MC Soluzioni Informatiche · Procedure AI

Your company's procedures, carried out on every computer

An agent that doesn't advise: it acts. It carries out the procedures written with you and approved by you, and outside those it touches nothing — it calls a person. One engine; for each department the catalogue of procedures changes.

How we workThe verticalisations

YOUR DEPARTMENTS HR Finance Purchasing Production yours ONE ENGINE ON EVERY COMPUTER approved
30years inside our clients' departments, writing the software that makes them work
5languages understood without translating the catalogue, measured on 15 October 2026
1.3sto understand a request and pick the right procedure
0scripts not approved by a person: a guarantee of construction, not a promise

An assistant answers. An agent acts.

Almost everything called “AI in the enterprise” today answers questions: it reads documents, summarises, suggests. Then the thing still has to be done, and a person does it.

Here the order is reversed. On every person's computer runs an agent that knows how to do a precise list of things, because we wrote those things together with you and you approved them one by one.

Understands

The request arrives as it comes: badly written, vague, in another language.

Chooses

Among your approved procedures, the one that matches. None matches? It says so.

Runs it

Checks first, actions in order, consent wherever it touches someone's data.

Writes it down

What it checked, what it did, how it went. Then it closes.

Outside the approved procedures the agent does nothing: it hands the case to a person, with what it has already checked written down.

The model does not decide what to do on the computer

It only decides which of your procedures matches the request. That is the difference that counts when you have to explain it to whoever handles security, and when it has to be written down in the documents the AI Act asks for: you do not have to prove what a model will do, because the model has nothing to invent.

A person decides

  • What the agent may do, procedure by procedure
  • The exact text of every script, read and approved
  • Who authorises, and where the request stops and waits
  • Everything that takes judgement: appraisals, exceptions, negotiations

The agent runs it, on every computer

  • Understand the request
  • Choose the procedure
  • Run the checks
  • Execute in order
  • Ask for consent
  • Verify the outcome
  • Write down what happened
  • Call a person

Thirty years of processes, not a bet

In thirty years we have written hundreds of applications for dozens of companies, and to write them we went inside their departments: finance, HR, purchasing, production, quality, warehouse. Many of those processes we have already automated.

What changes now is not whether a process gets automated: it is who runs it. No longer a program running overnight on a server, but an agent on the computer of the person doing the work, which understands the request and starts the right procedure.

The departments, one by one

Where the intelligence runs

All of it sits on Private AI, which is why it can be done in companies that cannot send data outside: the model runs inside your perimeter, or in a private cloud reserved for you.

Where it sits is a choice, and the choice is yours — not a line in somebody else's terms of use.

On 15 October 2026 we measured understanding against a catalogue written in Italian only, with requests in Italian, English, French, German and Spanish that were not in the catalogue: 38 out of 38, 1.3 seconds on average. It understands meaning; it does not compare words.

You start from a process, not from a platform

Tell us about a process that repeats and that steals someone's time today. We'll tell you whether it can be written as a procedure, how much it weighs, and which parts stay in a person's hands anyway. If it isn't worth it, we say so straight away.

Talk to usHow we work