Connecting actions to auto assist

Letting a procedure do things, not only propose words. This is where auto assist starts saving real time.

Letting a procedure do things, not only propose words. This is where auto assist starts saving real time.

Connect one

1. Make sure the action exists and works on its own.
2. Open the procedure.
3. Reference the action at the step where it belongs.
4. Say what to do with the result, and what to do if it fails.
5. Test both outcomes.

What changes

Without actions, a procedure proposes text and the agent goes elsewhere to look things up. With them, the lookup happens in the ticket and the proposed reply already contains the answer.

That is where the minutes are: not in the typing, in the switching between systems.

Read before write

Start with actions that fetch. The agent approves a reply containing real data, and the worst case is that no data arrives.

Actions that change something are worth adding once the fetching ones have been running cleanly.

Say what the result means

"If the order shows as delivered, ask whether it was left with a neighbour. If it shows as in transit, give the expected date." The action returns a value; the procedure has to say what each value implies.

Failures need a step

System unavailable, nothing found, ambiguous. Each needs a defined next step, and handing to a person is a good one. Without it the procedure stalls halfway with the agent unsure what was meant to happen.

The agent still approves

Including anything the action changes. That approval is what makes it acceptable to connect actions that touch orders or money.

See also

• What actions are and what an AI agent can do with them
• Creating a custom action
• Writing a procedure