What agents can read and do

Before the pilot, the team agrees which sources and permissions to use, names a reviewer and sets the conditions that pause work. The design below describes what will be checked for your workflow.

Four rules for a pilot

Who can see a document

Access follows company, project and role permissions. For a pilot, agree the data owner, which fields can be used, how to store them and who can see both the original and the result.

Who can approve sending

In the demo, the agent prepares a draft for a staff member to review. Permission to send a message or update a record is agreed for the specific system.

What happens when sources disagree

The agent shows which facts conflict and where they came from, then asks the reviewer for a decision. The design includes a log of the run, source, result version and any reason it paused.

What happens at the spending limit

The spending limit is agreed before the pilot. The design calls for each new billable action at the limit to wait for a decision from an authorized person.

What people see when work stops

In the monitoring design, an administrator sees why a run stopped. The source document requires a separate permission. An employee with that access checks the material and decides whether work can continue.

Questions and answers

What is checked before using live data?

On agreed samples, we check who can open the source and the result, how approval is requested, and what happens when data is missing or the spending limit is reached.

How can we start with less data?

Choose one task and agree on an anonymized file or export. Then decide whether the workflow needs access to a live system.

Start with one repeatable workflow

Choose work you can compare before and during a pilot, such as checking estimates, reconciling delivery notes or preparing client emails.

How the pilot works