Two approaches
There are two common ways to add conversational AI to business work. The first treats the assistant as a translator: it understands a request, then calls out to your existing systems through their APIs, often without a permission-aware record of its own.
The second treats conversation as one interface onto the same application — the assistant and the app read and write the exact same record under the same permissions. [prompt.exc] takes the second approach. A leave request filed in chat is the same leave record a manager opens in the HR app.
Why the difference matters
When conversation and application are separate, permissions must be rebuilt for the conversational layer, and people lose context moving between the two: a request made in chat may look different, or be missing, in the system of record.
Sharing one record removes both problems. Permissions and context are enforced once, in the application, not maintained twice.
What it looks like for a real request
An employee asks for two days off in chat. Their balance and the team’s policy are checked, then the request is filed — reserving the days, not booking them. The manager opens the HR app and sees the identical record, with the same scheduling-conflict flag. Whoever acts next, there is one record to update.
Related