Curious where RO teams are drawing the line on sales reps using AI agents. We use Cursor, Claude Code, MCPs, etc. to move faster. But giving reps that same autonomy is a different question when it means AI writing to Salesforce, Outreach, Slack, or other production tools. Would you trust any rep-owned workflows if permissions, approvals, audit logs, and rollback were built in or is that a hard no?
Jorge, my take from building this stuff. The line is at audit + reversibility, not autonomy itself. If a rep-owned AI workflow writes to Salesforce, you need: (1) every write is logged with the prompt, output, and rep ID, (2) every write is reversible within X hours, (3) there's a rate limit so a runaway loop can't destroy 1000 records in 30 seconds, (4) high-stakes actions like closing deals or changing forecast amounts require human confirmation before execution. With those four things, I'd trust most rep-owned workflows. Without them, hard no regardless of how sophisticated the agent is. The bigger risk isn't a malicious rep, it's an unintended cascade. AI agent drafts a follow-up email, sends it, prospect replies, agent auto-routes the reply, kicks off another workflow. Each step is reasonable, the chain is catastrophic.
Daniel F. Fully agree on the line being audit + reversibility, #4 is where gets genuinly hard. High stakes isn't a static list, when you start to merge in different customers, sales motions/frameworks etc. it gets rather complex and dynamic. Someone has to encode that judgment and today it lives in scattered business logic across platforms. But curious, if you have built something like this or used any third-party tools/products to accomodate these requirements?
