Automating your business with small vibe-coded tools and scripts
Read in : Français

Contents
- Which tasks are worth automating?
- Typical examples for freelancers and small businesses
- Which format: Apps Script, Python, an extension or a no-code tool?
- Google Apps Script is the best place to start
- A Python script for files on your computer
- A browser extension for repeated gestures on the web
- n8n or Make when you need to connect several services
- A copy-paste prompt for your first script
- How do you test a script safely?
- Scheduling: who runs the script, and when?
- The traps to avoid
- Quick decision tree
- Frequently asked questions
- Frequently asked questions
- Sources
To automate your business with AI, you rarely need big software: most of the gains come from small scripts of 50 to 200 lines that do one thing well. Chasing unpaid invoices, renaming 300 files, copying form answers into a clean spreadsheet: you describe the task, the AI writes the script, you test it on fake data, then you let it run. Here is how to pick the right format, the prompt to use and the limits to respect.
Which tasks are worth automating?
A task deserves a script when it comes back at least weekly, always follows the same steps and handles structured data (a spreadsheet, files, emails with a predictable format). If you have to think every time, automation mostly automates your mistakes.
The quick test is three questions:
- Can you describe the task in five steps or fewer, with no "it depends"?
- Do the inputs always have the same shape (columns, file names, email subject)?
- Can you check the result in under a minute?
Three yeses: good candidate. One no: standardize the task first (a file template, a form), then automate it. The pillar article on 25 real vibe coding use cases has more ideas, grouped by type of business.
Typical examples for freelancers and small businesses
- Every Monday, generate a summary of pending quotes from a Google Sheet.
- Send a reminder to clients whose invoice is more than 30 days old.
- Rename and file supplier invoice PDFs by date and supplier name.
- Pull orders from an inbox into a tracking sheet.
- Add a browser button that copies a customer record in the right format.
Which format: Apps Script, Python, an extension or a no-code tool?
The right format depends on where your data lives. If everything is in Google Workspace, Apps Script is the shortest path; for files on your computer, a Python script; for an action inside a web page, a browser extension.
| Your need | Suggested format | Where it runs | Watch out for |
|---|---|---|---|
| Google Sheets, Gmail, Calendar, Drive | Google Apps Script | Google's servers | Daily quotas, 6 min per run |
| Local files (PDF, CSV, images) | Python script | Your computer | Install Python once |
| Repeated action in a web page | Chrome extension | Your browser | Keep permissions narrow |
| Connecting several online services | n8n or Make | Cloud or server | Usage-based cost, visual logic |
| Scheduled task on a code repository | GitHub Actions | GitHub's servers | 5-minute minimum interval |
Google Apps Script is the best place to start
Apps Script is Google's platform for automating Workspace: you write JavaScript, there is "nothing to install", the editor runs in your browser and scripts execute on Google's servers, according to the official Apps Script overview. Open your spreadsheet, go to Extensions, Apps Script, and paste the code the AI wrote.
Time-driven triggers can run a function as often as every minute or as rarely as once a month. The exact time can drift: for a 9 a.m. trigger, Google picks a time between 9 and 10 a.m. and keeps it from day to day, per the installable triggers documentation.
Custom functions (a =MYFUNCTION() formula in a cell) have two more limits: they must return within 30 seconds and cannot call services that need authorization, such as Gmail, per the custom functions documentation. To send emails, use a custom menu instead.
A Python script for files on your computer
Python is the right choice when the data sits on your disk: renaming files, merging CSVs, pulling text out of PDFs. A coding agent like Claude Code can write the script, run it in your folder and fix the errors it sees. The Claude Code documentation describes it as a tool that reads your code, edits files and runs commands, in the terminal, an IDE, a desktop app or the browser. To get going, follow the Claude Code beginner's guide.
A browser extension for repeated gestures on the web
If you copy the same details from one site to another ten times a day, a small Chrome extension can do it in one click. That deserves its own tutorial: build a Chrome extension with AI.
n8n or Make when you need to connect several services
To chain "new form entry, then CRM record, then Slack message", a visual automation tool like n8n or Make saves you from writing and maintaining authentication code for every service. n8n can also be self-hosted: without a license key it runs as the free Community edition, according to the n8n hosting docs. Vibe coding keeps the edge for custom logic that no ready-made module covers.
