Most workflow automation doesn't need to be complicated. A single Python script, a Claude API call, and a cron job will handle 80% of the repetitive tasks most small businesses deal with. This post walks through exactly how to build that — starting from scratch.

The pattern

Almost every AI automation follows the same shape:

  1. Pull data from somewhere (a file, a database, an API, an email inbox)
  2. Send it to an LLM with a clear instruction
  3. Do something with the output — save it, send it, log it

That's it. The complexity in real systems comes from handling edge cases, retries, and scheduling — not from the core logic.

Setting up

You'll need Python 3.9+ and the Anthropic SDK:

pip install anthropic

Create a .env file in your project root:

ANTHROPIC_API_KEY=sk-ant-...

Load it in your script with python-dotenv, or just set the env variable directly in your shell.

A real example: daily report summarizer

Say you run a small business and every morning you read through a CSV of the previous day's transactions to spot anything unusual. Takes 20 minutes. Here's how to automate it.

Step 1 — read the data

import csv
import os
from anthropic import Anthropic

client = Anthropic()

def load_transactions(filepath):
    with open(filepath, newline='') as f:
        reader = csv.DictReader(f)
        rows = list(reader)
    return rows

Step 2 — build the prompt

The quality of your output depends heavily on how clearly you frame the task. Be specific about what you want back.

def build_prompt(transactions):
    data = "\n".join(
        f"{row['date']} | {row['description']} | ${row['amount']}"
        for row in transactions
    )
    return f"""You are reviewing daily business transactions.
Below is a list of transactions from yesterday.

{data}

Summarize in 3–5 bullet points:
- Any unusually large or small amounts
- Patterns worth noting
- Anything that looks like it needs follow-up

Be concise. Plain English. No financial advice."""

Step 3 — call the API

def get_summary(prompt):
    message = client.messages.create(
        model="claude-haiku-4-5-20251001",
        max_tokens=512,
        messages=[{"role": "user", "content": prompt}]
    )
    return message.content[0].text

Model choice matters for cost. For daily summaries like this, claude-haiku is fast and cheap — fractions of a cent per run. Save Sonnet or Opus for tasks that need deeper reasoning.

Step 4 — do something with the output

Saving to a file is the simplest option. From there you can email it, push it to Slack, log it to a database — whatever fits your workflow.

from datetime import date

def save_report(summary):
    filename = f"reports/daily-{date.today()}.txt"
    os.makedirs("reports", exist_ok=True)
    with open(filename, "w") as f:
        f.write(summary)
    print(f"Saved: {filename}")

if __name__ == "__main__":
    rows = load_transactions("data/transactions.csv")
    prompt = build_prompt(rows)
    summary = get_summary(prompt)
    save_report(summary)
    print(summary)

Running it automatically

On Linux (or WSL2), add a cron job to run it every morning at 7am:

crontab -e

# Add this line:
0 7 * * * /usr/bin/python3 /home/user/reports/daily_summary.py >> /var/log/daily_summary.log 2>&1

On Windows, use Task Scheduler. On a server, a systemd timer works well for more control over retry behavior.

Where to go from here

This pattern scales to almost anything:

The key is keeping each automation focused on one job. Once it's running reliably, adding a second task takes minutes — you already have the scaffolding.

If you're trying to automate something specific and running into trouble, get in touch — happy to help think through it.