As AI Executes More Service Work, Capability Has to Shift Upstream

When AI takes over routine service steps, I would not start by asking what Agentforce training employees need. I would start by asking what work is left for them to own.

Air India is expanding Salesforce Agentforce beyond refund processing into broader customer-service workflows, including multi-intent email resolution, passenger name changes and knowledge support. Salesforce reports that refund turnaround fell from about 14 days to four hours and name-change processing from about three days to 30 minutes. Those are company and vendor reported results, so I would not treat them as independent proof. They are still useful for seeing how the work may change.

If the system handles more of the drafting, validation, routing and routine completion, the employee is left with more exceptions and more judgment. Someone has to recognize when the answer is wrong, know when policy requires escalation and decide when to stop the automation and take over. That requires domain knowledge, but it also requires authority. A course can build knowledge and practice. It cannot repair an escalation path or give someone decision rights their role does not have.

Microsoft has since published another enterprise example that points in the same direction. Its cloud supply-chain organization says it simplified workflows before deploying more than 111 agents across planning, sourcing, fulfillment and logistics. Microsoft reports average planning cycle time falling from about 10 business days to less than 2.5 across five monthly cycles. In a separate nine-person product team, people worked with agents in roles Microsoft described as meta-engineers, meta-designers and meta-PMs.

That does not prove Air India has redesigned its service jobs, and Microsoft's numbers are also internally reported. What it does show is why I would look at work design before building an AI training plan. Once agents execute more of the standard process, the important human capability often moves toward exception handling, verification, escalation and accountability.

For any similar deployment, I would ask for the exception data, override rules and escalation paths. I would want to see which decisions employees still own, how often the system gets those cases wrong and what good human intervention looks like. That tells me far more about the capability requirement than a feature list.

Some old training may disappear because the system now performs those steps. Other practice may become more important because employees are left with the cases automation cannot safely finish. The role, authority and performance standard should be redesigned first. Training should follow that design.