Salesforce Agentforce is an AI agent platform that enables businesses to deploy autonomous agents capable of handling customer interactions, sales tasks, marketing workflows, and internal operations with less human involvement in routine, well-defined steps. Introduced in 2024 and expanded through later Agentforce releases, it represents Salesforce’s shift from AI copilots that assist people to AI agents that independently execute multi-step work.
Agentforce: From AI Copilot to Autonomous Agent
Earlier Salesforce AI tools — Einstein bots and Copilot — assisted users by surfacing information or drafting content on request. Agentforce operates differently: agents take actions across multi-step workflows, make decisions within guardrails, and resolve tasks end-to-end, only involving a human when the task exceeds defined scope or confidence.
Before — AI Copilot
User asks a question. AI suggests a response. User reviews and decides. Human executes the action manually.
Now — AI Agent (Agentforce)
Trigger received. Agent retrieves context, reasons through steps, executes actions within guardrails. Resolves or escalates based on configured rules.
Autonomous execution
Agents complete multi-step tasks with less human input for routine, well-defined processes.
Unified data access
Agents draw from Salesforce CRM and Data 360 to inform each decision with relevant, up-to-date context.
Built-in guardrails
Topics, actions, and escalation rules define what agents can and cannot do, keeping behaviour within operator-defined boundaries. For a broader look at what this enables, see Salesforce automation for revenue operations.
Low-code configuration
Agent Builder allows non-developer teams to create and configure agents without writing custom code for most standard use cases.
Concurrent handling
Multiple agents can run in parallel, subject to Salesforce licensing terms and org configuration.
Native Salesforce integration
Agents work within existing Salesforce orgs, accessing Flows, Apex, and external APIs through standard connectors.
Agentforce Features and How It Works
Agentforce is built on six core features. Together they form a execution flow — trigger, context, reasoning, execution, and resolution.
Six Agentforce Platform Components
Atlas Reasoning Engine
Salesforce’s LLM layer. Processes context and determines the right action sequence for each task.
Agent Builder
Low-code interface for defining agent role, topics, actions, and escalation conditions.
Topics & Actions
Topics set what the agent handles; Actions define what it can do — Flows, APIs, record updates. See: Salesforce Flow, REST API.
Data 360 (formerly Data Cloud)
Unified data layer. Harmonises CRM records, interaction history, and external data for agent context.
Prompt Builder
Reusable prompt templates grounded with live CRM data to improve agent response relevance.
Guardrails & Escalation
Configured boundaries on agent behaviour. Out-of-scope tasks escalate to humans with full context.
Agentforce Agent Types
Six pre-built Agentforce templates — trigger, action, and output for each. All configurable via Agent Builder, with custom logic possible via Salesforce Flow or Apex.
Service Agent
Customer service resolution
SDR Agent
Sales prospecting & outreach
Sales Coach Agent
Rep coaching & practice
Campaign Agent
Campaign performance management
Commerce Agent
Shopper assistance & orders
Custom Agent
Any org-specific process
Agentforce Use Cases by Operator Type
Which Agentforce Use Case Fits Your Business?
Select your situation to see the recommended starting point and readiness checks.
