Salesforce Agentforce: AI Agents Move Into CRM and Customer Workflows

Salesforce's Agentforce https://www.salesforce.com/agentforce/what-is-agentforce/ is the enterprise AI agent deployment that is probably closest to what most large organisations will actually experience as their first production AI agent rollout.

The digital labour framing being provocative is intentional. Positioning AI agents as doing work rather than assisting with work changes the evaluation from capability question, what can it do, to deployment question, what should it do, who authorises it, and what happens when it makes a mistake.

The customer-facing workflows being the initial deployment context is the right place to start for this evaluation. Sales lead qualification, customer support triage, onboarding sequences, and follow-up automation are high-volume, relatively structured processes where the AI agent can be constrained to a well-defined task set and where the cost of errors is real but recoverable. That is different from deploying agents in financial decision-making or sensitive internal processes where errors have different consequences.

The human escalation path being designed into the customer-facing workflow is the architecture that determines whether Agentforce is a useful automation tool or a liability. An agent that can identify when it is in a situation outside its competence and escalate gracefully is a different product from one that handles everything with equal confidence.

Which customer workflow should be automated first and what requirement would you have before removing human review from that process?