AI agents and automation in plain language: Zapier, Make and the risks

What do automation and agents mean?

An ordinary AI assistant answers when you ask. Automation and agents go a step further: they do things on your behalf. It is worth understanding the difference, because the benefits and risks are different.

• Automation is a pre-built chain: "when A happens, do B and then C". For example, when someone fills in the contact form on your website, the details are saved to a spreadsheet and you get a notification on your phone. The chain works the same way every time.
• An AI agent is given a goal and decides for itself which steps to take. For example: "find three hotels in Tampere for next Tuesday, under a certain price, near the exhibition centre, and put together a comparison". The agent can browse the web, open pages, fill in forms and use other programs.

Automation is predictable and suits recurring routines. An agent is more flexible, but what it does needs closer checking. Small business owners are usually best off starting with automation where AI is one step among others.

Zapier and Make: automation without code

Zapier and Make are the most popular services for connecting programs without programming. They work with thousands of apps, including Gmail, Outlook, Google Sheets, Excel, web shop platforms, booking systems and social media. Both let you add an AI step that, for instance, summarises, categorises or writes a draft.

Many programs also have their own automation. Microsoft 365 has Power Automate, and many web shop and accounting programs can send reminders by themselves. Check first whether the feature you need already exists in a program you use.

What can AI agents do for you?

The major AI assistants, such as ChatGPT, Claude and Gemini, now have agent features. They can use a browser, research across several sources, fill in forms, work with files and connect to your email or calendar if you give permission. An agent usually asks for confirmation before important actions such as buying something or sending a message.

In a small business, an agent can help with things like:

• collecting competitors' price lists and services into a comparison table
• finding and summarising public tenders in a particular field
• gathering suppliers' product information into one file
• arranging a meeting time based on your calendar
• organising and renaming files into folders

The risks: what can go wrong

When AI acts on its own, mistakes have bigger consequences than in a text draft. Know the risks before giving an agent access to anything important.

• Wrong actions. The agent may misunderstand the task, pick the wrong file or send a message to the wrong person.
• Hidden instructions online. A web page or email can contain concealed text that tries to steer the agent into doing something other than what you asked. This is called prompt injection.
• Overly broad permissions. If the agent can access your entire mailbox, bank or customer register, a single mistake can leak data or cost money.
• Silent failures in automation. A chain can stop working when one of the programs updates, and you may not notice for weeks.
• Data protection. When information passes through several services, you need to know where personal data ends up.

There is more on security in security and privacy and AI scams.

Your first automation: from enquiry to draft reply

A good first automation saves time but does nothing irreversible. This example suits almost any service business with a contact form on its website.

Goal

When a customer submits an enquiry, the details are saved to a spreadsheet, AI categorises the request and writes a draft reply, and you get a notification. You read the draft and send it yourself.

Steps in Zapier or Make

1. Trigger: new form submission (for example your website form, Google Forms or Microsoft Forms).
2. Save: add a row to Google Sheets or Excel: date, name, contact details, message.
3. AI step: give the AI the message and ask for a category and draft reply using the instruction below.
4. Draft: save the reply as a draft in your email; do not send it automatically.
5. Notify: send yourself a message that a new enquiry is waiting.

Test the automation by submitting a few forms yourself with different messages. Check whether the categories are right and whether the drafts sound like you. Watch it closely for the first few weeks before you rely on it.

Let AI plan the automation with you

If you do not know where to begin, ask the AI itself. Tell it which programs you use and which tasks recur every week. It will suggest what is worth automating and write step-by-step instructions for Zapier or Make. Examples: a cancellation in the booking system notifies customers on the waiting list, site photos are saved to the right customer's folder, or a new review lands in a spreadsheet with a draft reply.

Where to go from here

Once your first automation works, think about other recurring tasks: saving invoice attachments to the right folder, a welcome email for new customers, a weekly sales report or sharing a post across several channels. Keep a list of the automations you have built so you know what is running if something stops working.

Also read AI for small business and customer service and sales. If you want to understand why AI sometimes behaves unexpectedly, see what AI cannot do.