Step 1 — Primary challenge
Step 2 — Your Salesforce situation
Readiness checks
- Case routing setup and field completeness in Service Cloud
- Knowledge article quality for your highest-volume query types
- Data 360 customer profile completeness
- Defined escalation path for cases the agent cannot resolve
Key decisions up front
- Which Service Cloud edition covers your case management requirements
- Whether Data 360 is included in the initial build or phased in later
- How human escalation works before the agent goes live
- Which query type to automate first, based on volume and logic consistency
How to structure a useful pilot
- Pick one high-volume, low-complexity query type with clear resolution logic
- Set measurable success criteria before go-live (deflection rate, resolution time)
- Run for four to six weeks — enough for reliable performance data
- Keep human escalation active and monitored throughout the pilot
When custom is needed
- Resolution requires data from a system outside standard Salesforce objects
- Multi-step approval logic is part of the resolution workflow
- Custom topic definitions are needed to match your service categories
Readiness checks
- Lead data completeness and field population for personalisation
- Existing routing and assignment rules the agent will operate alongside
- Which outreach sequence or lead source to automate first
- Rep adoption plan for Sales Coach use during onboarding or deal reviews
Why sequencing matters
- SDR Agent personalisation quality depends directly on CRM data completeness
- Lead capture, routing, and enrichment should be configured before the agent is activated
- Implementing both together avoids rework from separate data architecture decisions
How to evaluate before committing
- Select a specific lead source with consistent volume and clear conversion logic
- Define your measurement metric: leads to meetings, or meetings to qualified opportunities
- Run agent-handled leads alongside a control group for clean comparison data
When custom is needed
- Qualification criteria include data from external sources (intent data, firmographics, ERP)
- Handoff logic involves multi-step scoring or approvals
- Outreach personalisation requires fields not natively available in Sales Cloud
Readiness checks
- Data 360 data stream quality and segment definition completeness
- Which campaign type to automate first (email open-rate optimisation is a common entry point)
- A/B testing baseline data for measuring agent-driven improvement
Why data readiness determines agent quality
- Campaign Agent quality is directly proportional to Data 360 segment completeness
- Identity resolution and data ingestion decisions affect every agent response downstream
- Configuring Marketing Cloud and Data 360 before Agentforce avoids segment rework
How to structure a pilot
- Select a campaign with existing performance benchmarks (open rate, CTR, or conversion)
- Run agent-managed version against a control for four to six weeks
- Use results to decide whether broader rollout is justified
When custom is needed
- Multi-channel orchestration outside standard Campaign Agent templates
- Approval workflows between marketing and other teams before campaign activation
- Integration with ad platforms or analytics outside the Marketing Cloud ecosystem
How to scope operational automation
- Map the target process: trigger, data sources, actions, escalation path
- Best candidates are high-volume, consistent-logic processes with clear outcomes
- Start with one process — multi-process agents create maintenance complexity
Why sequence matters
- Salesforce objects, Flows, and integrations the agent depends on must exist first
- Data architecture decisions made at setup affect what the agent can access
- MuleSoft connections to external systems should be scoped alongside the platform build
Criteria for suitability
- High volume, repeatable logic, low exception rate — good candidates
- Clear trigger, defined inputs, consistent outcome — good candidates
- Frequent edge cases or high judgment required — defer to humans initially
How to define scope before build
- Document what the agent can do, cannot do, and when it must escalate
- Define actions using existing Salesforce Flows or Apex for complex logic
- Test against real process data in a sandbox before production deployment
If the selector does not display correctly, scroll to the comparison table below.
Agentforce vs Other Approaches
How Agentforce compares with alternatives for handling customer interactions, sales, and operational tasks. See also: Salesforce automation for revenue operations.
Scroll horizontally on smaller screens
| Dimension | Agentforce | Rule-based chatbots | Traditional automation | Manual processes |
|---|---|---|---|---|
| Decision-making | Multi-step reasoning within guardrails | Scripted decision trees only | Pre-defined rule execution | Human judgment required |
| Novel scenarios | Handled via reasoning, within topic scope | Falls to fallback or human | Process fails or errors | Flexible but inconsistent |
| Data access | Unified via Data 360 and CRM | Limited, single-channel typically | System-specific, siloed | Manual lookup required |
| Salesforce integration | Native — Flows, Apex, MuleSoft | Varies by vendor | Point-to-point only | None |
| Concurrent handling | Scales with licensing and org config | Moderate | High but structurally rigid | Headcount-limited |
| Setup complexity | Low-code + Data 360 configuration | Medium | Typically dev-heavy | None |
| Escalation handling | Context-preserving handoff | Context lost on escalation | No escalation path built in | Inherent |
| Ongoing maintenance | Topic, action, and prompt updates | Script and flow updates | Rule and integration updates | Process documentation |
Agentforce Pricing
Agentforce uses a consumption-based pricing model rather than a flat per-seat licence. The primary unit is conversations or actions handled by agents, tracked through a credit system. Total cost depends on agent type, volume, and your existing Salesforce product licences.
Published Pricing
What Affects Total Cost
- Number of agent types deployed (Service, SDR, Campaign, Custom, etc.)
