Agentforce explained
Agentforce is Salesforce's platform for building, testing, deploying and governing AI agents for employees and customers.
Agentforce is Salesforce's platform for building, testing, deploying and governing autonomous AI agents for employees and customers. It replaced Einstein Copilot, which became Agentforce Assistant. In October 2025 Salesforce used "Agentforce 360" as a wider brand, and in 2026 it returned to names such as Sales Cloud and Service Cloud for the apps, while Agentforce remains the name of the agent platform.
What it is for
Agents handle tasks such as answering customer questions, qualifying leads, resolving IT and HR requests, and helping shoppers, using your Salesforce data and actions. They can work alone or hand over to people.
Core components
| Component | What it does | Status |
|---|---|---|
| Agentforce Studio and Agentforce Builder | Build, test and monitor agents; needs the Manage AI Agents permission | Available |
| Agent Script | Mixes deterministic logic with LLM reasoning | Available |
| Atlas Reasoning Engine | Plans and reasons over tasks | Available |
| Multi-Agent Orchestration | Coordinates several agents | GA |
| Agent Optimizer | Improves agent performance in Agentforce Builder | GA announced for October 2026 |
| Agentforce Voice and Observability | Voice channel and monitoring | Available |
| Long-Horizon Runtime | Longer-running agent work | Pilot; GA announced for November 2026 |
| AgentExchange and MCP support | Marketplace and tool connections | Available |
Since the week of 13 July 2026, new agents can only be created in the new Agentforce Builder. Legacy agents keep running and stay editable.
Agent types
Salesforce lists prebuilt agents including Casey (help and service), Paige (IT and HR), Carter (shopper), Marshall (supply chain), Piper (inbound pipeline) and Fin (customer experience), all GA. Hunter, an outbound sales agent, is a pilot with GA announced for November 2026. Generic types such as Service Agent, SDR and Sales Coach still appear, and Agentforce Coworker is a GA assistant in Lightning that calls your deployed agents.
Trust and governance
The Einstein Trust Layer provides grounding in CRM data, data masking, toxicity detection, an audit trail and zero data retention agreements with LLM providers. Some newer Salesforce copy calls it the Agentforce Trust Layer. See our Einstein and AI features guide.
How it connects with other Salesforce products
Agents run inside Sales Cloud, Service Cloud, Commerce Cloud and Marketing Cloud, and use Data 360 for unified context. Salesforce says every customer can start with Agentforce through Salesforce Foundations; commercial models include Flex Credits, Resolutions and per-user licensing.
Planning an implementation
- 01
Pick one job
Choose a narrow, measurable task with clear handover rules.
- 02
Prepare data and actions
Make sure records, knowledge and Flows the agent needs are accurate.
- 03
Write guardrails
Define topics, instructions and what the agent must not do.
- 04
Test thoroughly
Test with realistic conversations, including edge cases.
- 05
Monitor and improve
Review conversations and observability data after launch.
Common mistakes
- Starting with a broad "do everything" agent.
- Connecting agents to poor or outdated knowledge.
- Planning around pilot features as if they were GA.
How GeekMidst can help
Our Agentforce and AI automation service covers use-case selection, build and testing. Our Flow automation services prepare the actions agents call, and our Data Cloud consulting prepares grounding data. Read our responsible AI approach.
Questions we hear
Is Agentforce the same as Einstein Copilot?
Can I still use agents built in the old builder?
Is Agentforce free?
Do agents need Data 360?
Related reading
When to use Apex instead of Flow
Choosing how agent actions are built
→Record-triggered Flow best practices
Reliable automation for agents
→Security and data handling
How we handle your data
→Service agent with human handoff
Illustrative blueprint
→Sources
Planning your first agent?
Tell us the task you want an agent to handle, and we will help you scope and test it.
