes outputs (money, legal, customer-facing messages) should keep a human review step — the workflow flags, the human approves.
- Workflows depend on third-party APIs and AI providers; when those change, the workflow needs maintenance. That's built into how I ship — with clear logs and isolated integration points.
- AI quality follows input quality — messy or missing data means the first milestone is usually cleaning the inputs, not adding more AI.
Which AI models do you work with?
OpenAI, Claude, and Gemini — the choice depends on the task: some workflows need strong reasoning, others need speed or cost efficiency. I benchmark against your real data rather than defaulting to one brand.
Do you build with n8n?
Yes — n8n for orchestration and visual workflows, Python for the logic that needs real engineering, and browser automation (Playwright/Puppeteer) when a service has no API. The stack follows the problem, not the other way around.
Will this replace my team?
No — it removes the repetitive part of their work. The judgment calls stay with people; the workflow handles the reading, drafting, routing, and chasing.
Have a process worth automating?
Tell me what's eating your time — I'll tell you honestly whether it's a fit and how I'd approach it.