Agentforce — Independent Software Review

Empower your business with autonomous AI agents.

Compliance Transparency Index

Grade: A — Score: 85/100

Best For

• Existing Salesforce organizations that want AI agents grounded in CRM, Knowledge, Data 360, Flow, and Apex without moving business logic to a separate agent platform.
• Customer service teams that need agents to answer questions, update records, execute workflows, and escalate within Salesforce-governed channels.
• Sales, service, field service, and employee teams that want cloud-specific Agentforce add-ons with unmetered employee agent usage.
• Salesforce admins and developers that want both low-code Agentforce Builder and pro-code options including Agent Script, Apex, APIs, and MCP.
• Enterprises that need an agent ecosystem with AgentExchange, native MCP support, granular Salesforce permissions, and third-party LLM zero-data-retention controls.

Not Ideal For

• Organizations that do not already use Salesforce and want a standalone agent platform without CRM edition, cloud add-on, or Salesforce data dependencies. Foundations is positioned as an add-on to an eligible Salesforce org, and required licenses vary by agent type.
• Teams that require one simple fixed recurring price for every agent workload. Agentforce mixes Flex Credits, Conversations, per-user licenses, cloud add-ons, and other consumption services, and Salesforce notes that pricing examples can exclude Data 360 credits and other costs.
• Architecture teams that require one unambiguous public prerequisite for Data 360 before implementation. Salesforce Trailhead says Data 360 must be provisioned and enabled for all Agentforce use, while the current Agentforce DX guide says agents that do not require Data 360 can skip enabling it.
• Small teams that mainly need lightweight cross-app agents and do not need Salesforce-native CRM permissions, multi-agent governance, or cloud-specific workflows. The platform's strongest differentiation comes from its integration with the wider Salesforce stack.

Operational Overview

Pricing Structure

Salesforce Foundations: $0 add-on for eligible Salesforce customers

• Agentforce Builder
• Prompt Builder
• Agent Script
• Agentforce Coworker
• Agentforce Vibes

Flex Credits: $500 USD per 100,000 credits

• 20 credits per standard Agentforce action
• 30 credits per Agentforce Voice action
• Customer-facing and employee-facing agents
• Digital Wallet usage tracking

Conversations: $2 USD per conversation

• Customer-facing agents
• Digital Wallet
• Pre-Purchase buying model
• Cannot be used with Flex Credits in the same org

Agentforce User License: $5 USD/user/month plus required Flex Credits

• Employee Agentforce access
• Limited Salesforce CRM objects
• Metered employee usage

Agentforce Add-ons: $125/user/month

• Available for Sales, Service, and Field Service
• Unmetered employee Agentforce usage
• Predictive, generative, and agentic AI
• AI analytics and Prompt Builder

Agentforce Industries Add-ons: $150/user/month

• Available for Industry Clouds
• Unmetered employee Agentforce usage
• Industry-specific AI
• Includes Agentforce Sales and Service add-on capabilities

Agentforce 1 Editions: From $550/user/month

• Available for Sales, Service, Field Service, and Industries
• Agentforce add-on included
• 2.5 million Flex Credits per org per year

Alternative Consideration

Consider switching to Microsoft Dynamics 365: Offers similar AI capabilities but may lack the same level of integration with existing Salesforce tools.

Frequently Asked Questions

How does Agentforce compare with Microsoft Copilot Studio?

Agentforce is built around Salesforce CRM data, permissions, Flow, Apex, Data 360, and Agentforce Builder, while Microsoft Copilot Studio is centered on Microsoft 365, Power Platform connectors, Dataverse, and Azure-linked environments. Salesforce lists Flex Credits at $500 per 100,000 credits, with a standard Agentforce action consuming 20 credits, while Microsoft lists Copilot Studio capacity packs at $200 per month for 25,000 Copilot Credits and also offers pay-as-you-go. Those credit units are not directly equivalent, so the stronger fit usually depends on whether the organization's core business data and workflows already live in Salesforce or Microsoft.

