AI Agents Displacing Salesforce
Sierra vs Decagon
The real threat to Salesforce is not that AI agents sit on top of the CRM, it is that they can become the worker, the workflow, and eventually the place where customer context is created first. Sierra and Decagon start by handling support conversations, but once the agent is also collecting data, deciding next actions, and writing outcomes back into CRM, the system of record risks becoming a passive database behind a more valuable system of action.
-
The wedge is economic and operational. Third generation support agents resolve roughly 60 to 80% of conversations, often at about $0.99 to $1.50 per resolution versus roughly $10 to $15 for human support, which lets buyers cut BPO spend first and worry about software stack changes later.
-
This starts as disintermediation, not rip and replace. Many deployments run on top of Salesforce, Zendesk, or Intercom at first, with agents reading knowledge bases, handling the conversation, then writing notes, fields, refunds, or status changes back into the incumbent system through APIs.
-
The expansion path goes beyond support. Sierra is already framed as moving into outbound sales, collections, claims follow up, and supplier workflows, while Intercom shows the incumbent response by bundling an agent with the help desk so it can stay close to both the human operator and the underlying data.
Over time, the winning product is likely to be the one that owns the live conversation loop and the write path into many business systems. If AI agents keep getting better at tenant specific integrations and real time action taking, systems like Salesforce remain important as databases, but far less central as the software people actually use to get work done.