Who owns it when an AI agent gets it wrong? (A question I keep coming back to.)
We spend a lot of energy on the happy path — the agent qualifies the lead, drafts the reply, updates the record. Far less on the moment it does all of that confidently and incorrectly, maybe in front of a customer.
Agents act in seconds. The human judgment to catch and override them scales in months. That gap seems to be where a lot of rollouts quietly fall apart.
I'm curious how teams here are handling it in practice:
1. Who actually owns or reviews an agent's decisions in your org — RevOps, the line manager, a dedicated "AI steward", or nobody yet?
2. How fast can a human step in when an agent goes off the rails — and is there a clean rollback?
Not after the vendor answer — genuinely want to hear how Salesforce teams are governing this on the ground. What's working, and what isn't?
#Agentforce #Salesforceadmins
We are rolling out a new Agent that, in testing, has a 93% success rate in routing emails. These are results than expected for the introduction but the busines says, what about the 7% that are "wrong".
In our partnership with Salesforce they asked/told us early in the project, you will need to have an "Agent Manager". This person or persons would be responsible for ensuring the agent stays within it's guardrails, monitors feedback from users etc. etc. etc. . Not quite a baby sitter but something close to it. So who should be in that role, someone from the business, an Admin or Developer??? We haven't identified an Agent Manager.... yet.