A field-services company had a Salesforce org with eleven custom objects, three approval processes, and a validation rule for almost everything. Their question was whether an AI agent could answer customer calls about job status without breaking any of it. It could, and the way it was done is the pattern this guide describes.
Key takeaways
- An AI agent should act as a scoped integration user so every validation rule, trigger, and approval process still applies.
- Map the agent to the objects that matter for the conversation, so cases cover service, leads and opportunities cover sales, plus your custom objects.
- Choose between a custom agent and Agentforce based on channel, integrations outside Salesforce, and control over cost and model.
Service conversations from the case record
For a service agent, the case and its related records are the source of truth. The agent verifies the caller, reads the open case, asset, and entitlement, answers status and how-to questions, and creates or updates the case with the transcript. Routing follows your existing assignment rules.
Sales conversations from the lead and opportunity
For sales, the agent qualifies inbound leads by your criteria, converts them following your process, updates opportunities after calls, and follows up on stalled deals. Territory and owner rules stay in Salesforce; the agent respects them.
Custom objects and the rules around them
The strength of a custom agent is that it can read and write your custom objects, such as jobs, sites, or contracts, and it writes through the API as an integration user, so validation rules, triggers, and approval processes apply exactly as they would for a human. Test in a sandbox first, always.
Custom agent or Agentforce?
Agentforce is a strong option for standard use cases on eligible editions. A custom agent makes sense when you need voice on your own phone system, integration with tools outside Salesforce, a specific conversation design, or control over model choice and cost. Many organizations use both: Agentforce for in-platform tasks and a custom agent for the phone line.
Governance that keeps IT comfortable
Least-privilege permission sets, field-level security on sensitive data, full logging, and a shadow period before go-live make the security review straightforward. Document what the agent may do and have the business owner sign off before the first live call.
Common questions
Can an AI agent work with Salesforce custom objects?
Yes. A custom AI agent reads and writes standard and custom objects through the API as a scoped integration user, so your validation rules, triggers, and approval processes still apply.
Should we use Agentforce or build a custom Salesforce AI agent?
Use Agentforce for standard in-platform use cases on eligible editions. Build custom when you need voice on your own numbers, integrations beyond Salesforce, a bespoke conversation design, or control over model and cost. The two can coexist.