Coding with AI for Non-Programmers: Method, Testing and Security

AI assistants write code surprisingly well. That opened a new door for every professional: building small tools tailored to their own work (a complex formula, a macro, a script that organizes files, an automatic report) without waiting on IT. But "asking AI for code" and "having a reliable tool" aren't the same thing. The difference is the method, and that's what you'll learn today.

What you can build

You can use any assistant's chat, coding assistants built into code editors (such as GitHub Copilot or Cursor) or agents that work directly on a project's files (such as Claude Code or Codex; lesson 9). To start, the chat is enough.

Big tech teams have already embraced it. In October 2024, Google CEO Sundar Pichai said that more than a quarter of all new code at Google was being generated by AI, then reviewed and accepted by engineers. Notice the second half of that sentence: reviewed and accepted. For you, at a smaller scale, the promise is the same: less time writing code, as long as you spend time testing it.

The six-step method

1. Describe the problem with sample data

Nothing improves generated code more than showing the exact input and the output you expect. A complete prompt:

I need a Python script to flag overdue invoices.

INPUT: a CSV like this (the columns are always the same):
number,client,issue_date,credit_days,amount,paid
INV-102,Lakeview Offices,2026-08-20,30,850.50,no
INV-103,Hotel Marisol,2026-09-05,15,640.00,no

RULES: an invoice is due on issue_date + credit_days. It's overdue
if it isn't paid and that date has passed.

OUTPUT: a list of overdue invoices sorted by days late (number,
client, due date, days late and amount) and, at the end, the total
overdue per client.

CONTEXT: I'm not a programmer. I use Windows and Python 3.12. I can
only use the standard library. I want to run it like this:
python overdue.py invoices.csv, with an optional date for testing.

2. Ask for the plan before the code

"Before writing any code, explain in 5 steps how you'll solve this and what assumptions you're making." That way you catch misunderstandings while they're cheap to fix: for example, if the AI assumes credit days are business days rather than calendar days.

3. Ask for readable code and instructions to run it

Ask for clear names, short comments, small functions and exact steps to run it. A typical result, already tested:

"""List overdue invoices and the total owed by each client.

Usage: python overdue.py invoices.csv [YYYY-MM-DD]
If you don't pass a date, today's date is used.
"""
import csv
import sys
from collections import defaultdict
from datetime import date, datetime, timedelta


def read_invoices(path):
    with open(path, encoding="utf-8") as f:
        return list(csv.DictReader(f))


def overdue_invoices(invoices, today):
    overdue = []
    for inv in invoices:
        if inv["paid"].strip().lower() == "yes":
            continue
        issued = datetime.strptime(inv["issue_date"], "%Y-%m-%d").date()
        due = issued + timedelta(days=int(inv["credit_days"]))
        if due < today:
            overdue.append({**inv, "due": due, "days_late": (today - due).days})
    return sorted(overdue, key=lambda x: x["days_late"], reverse=True)


def main():
    if len(sys.argv) < 2:
        print(__doc__)
        sys.exit(1)
    today = (datetime.strptime(sys.argv[2], "%Y-%m-%d").date()
             if len(sys.argv) > 2 else date.today())
    overdue = overdue_invoices(read_invoices(sys.argv[1]), today)

    by_client = defaultdict(float)
    for inv in overdue:
        print(f"{inv['number']}  {inv['client']:<18} due {inv['due']}  "
              f"{inv['days_late']:>3} days  {float(inv['amount']):>9,.2f}")
        by_client[inv["client"]] += float(inv["amount"])

    print("\nTotal overdue by client:")
    for client, total in sorted(by_client.items(), key=lambda x: -x[1]):
        print(f"  {client:<18}{total:>10,.2f}")


if __name__ == "__main__":
    main()

4. Test with data you know

Prepare a small file where you know the right answer. With five invoices, of which you know only INV-102 and INV-103 are overdue as of October 31, 2026:

python overdue.py invoices.csv 2026-10-31

INV-102  Lakeview Offices   due 2026-09-19   42 days     850.50
INV-103  Hotel Marisol      due 2026-09-20   41 days     640.00

Total overdue by client:
  Lakeview Offices      850.50
  Hotel Marisol         640.00

It matches. Now the edge cases: what if someone typed "Y" instead of "yes" in the paid column? The script compares against "yes," so it would treat that invoice as unpaid and report it as overdue. That's the kind of error the AI can't guess, because it doesn't know your real data. Always test with: blank cells, "Y," "Yes" and "Paid" variants, dates in U.S. format (09/15/2026), amounts written as "$1,250.00" and a file with no rows.

5. Fix it with the exact error

When something fails, paste the complete error message into the chat along with what you were doing. "It doesn't work" doesn't help; this does:

Running it on the real file, I get this error:
ValueError: time data '09/15/2026' does not match format '%Y-%m-%d'
In the real file, some dates look like 09/15/2026 and others like
2026-09-15. Also, the paid column contains "yes", "Y" and "Paid", and
some amounts are written as "$1,250.00". Fix all three and tell me
what you changed.

6. Ask for an explanation and a review

Before you rely on it: "Explain what each function does as if I didn't know how to code" and "Review this code like an experienced developer: what could break with real data? Is there any security risk?" Understanding the essentials of your tool lets you maintain it when your needs change.

Security: what never to do (and what to do instead)

Is it worth it? Calculate the savings

Before you invest an afternoon in a tool, do the math: minutes per week × 52 ÷ 60 = hours per year. Three examples:

If building and testing the tool takes you 4 hours, each of these pays for itself within a few months. But count maintenance too: when the report format or the date column changes, you'll have to adjust it.

When to call a developer

Coding with AI is ideal for personal tools or tools for a small team. Get professional help when the tool handles money or customers' personal data, when many people depend on it, when it has to connect to critical company systems or when it must meet security or compliance standards. A prototype you built with AI is also the best way to show a developer exactly what you need.

def is_paid(value):
    return value.strip().lower() in ("yes", "y", "paid", "true")

def parse_date(text):
    for fmt in ("%Y-%m-%d", "%m/%d/%Y"):
        try:
            return datetime.strptime(text.strip(), fmt).date()
        except ValueError:
            pass
    raise ValueError(f"Unknown date format: {text!r}")

def parse_amount(text):
    return float(text.replace("$", "").replace(",", "").strip())

# In overdue_invoices:
#   if is_paid(inv["paid"]): continue
#   issued = parse_date(inv["issue_date"])
# And in main(), use parse_amount(inv["amount"]) instead of float(...)

Summary

• With AI you can build formulas, macros, Apps Script, Python scripts and simple web pages without being a programmer.
• The method: sample data, plan first, readable code, tests with known answers, exact errors and a review.
• Edge cases (variants like "Y," blanks, U.S. date and dollar formats) are where generated code breaks.
• Back up your data, keep keys and confidential data out of the chat and verify packages before installing.
• For critical tools, your prototype is the starting point for a developer.

Next lesson: How to Make Money with AI: Services, Pricing and Your First Clients. In the course's final lesson you'll turn everything you've learned into income: services businesses pay for, how to price them and how to land your first clients.