Does Agentforce require Data 360 to work?

Salesforce's current public documentation gives conflicting prerequisite guidance. A Trailhead module says Data 360 must be provisioned and enabled for all Agentforce use and that Agentforce Data Library and Einstein Trust Layer features do not work without it, while the current Agentforce DX setup guide says agents that do not require Data 360 can skip enabling it. Buyers should verify the requirement for their exact agent type, environment, grounding method, and Trust Layer or Data Library use before implementation.

Can Agentforce reuse existing Salesforce Flow, Apex, and prompt templates?

Yes. Agentforce actions can call Salesforce Flow, Apex, and Prompt Builder templates, and Agent Script can expose those actions to the reasoning engine or require them at specific steps. This lets an organization reuse existing Salesforce business logic rather than rebuilding every workflow specifically for an AI agent.

Can Agentforce connect to external systems through Model Context Protocol?

Yes. Salesforce has a native Model Context Protocol client in Agentforce that can register third-party MCP servers, allowlist their tools, and add those tools as agent actions in Agentforce Builder. Agentforce Gateway can then apply policies to specific agents and MCP server tools, while MuleSoft and Salesforce-hosted servers can also be discovered through the API Catalog.

How does Agentforce test for hallucinations and unsafe agent behavior?

Salesforce provides Agentforce Testing Center in sandbox environments to evaluate topic classification, action sequences, multi-turn conversations, response quality, prompt injection cases, and agent behavior against expected results. Salesforce also documents Trust Layer controls including prompt defense, toxicity detection, grounding, and configurable agent instructions or guardrails. Testing consumes requests and credits and can modify data, which is why Salesforce recommends running these tests in a sandbox rather than production.

Can Agentforce be deployed to web, messaging, Slack, email, and voice channels?

Yes, but channel support depends on the agent type and related Salesforce products. Salesforce documents Employee Agent deployment to Lightning Experience, Salesforce Mobile, Slack, and web chat, while Service Agents can use enhanced messaging, email, telephony, and other service channels. Agentforce Voice extends supported agents to phone, web, and mobile voice experiences and can require Salesforce Voice or partner telephony configuration.

Do third-party LLM providers train on Agentforce customer data?

Salesforce says the Einstein Trust Layer applies zero data retention to third-party LLM providers used by Agentforce, so those providers do not retain the data or use it to train their models after the response is returned. Salesforce also says global predictive models do not apply to the Agentforce platform and agent actions. Other Salesforce AI features added to an agent can have separate data-use rules and opt-out controls, so that protection should not be generalized to every AI feature in the Salesforce ecosystem.

Which LLMs can Agentforce use in the current and legacy builders?

Salesforce says agents created in the new Agentforce Builder use GPT-4.1 under Salesforce Default, while agents created in the legacy builder use GPT-4o. Salesforce also documents an AWS-hosted option using Anthropic Claude Haiku 4.5 and a Google Gemini option using Gemini 3.5 Flash, while agent actions can call other predefined or Salesforce-managed models. Salesforce's current documentation does not support bringing an arbitrary external model directly into Agentforce Default.

How are Employee Agents and Customer Agents secured differently in Agentforce?

Salesforce documents two different security models. Employee Agents typically run in the context of the logged-in Salesforce user and inherit that user's licenses, permission sets, field-level security, and sharing rules, while Customer Agents use a dedicated Agent User for public channels such as messaging or web chat. Salesforce recommends granting that Agent User only the specific permissions needed for its actions.

How can administrators monitor Agentforce usage and performance after deployment?

Salesforce provides Digital Wallet for near real-time consumption tracking, usage trends, configurable alerts, and tags that show which agents or resources are driving usage. Agentforce Observability adds centralized monitoring for agent performance, health, adoption, and consumption. Together they let administrators review both operational behavior and usage-based spend after agents go live.

AI Visibility Report

How AI agents (ChatGPT, Perplexity, Claude, others) read this review page in the past 7 days. Updated weekly. View Agentforce AI Visibility Report.