Build it

Automating business workflows with Python (and where AI actually fits)

July 27, 2026 · 7 min read

The most valuable AI skill in my toolbox isn't prompting a chatbot. It's using AI to write Python scripts that then run my workflows without AI — free, instant, and identical every single time. If your data is consistent, this is almost certainly the most efficient automation available to you.

Why Python, and why now

Python has been the workhorse of business automation for two decades: it reads spreadsheets, moves files, fills forms, talks to websites, and builds reports. What changed is who can wield it. A year ago I knew next to nothing about coding. But modern AI models are exceptional at writing Python — so the barrier moved from "learn to program" to "learn to describe your work precisely." The second skill is much closer to what you already do every day. (It's also learnable.)

The efficiency argument: one build, zero tolls

There are two ways to put AI on a repetitive task:

• The token treadmill: call an AI model every time the task runs. Every run costs money, takes seconds, and can vary — because models are probabilistic by nature.
• The script route: use AI once, to write a Python script that encodes the rules. Every run after that costs nothing, finishes in milliseconds, and produces the same answer every time.

When the data is consistent — same file layout, same fields, same rules — the script route wins on every axis: cost, speed, and reliability. This is the no-token-tax principle, and "consistent data" is precisely the condition that makes it safe.

Two real examples from my own week

Forms and orders. I used to fill the same web forms and order screens by hand — identical steps, different values, week after week. Now a script does it: it reads the values from where they already live and fills the forms without me. What used to be an evening of careful typing is a button press with zero transcription errors.

The report that used to crash. One of my recurring reports lived in a massive spreadsheet so dense with formulas that updating it would routinely crash the file. The fix wasn't a better spreadsheet — it was retiring the formula forest entirely. A Python script now reads the source data, does all the computation in code, and writes out the finished report in seconds. Same numbers, same layout, none of the drama. Both of these, and more, are on the Proof page.

How to spot a script-ready workflow

Walk through your week and look for tasks with these fingerprints:

• You could train a new hire with a checklist. If the instructions fit on a page of "then click, then copy, then paste," a script can follow the same page.
• The input arrives in the same shape every time. Same spreadsheet columns, same file naming, same portal. Consistency in, determinism out.
• You do it more than weekly. Frequency is what turns a small saving into a compounding one.
• Errors are expensive but the steps are dumb. Careful-but-boring is exactly the work computers were born for.

How the build actually goes

The process, honestly described: you explain the task to an AI coding assistant the way you'd explain it to that new hire. It writes a script. You run it on real examples. It breaks — something about your data you forgot to mention. You tell the AI what happened; it fixes the script. Three or four rounds of this and the script handles everything you actually see. The AI supplies the syntax; you supply the judgment about what "correct" means — which is why the person who knows the work is the right person to direct the build, not a distant developer.

And when a workflow isn't perfectly consistent — when there's messy human text in the middle — you don't abandon the pattern. You keep the scripted spine and add a single AI call at the messy joint. That's the hybrid design, and it's the subject of its own post.

Learning to program sounds daunting. But you're not learning to program — you're learning to direct a very fast programmer who never gets tired. The judgment stays yours; only the typing left the building.