Human, delegated or autonomous: how to log what each AI does

As soon as an AI agent can create tasks or update opportunities, an inevitable question arises: who made this change? In a law firm, an advisory firm or a sales team, the answer has legal and trust implications.

Three types of action

• Human: done by a person in Unelia and logged under their name.
• Delegated to an AI: an assistant acts on a person's behalf, never with more permissions than they have.
• Autonomous: an agent works on its own within the permissions of its role.

Why the distinction matters

A delegated action carries the responsibility of the person who requested it. An autonomous one belongs to the agent and to whoever configured its role. Lumping both together as "the AI did it" makes auditing impossible later on.

Traceability doesn't slow down AI adoption: it's what lets you adopt it without fear.

What gets logged

12:04  Delegated AI  Claude on behalf of Marta
       Created the task "Send proposal" · Permission: Sales
10:15  Blocked    Sorting agent
       Tried to change an amount · No permission

Each entry includes the actor, the action, the permission used and, if it was blocked, the reason. Blocked actions are as valuable as allowed ones: they show that the limits work.