AI agent: a simple definition
An AI agent gets a goal, not just a question. To reach it, it works through steps: it reads, it searches, it picks the next step and it uses your tools. At its core is a language model, the kind that runs ChatGPT, Claude, Gemini or Mistral. You give it instructions, documents and access.
Take a quote request that comes in by email. The agent reads it, finds your rates, checks your availability in the calendar and drafts a quote. You read it over, fix it if needed, and send it.
Access to your tools is what makes the difference. A language model on its own writes text. An agent uses it to move a task forward.
AI agent or chatbot: the difference
A chatbot answers questions. An agent does something with the answer: it books an appointment, updates a customer record or follows up on a quote that got no reply.
| In practice | Assistant (chatbot) | AI agent |
|---|---|---|
| What it does | It answers from your content. | It answers, then it acts in your tools. |
| An example | “How late are you open on Saturday?” | “I’d like an appointment on Thursday.” It checks the calendar, suggests a time slot and books it. |
| What it’s connected to | Your website, your PDFs, your usual answers. | On top of that: your calendar, your CRM, your email or your support tickets. |
| Built by us | CHF 3,000 | CHF 4,000–6,000 |
Starting with the assistant is often the right order. The first few weeks of conversations show which actions come up often enough to be worth handing to an agent.
Examples of AI agents in business
- Handling inquiries. A request comes in through your website or by email. The agent creates the record in your CRM, replies with the right information, and lets you know when a person needs to take over.
- Appointments. It offers open time slots, books one, then sends the confirmation and the reminder.
- Quotes. From a request, it pulls your rates and your terms, and prepares a draft for you to review.
- Invoices and forms. It reads supplier invoices, pulls out the amount, the due date and the supplier, and prepares the entry in your accounting software.
- Sales prep. Before a meeting, it gathers what’s public about the company you’re about to see and turns it into a one-page brief.
- Your team’s questions. It finds a procedure, a contract clause or the price of a part in your files, and it cites the document the answer comes from.
An accounting firm, a property management firm, a construction company and a law firm don’t have the same tasks. Yet the starting point is the same: a task that comes back every week and follows roughly the same steps each time. You’ll find more examples, industry by industry, on our page about AI in small businesses.
An AI agent for a small business: where to start
A small business doesn’t need an agent platform. It needs one specific task done well. Here’s the order that works most often.
- Pick one task. It comes back every week, it follows known rules, and nobody likes doing it.
- Gather what the agent needs to know. Your rates, your terms, your usual answers and your procedures. If none of it is written down, that’s the first job.
- Set the limits. What it can do on its own, what it needs approval for, and what it must never say.
- Test on real cases. Real requests from your customers, before anything launches.
- Measure. The time saved each week, and the answers that needed fixing.
If you’re torn between several tasks, the AI and automation audit ranks them for you, with the time saved in hours and in francs. It costs CHF 1,000, and it’s credited in full if you start an AI project within 30 days.
Which AI agent to choose
There are three ways to get an AI agent. They work well together: many businesses use the first one every day, and hand one specific task to the second or the third.
The agents in off-the-shelf assistants
ChatGPT, Claude, Gemini, Microsoft Copilot and Mistral all have agent features: searching the web, reading your files, writing, and sometimes acting in your email or your calendar. They’re the right choice for everyone’s day-to-day work. To choose between them, the most useful test is often the tools you already have: Copilot lives in Microsoft 365, Gemini in Google Workspace. The subscriptions are compared on our page ChatGPT for business.
Agents built with an automation tool
n8n and Make let you build an agent that works in the background. It receives an email or a form, uses an AI model to understand it, then acts in your software. It’s the right choice for a task that comes up often and touches several tools. See AI automation for small businesses.
An agent built for your business
When the agent talks to your customers, works from your documents and acts in your tools within precise limits, it has to be built and tested for you. That’s our job: an assistant for CHF 3,000, a workflow agent for CHF 4,000 to 6,000.
Free or paid AI agents
The free versions of ChatGPT, Claude, Gemini and Copilot are enough to try things out and to see what an agent can do. For business use, their limits show up fast.
- Your data. The rules on how your data is used aren’t the same in the free plans and in the business plans. Read them before you put a client file in.
- Your tools. An agent that acts in your CRM or in the team calendar needs access, rules and testing. That work isn’t free anywhere.
- Control. For your own use, a mistake is easy to spot and fix. In front of a customer, you need written limits, testing and someone who reads over the conversations.
In a small business, most of what an agent costs is the work of preparing and testing it.
How far can an AI agent run on its own?
An agent can work on its own. It starts when an email arrives, at night or every Monday morning, without anyone launching it. But autonomous doesn’t mean unchecked.
We set the autonomy task by task. Filing an email or creating a customer record can happen on its own. Sending a quote, answering a complaint or touching a payment goes through an approval. Every action is written to a log, and an alert goes out when a step fails.
An agent gets more freedom once its results have been checked on real cases.
The risks of an AI agent, and how to limit them
| The risk | What we do |
|---|---|
| It makes up an answer. | It answers only from your content, and it says when it doesn’t know. Before launch, an assistant is tested on 50 real questions. |
| It goes one step too far. | It only has the permissions its task needs, and anything that commits your business goes through you. |
| A malicious text, in an email or on a web page, tries to hijack it. | Limited permissions and an approval step for sensitive actions shrink what such a text could get it to do. |
| Your data ends up where it shouldn’t. | The AI provider and the settings are chosen based on your data. A public FAQ and client files don’t call for the same precautions. |
| You depend on us. | The accounts, the knowledge base and the automations are in your name, and documented. |
What an AI agent costs
With us, an assistant that answers from your content costs CHF 3,000. A workflow agent, connected to your tools and working through several steps, costs CHF 4,000 to 6,000. Ongoing care for an assistant then costs CHF 300 a month.
CalculatorCalculate the price of your AI agentPick your options: the price shows up right away.In Switzerland, published entry prices for an agent or a chatbot range from CHF 450 to 60,000, with a median of CHF 2,850. The top of the range is for installations on the company’s own servers. For the monthly running cost, the published median is CHF 250. Published prices, surveyed in November 2026, mostly before VAT. Ours are final, with no VAT on top.
Our AI agent cost page details, step by step, what an assistant’s price covers, along with the options.
When another provider is a better fit
For an agent that has to run on your own servers, handle thousands of requests a day or plug into heavy enterprise software, a large agency or a systems integrator is better placed. They have the teams and the infrastructure for it. We’ll tell you so in our very first conversation.