Skip to content
Critical Mug
(Artificial intelligence)

AI agents at work on company data

An AI agent is useful when it can act on the company's real data, with clear rules on what it may do. We build assistants and agents connected to your data model, with defined permissions and human supervision where needed.

We work from Florence, with companies across Italy.

When you need it

  • The front office answers the same questions every day.
  • Getting a figure out of the management system means asking whoever knows how to run the queries.
  • Incoming mail has to be read, sorted and turned into tasks by hand.
  • You tried a generic chatbot and it gave untrue answers.

What we build

Assistants on official information

They answer customers and colleagues using the company's up-to-date data: schedules, availability, case status. Cases that need a person are handed over to a person.

Questions to your data in plain language

Management and team leads query company data by writing in plain language, and get answers built on the real figures.

Operational agents with progressive autonomy

Agents that prepare, check and propose actions. Autonomy is granted one action at a time, after seeing how they work.

Custom tools and MCP servers

Agents reach the data through tools built on your data model, with role-based permissions and a log of every operation.

How we do it

An agent is worth as much as the tools it is given. Every agent receives a set of tools built on your data model: each does one thing, with defined permissions, and leaves a trace of every call. We use the models best suited to the task and replace them when better ones come out, because the logic lives in the tools and in the data.

Further reading on luciopatone.it: Asking your management system questions in plain language

In operation

AI assistant on incoming email

(Organization with a front office)

It answers recurring requests with official information and hands over to people the cases that need them.

Frequently asked questions

Can AI make up answers?

A generic model can. An assistant built this way answers from the company's official data and, when the information is missing, says so and hands the request to a person.

Is our data used to train the models?

No. We choose services and settings that exclude training on customer data. For the most sensitive data we are experimenting with open-weight models on infrastructure under the company's control.

Where do we start?

From a repetitive, well-defined task with data already available. A first assistant goes live in two to four weeks and shows in practice what is worth automating next.

Contact

Let's talk about your company

Tell us in a few lines which process you want to improve: we will get back to you to arrange a first meeting.

By submitting the form you accept the privacy notice: we use your data only to reply to you.