A copy-paste prompt for your first script
A good automation prompt describes the input data, the expected result, the edge cases and how to test. Here is a template for Apps Script:
You are a careful Google Apps Script developer.
Context: I have a Google Sheet "Invoices" with columns
A: Client, B: Email, C: Amount, D: Issue date (YYYY-MM-DD),
E: Status ("Paid" or "Pending"), F: Last reminder (date or empty).
Goal: every Monday morning, send a polite reminder email to clients
whose status is "Pending" and whose issue date is more than 30 days
ago, unless a reminder was sent less than 7 days ago. After sending,
write today's date in column F.
Constraints:
- Test mode by default: a TEST_MODE = true constant that writes to
a "Log" sheet instead of sending emails.
- Skip incomplete rows and note them in the Log.
- No more than 50 emails per run.
- Comment every function in plain English.
Deliver: the full code, the steps to create the weekly trigger,
and 5 rows of fake data to test with.
This prompt follows the 5-step method for vibe coding prompts: context, goal, constraints, output format, test.
How do you test a script safely?
Always test on a copy of the data, never on the live file. The routine is five steps:
- Duplicate the spreadsheet or folder and work on the copy.
- Run the script in test mode and read the log line by line.
- Add awkward cases on purpose: empty cell, wrong date format, duplicate row.
- Switch to live mode on two or three rows only, with your own email as the recipient.
- Turn on the trigger, then check the log after the first scheduled run.
When a script crashes, paste the exact error message into the AI chat, along with the line it points to. If the AI goes in circles after three attempts, go back to a version that worked: that is one of the 12 mistakes that sink vibe coding projects.
Scheduling: who runs the script, and when?
A useful script runs without you, but you need to know where it runs and what happens when it fails.
- Apps Script: time-driven trigger, running under the account that created it. Failures arrive by email and in the Executions tab.
- Local Python script: Task Scheduler on Windows,
cronon Mac or Linux. It only runs while the computer is on. - GitHub Actions: the
scheduleevent runs at most every 5 minutes, can be delayed under heavy load and is disabled in a public repository after 60 days without activity, per the GitHub Actions events documentation.
The traps to avoid
Most failed automations break for boring reasons, not technical ones.
- Credentials in plain text. An API key or password pasted into a script ends up in a screenshot or a public repo one day. Use Script Properties (Apps Script) or a
.envfile kept out of anything you share. - The script nobody understands. Ask for comments and a short "how to use" note: in six months, you will be that nobody.
- Automating a fuzzy process. If the rule changes with the client's mood, keep a human approval step.
- Personal data. Client emails flow through the script: read only the columns you need and never paste real data into the AI chat, use fake examples.
- Forgotten quotas. 100 recipients a day on free Gmail runs out fast for a newsletter: wrong tool for that job.
Quick decision tree
- Your data lives in Google Sheets or Gmail: Apps Script.
- Your data is files on your disk: Python with a coding agent.
- You repeat a gesture in a web page: browser extension.
- You connect three or more online services: n8n or Make.
- The task needs judgment every time: automate the preparation only, keep the decision.
To go further and turn a spreadsheet into a full app, read replacing a spreadsheet with a real internal tool. On a zero budget, the guide to free vibe coding lists what you can do without paying.
Frequently asked questions
Frequently asked questions
Do I need to know how to code to automate my business with AI?
Not to write the script, partly to check it. The AI generates Apps Script or Python code from a precise description. Your job is to describe the inputs, the expected result and the edge cases, then test on a copy. Being able to read an error message and find the line it refers to is enough for most small office scripts.
Is Apps Script free?
Apps Script comes with a Google account at no extra subscription. Consumer accounts have quotas: 6 minutes per execution, 90 minutes of trigger runtime per day and 100 email recipients per day, according to Google's official quotas page. Google Workspace accounts get higher limits, such as 1,500 email recipients per day. Google says these quotas can change without notice.
Is a script better than a tool like n8n or Make?
A visual tool like n8n or Make is easier for connecting several services that already have ready-made modules: forms, CRM, messaging. A vibe-coded script is better for custom logic, local files or processing no module covers. Many small businesses combine both: the visual tool orchestrates, a small script does the specific calculation.
How do I stop an automated script from doing something stupid?
Ask the AI for a test mode that writes to a log instead of acting, work on a copy of the data, cap the number of actions per run and start with your own email as the recipient. Only turn on scheduling after one live run you checked by hand. Keep a working version before every change.


