ARTIFICIAL INTELLIGENCE · CYBER-MED
An AI agent needs clear rules.
Starting with a specific task makes it easier to assess value, limitations and responsibilities.
Choose a process
A starting point could be searching company documents, preparing a draft or organising requests. The objective should be described in observable terms.
Define the sources
The assistant must work with authorised information and provide references that can be checked. If documents are incomplete or outdated, the output may be too.
Limit the actions
Reading, proposing and modifying a system are different activities. The agent’s permissions must reflect its assigned role and the checks required by the process.
Measure and review
Evaluation considers quality, errors and usefulness in real work. Traceability and supervision help improve the process and clarify who makes decisions.
Further reading
NIST — AI Risk Management Framework
For general information. Activities and controls must be assessed in their specific context.
The topic in 35 seconds
Silent educational video with Italian on-screen text. Italian and English subtitles and transcript are available.
Read the transcript
- 00:00–00:07 The task comes first, then the AI agent. A useful project starts with a task and clear rules.
- 00:07–00:14 A draft to prepare. For example, organise a request and propose a response for a person to check.
- 00:14–00:21 Which documents can it read? Authorised, up-to-date information with references the team can verify.
- 00:21–00:28 Who authorises sending? Drafting and sending are different actions. Define permissions and review steps.
- 00:28–00:35 Check the result. Assess quality, errors and usefulness. Let’s choose the first process: info@cyber-med.it
View the carousel: 5 cards
