AI agents for businesses: real work, with a person who approves

An AI agent is not a chat window that answers questions. It is a program that does tasks inside your systems: it reads an email, finds the data, drafts the reply or the record and leaves it ready for someone on your team to approve.

1What an AI agent is in a company

An agent combines a language model with specific tools: read access to an inbox, permission to create a draft in the CRM, a query to your management software. With those it can carry a task from start to finish, rather than just suggesting text.

The difference from classic automation is that the agent understands documents and messages written any old way. The difference from a chat assistant is that it acts, which is why you have to decide carefully what it may do on its own and what it may not.

2Which tasks fit and which do not

Agents fit well with repetitive tasks that are passed from hand to hand today:

  • Sorting incoming email and preparing the reply or task it calls for.
  • Reading invoices, delivery notes or tender documents and leaving the data ready for review.
  • Keeping the CRM up to date from emails and meetings.
  • Preparing periodic reports with data from several systems.

3What we do not let it do alone

An agent should not pay, sign, delete data or send anything to a customer without a person approving it. Those decisions stay with your team, and the system is designed to make approving quick, not to skip it.

Nor do we recommend agents when the task is rare, when a mistake is very costly or when a fixed rule does the job just as well. In those cases we will tell you.

4How we put it into production

Each agent has permissions limited to what it needs and nothing more. Every action is logged: what it read, what it proposed, who approved it and when. If something goes wrong, it can be reconstructed step by step.

We start with a single task, measured before and after, with a person reviewing everything it does. Only when the result is reliable do we extend it to the next one. By default data is processed on servers in the European Union, and we put that in writing.

5Agents already at work

  • Certia Group: A team of AI agents inside an engineering firm

    Certia Group

    A team of AI agents inside an engineering firm

    For Certia Group, an infrastructure engineering firm, we built a team of AI agents into its daily work, handling email, the CRM and reports, with a person approving every step that matters. In use.

    In use

    See the project page

  • Kiero: An AI employee on the business’s WhatsApp

    Kiero

    An AI employee on the business’s WhatsApp

    Our own product: an AI employee in the business’s WhatsApp that replies with the catalogue and opening hours, books appointments and takes payment in the chat. When a person is needed, it hands over. Live at kiero.es.

    Live at kiero.es

    See the project page

See all projects

6Frequently asked questions

Does the agent replace someone on my team?

It takes repetitive work off their hands, but the decisions that matter are still approved by a person. That is how we design it from the outset.

What happens if the agent gets it wrong?

That is why a person reviews before anything leaves the company, and every action is logged so it can be corrected and the agent adjusted.

Where is my data?

By default, on servers in the European Union. Which data the agent sees and which providers are involved is set out in writing in the proposal.

Does it work with the software I already use?

Usually yes: the agent works on your email, your CRM or your management software. In the diagnosis we see what can be connected and how.

How is it different from a WhatsApp assistant?

A WhatsApp assistant talks to your customers in a conversation. An agent works behind the scenes, on your company’s systems, and prepares tasks for your team to approve. Sometimes a project includes both.

Tell us which task keeps repeating

Describe the work that passes from hand to hand. We will tell you whether an agent is the right tool or whether something simpler will do.

Tell us your problem