- Monthly conversation or interaction volume across all agents
- Existing Salesforce Cloud licences — Service Cloud, Sales Cloud, Marketing Cloud
- Data 360 configuration scope — data streams, identity resolution, ingestion
- Custom agent development — complexity of topics, actions, and integrations
- Prompt Builder template count and ongoing prompt optimisation
Agentforce Implementation Considerations
Implementing Agentforce involves Data 360 configuration, agent setup, integration work, and post-launch tuning. The scope depends on your existing Salesforce footprint, the number of agent types being deployed, and how many external systems need to connect.
What a typical Agentforce engagement covers
Common phases regardless of implementation partner
Common implementation scope
- Data 360 data stream setup and CRM data harmonisation
- Agent Builder configuration for target agent types
- Custom agent development where standard templates do not cover the workflow
- Salesforce integration services for external system connections via MuleSoft
- Prompt Builder template creation and guardrail definition
- Salesforce managed services for post-launch tuning and support
Common gaps that extend timelines
- Incomplete CRM data requiring cleansing before Data 360 ingestion
- Knowledge base gaps that limit Service Agent resolution quality
- Undefined escalation rules — Agentforce requires these before go-live
- Missing external system connections needed for agent actions
- Unclear scope on which process to automate first
Typical implementation sequence
Discovery and scoping
Map target workflows, define agent topics and actions, assess Data 360 readiness, and confirm integration requirements before build begins.
Data 360 configuration
Set up data streams, harmonise CRM records, and configure identity resolution so agents have the unified context they need.
Agent configuration and custom build
Configure agents in Agent Builder, build custom agents where standard templates do not cover the workflow, and create Prompt Builder templates.
Testing and UAT
Test agent reasoning and actions across defined scenarios including edge cases, escalation paths, and concurrent interaction handling.
Go-live and post-launch tuning
Deploy to production, monitor topic handling rates and escalation volumes, and tune prompt templates and guardrails based on real interaction data.
Frequently Asked Questions About Agentforce
Common questions about Salesforce Agentforce — pricing, Data 360 requirements, agent types, implementation timelines, and customisation options.
Agentforce is Salesforce’s AI agent platform, introduced in 2024, that lets businesses deploy autonomous agents capable of handling customer service, sales, marketing, and operational tasks with less human involvement in routine, well-defined steps. Unlike earlier Salesforce AI tools that assisted users on request, Agentforce agents independently execute multi-step workflows, access Salesforce data via Data 360, and take actions in connected systems — only escalating to humans when a defined guardrail condition is met.
Einstein bots followed scripted decision trees and could not handle novel inputs. Copilot was an AI assistant that suggested responses on demand but required the user to act. Agentforce uses the Atlas Reasoning Engine to plan and execute multi-step tasks autonomously, draws from Data 360 for unified context, and takes real actions in Salesforce and connected systems — the key distinction being autonomous execution versus assisted suggestion.
Salesforce has published pricing for Agentforce Service Agent from $2 per conversation. Salesforce has also introduced Flex Credits, a consumption-based model for broader Agentforce usage. Final cost depends on agent type, usage volume, existing Salesforce licences, Data 360 configuration scope, and implementation requirements. Verify current pricing directly with Salesforce, as pricing models for AI platforms can change with new releases.
Data 360, previously known as Data Cloud, is commonly needed for richer, context-aware Agentforce deployments. Agents draw from Data 360 to access unified customer profiles and interaction history. Without it, agents operate with less contextual information, which can reduce resolution quality. Data 360 setup is typically scoped as part of an Agentforce implementation project.
Yes. Salesforce provides pre-built agent templates as a starting point, but Agent Builder allows organisations to create agents with custom topics, actions, Apex logic, and external system integrations for industry-specific workflows. The degree of customisation needed depends on how closely your target process maps to an existing template.
Agentforce pre-built agent types are available for Service Cloud, Sales Cloud, Marketing Cloud, and Commerce Cloud. Custom agents built via Agent Builder can be deployed within any Salesforce org. The Agentforce product lineup has expanded since 2024; confirm current compatibility with Salesforce for your specific product mix.
A focused engagement for one agent type on an existing Service Cloud org with partial Data 360 configuration can take four to eight weeks. Engagements involving Data 360 from scratch, custom agent development, or multi-cloud integration typically run eight to twelve weeks or longer. Timeline is most commonly extended by data readiness gaps, undefined escalation rules, or scope changes discovered during discovery.
