Short answer
Choose Agentforce when the work happens inside Salesforce, on clean Salesforce data, and your admins want native tooling. Choose a custom agent when you need deep cross-system context, your own telephony or voice stack, a choice of models, or behaviour Agentforce doesn’t support yet. Many companies run both: Agentforce for in-platform actions and custom agents for the work around it.
The two options
Agentforce is Salesforce’s platform for building AI agents that work natively in Salesforce. You define topics, actions and prompt templates, ground them in Salesforce data and Data 360 (formerly Data Cloud), and manage them with Salesforce’s admin, testing and permission tooling.
A custom agent is built on a model provider and an orchestration framework. It works with Salesforce through its APIs or an MCP server, and with any other system you connect: call recordings, product usage, a data warehouse or other tools.
Comparison
| Agentforce | Custom agent | |
|---|---|---|
| Strongest at | Actions on Salesforce records | Workflows spanning several systems |
| Data it sees | Salesforce and Data 360 | Anything you connect, Salesforce included |
| Channels | Salesforce-supported channels | Any channel, including your own telephony |
| Model choice | Options managed by Salesforce | Any model, switchable over time |
| Governance | Native permissions and testing tools | Built in: approvals, audit logs, evals |
| Commercials | Salesforce licences and consumption | Build, hosting and model usage |
| Time to first agent | Fast when data and automation are clean | Typically 2–4 weeks for a focused agent |
Readiness decides more than the platform
Agents that take actions raise the bar compared with predictive AI features. Whichever route you choose, check these first (our CRM readiness checklist covers them in detail):
- Duplicate and incomplete records on the objects the agent will use.
- Flows and triggers that would fight with agent updates.
- A least-privilege integration user and field-level write rules.
- One defined use case with an owner and a baseline metric.
A hybrid pattern that works
Use Agentforce for in-platform work such as case handling, record updates and guided selling inside Salesforce. Use custom agents for voice, research across external sources and anything that needs context from outside Salesforce. Connect them through a shared context layer so both act on the same view of each account.