Automation

Automation is deterministic. AI is probabilistic. The real skill is knowing which one to use and how to combine them.

Deterministic workflows vs probabilistic models

Python, n8n, Power Automate, and more

Using AI to build and improve automation

Automation ≠ AI

Deterministic workflows vs probabilistic models

Deterministic by nature: automation runs fixed rules. Same input always produces same output. No surprises, no hallucinations.
AI is probabilistic: LLMs and ML models produce different outputs for the same input. Useful for judgment tasks, unreliable for fixed processes.
Where automation wins: data pipelines, ETL, CI/CD, deployment, monitoring, scheduled tasks, batch processing — anything with a defined success condition.
Where AI wins: classification, extraction, summarization, generation, routing decisions — anything that needs contextual judgment.
The hybrid pattern: use AI to decide WHAT to do, use automation to do IT. AI routes, automation executes. This is the most reliable architecture.

From scripts to low-code to full-stack automation

Python — most versatile: scripts, data pipelines, API integration, file processing, automation glue. My primary tool.
n8n — open-source workflow automation, visual node-based editor, connects APIs without code.
Power Automate — Microsoft ecosystem: Office 365, Teams, SharePoint, Outlook automation.
Shell / Bash — file operations, system admin, deployment scripts on Linux/macOS.
CI/CD pipelines — automated test, build, deploy via GitHub Actions, Jenkins, or custom tooling.
Playwright / Selenium — browser automation for testing, scraping, and form-filling tasks.
SQL / ETL — automated data extraction, transformation, and loading between systems.

Stack

Python n8n Power Automate Bash / Shell GitHub Actions Playwright SQL Docker REST API Webhook Selenium

Using AI to build and improve automation

AI writes the script: use LLMs to generate automation scripts from natural language descriptions. Five minutes instead of an hour.
AI handles the edge case: when a data format changes unexpectedly, AI can classify and reroute instead of breaking the pipeline.
AI monitors automation: log analysis, anomaly detection, failure classification — AI watches the automation so you don't have to.
Human judgment + machine execution: the most effective pattern. Let AI decide, let automation execute, let humans supervise.

Approach

LLM-generated scripts AI + Deterministic hybrid PDCA Cycle Human-in-the-Loop Log analysis Anomaly detection

Automation runs on infrastructure that stays reliable.