Autonomy Models

Who this is for. Everyone. See the Roles and Access.

Autonomy is set for each function an Agent Co-worker runs. The AP Helpdesk function runs under one of two models, chosen on the AP Helpdesk tab of Define capabilities.


AP Helpdesk: two models

An AP Helpdesk agent runs under one of two autonomy models. The model is fixed for that agent: the two are alternatives, not settings that combine.

Governed Autonomy lets the agent complete work unattended, within bounds you set. You write response guidance rather than fixed text, and the agent composes each reply from it. How much it may complete is decided case by case, by a mandatory write checklist.

Collaborative Mode requires a person to review and approve before the agent executes. Replies come from stored templates chosen by intent, so the exact wording is fixed and predictable.

Governed Autonomy is the current model and the one new agents use. Collaborative Mode is being deprecated, and the choice appears only for organizations already running it.

This article compares the two, then explains what bounds a Governed Autonomy agent and why two similar invoices can be treated differently.

The autonomy level choice, showing Governed Autonomy recommended and Collaborative Mode marked legacy.

Figure 1. The two models side by side, with Collaborative Mode marked legacy.


How they differ

Table 1. The two models

Governed Autonomy

Collaborative Mode

How a reply is produced

The agent composes each reply from guidance you write.

The agent fills a stored template chosen by intent.

What you configure

What the reply should do: what to lead with, how to handle each item raised.

The template text, its variables, and its attachments.

Where wording is set

Response guidance for the category.

The Email Template panel of the Agent Replies group.

Capability sections

A smaller set.

A larger set, including per-intent templates and orchestration.


What this means day to day

Table 2. Practical consequences

If you want to

Under Governed Autonomy

Under Collaborative Mode

Change how a reply reads

Edit the response guidance for the category.

Edit the template for that intent, and repeat for every other intent that shares the wording.

Add a new kind of request

Add the category and the guidance.

Add the intent, then build its templates.

Predict the exact wording

You cannot. The agent composes each reply from the guidance; where the autonomy level and rules allow, it sends without a person.

You can. The template is fixed text with variables.

Handle an unusual message

The guidance still applies; the agent adapts.

No template matches, so nothing is drafted.

That last row is the substantive difference. Collaborative Mode produces predictable wording from a fixed set of templates; Governed Autonomy composes each reply, and you bound it with guidance, the autonomy level and rules rather than by pre-approving each one.


Which to choose

Governed Autonomy is the current model and the one new agents should use. Collaborative Mode exists for organizations already running it. The two models expose different capability sections on the Define capabilities step. This choice applies to the AP Helpdesk function only; it has no bearing on how the agent handles invoices.


Related information