---
title: "Licensed AI assistant: when it is and isn't enough"
description: "A licensed assistant helps one person with one task. When it is enough on its own, and what more it takes to entrust a process to AI."
url: https://www.joinx.it/en/resources/licensed-assistant-what-is-missing
---

# We already have a licensed assistant: when it is enough and when it is not

A licensed assistant helps one person with one task. When it is enough on its own, and what more it takes to entrust a process to AI.

JoinX · Last updated 18 November 2026

In a management committee you sometimes hear a question like “we have already given everyone a licensed assistant, what else do we need?”. It is a fair question, and the answer depends on what you expect from the tool. A licensed assistant helps a person draft an email or summarise meeting minutes, and for tasks like these it is often enough. When you want AI to work inside a process, for example triaging complaints or the credit assessment of a loan application, you need things the licence does not include, because they depend on that company's data and rules.

## First of all, make good use of what is there

When we carry out a [Mapping](https://www.joinx.it/en/solutions/mapping) we start from the tools the company already has, and many of the first initiatives are delivered with exactly those. The licensed assistant is therefore a resource to use to the full before buying anything else, and the difficulty usually lies in the use itself. A company may give it to everyone and find that few use it, because nobody has said which activities to entrust to it and who checks the result.

With [Adoption](https://www.joinx.it/en/solutions/adoption) we often begin with a workshop with the heads of functions, to choose the activities where the tool is really useful, and with an adoption plan that says which roles change and who is accountable. After a few weeks we check with those who took part whether the chosen activities have really become part of the work.

## When the assistant is enough

If the goal is individual productivity, that is, the analyst preparing the summary of a file sooner or the back office drafting a reply more quickly, the licensed assistant is often sufficient. In these cases the task begins and ends on one person's desk, that person checks the result before using it, and the work to be done is mainly about adoption and rules of use. Building anything else would make no sense, and we advise a company in this position to stop there.

There is still Article 4 of the AI Act, which asks those who use AI systems for measures to support the AI literacy of their staff, without requiring a set level to be guaranteed for each person. It applies to a licensed assistant too, as we explain in the article on [what the AI Act asks of companies](https://www.joinx.it/en/resources/ai-act-what-it-asks-of-companies).

## What you need when the work becomes a process

An assistant helps one person with one task, whereas a process runs across several people and several systems, handles many cases a day and leads to decisions that someone is accountable for. To entrust part of it to AI you need, first of all, data with a definition and an owner. With [Data Foundation](https://www.joinx.it/en/technology/data-foundation) we establish which source each data item in the process comes from and who owns it. You then need written rules, including the exceptions that today only a few experienced colleagues know, and in the [Company Brain](https://www.joinx.it/en/technology/company-brain) we gather them so that people and agents read them in the same way.

The agents work on top of this base, each with a mandate, a level of autonomy and a test bench of real cases, as we describe on the [Agents](https://www.joinx.it/en/technology/agents) module page. Their work reaches the person who has to approve it through [Workspace](https://www.joinx.it/en/technology/workspace), with the sources alongside. Lastly, [Governance](https://www.joinx.it/en/technology/governance) keeps the run log, so the answers an auditor asks for when faced with an agent in production are already written.

## The tools nobody has inventoried

Alongside the assistant bought by the company, we often find tools that a function has switched on of its own accord, such as a service that summarises meeting minutes or an email assistant. This is what is known as shadow AI: tools already in use but outside any control. In Mapping we find it in the interviews, cross-checking what emerges against the company's list of systems and licences. Then we enter each tool in the register, with an owner and a risk level, and the work on the AI Act also starts from that inventory.

## An example in a complaints office

Take an illustrative case. In a complaints office the operators use the licensed assistant to draft replies, which they then correct. Management, however, would like complaints to reach the operators already triaged and prepared, and here the assistant alone is not enough. Triage requires the office's policy with its exceptions, access to the complaints system with precise permissions and a person to approve the replies that grant the customer a refund.

In a case like this the assistant stays where it is, for the drafts the operators write every day, while with [Deployment](https://www.joinx.it/en/solutions/deployment) we build the agent for the process, with its rules and the queue in which the operator approves or corrects.

## The time the work around the process takes

The limit lies in people's time: writing the rules of a process and agreeing the data definitions takes time from the people who know the work, and they take that time away from their day-to-day activities. The first project pays for it, while the second finds the rules and definitions ready, written to be reused, as we explain on the [Compounding Intelligence](https://www.joinx.it/en/compounding-intelligence) page.

To decide where to start, Mapping lines up the initiatives and indicates which can be delivered with the tools already there and which require a project around a process, and if an area is not worth investing in we write that in the documents, with the reasons.

[All articles](https://www.joinx.it/en/resources#articoli)

## Work out what to entrust to the assistant you already have

If few people use the licensed assistant, we start with the heads of functions from the activities where it is really useful.

[Contact us](https://www.joinx.it/en/contact?tema=adoption)
