How Agentforce Works

Agents need three things to get work done: data, reasoning, and actions. With Agentforce, agents can connect to any data source and use it in real time to plan, reason, and evaluate. And Agentforce agents can leverage any workflow, automation, or API to complete tasks. Read on to learn how it all works.

Bring trusted data to your agents.

Data Cloud gives agents real-time access to the data they need to do work, without the need to copy data from existing warehouses. Create powerful agents that refer to your connected structured and unstructured data. That includes company knowledge articles, CRM data, external data lakes, and more.

In Agentforce, unstructured data including articles and policy docs is used to give an AI agent knowledge.

Metadata allows agents to know the context of your business and the actions available to them. Metadata is a core part of the Salesforce Platform: Every field, label, entry, and automation built on the platform is tagged with relevant metadata that Agentforce can read and understand. This metadata allows the agent to know exactly how to use a flow or what data it needs to pull in.

A view of the metadata available to the AI agent to reference.

Prompt Builder lets you build repeatable, tailored prompts that can deliver the exact data an agent needs to do work. These prompts can be used by agents to find and retrieve structured and unstructured data in real time through Retrieval Augmented Generation.

Learn more

In the Prompt Builder Template Workspace, a prompt gets configured.
In Agentforce, unstructured data including articles and policy docs is used to give an AI agent knowledge.
A view of the metadata available to the AI agent to reference.
In the Prompt Builder Template Workspace, a prompt gets configured.

Agentforce learns and reasons with the Atlas Reasoning Engine.

The Atlas Reasoning Engine uses advanced techniques like ensemble retrieval augmented generation (RAG), which combines the strengths of several RAG models, to find highly specific, accurate data for agents. Agents can search across structured and unstructured data to find similar language to the initial task, giving them the knowledge they need to respond and act with precision.

A view of the Atlas reasoning engine interface, in which a new data library is assembled.

The Atlas Reasoning Engine breaks down the initial prompt into smaller tasks, evaluating at each step and proposing a plan for how to proceed. For example, if an agent is handling a customer inquiry, it identifies the intent, searches for relevant data, creates a plan of action, and evaluates the effectiveness of the action. If unsatifactory, the agent continues to adapt and refine the plan by asking for additional information. This ensures that the initial prompt can be completed accurately.

A graph depicting how the Atlas Reasoning Engine learns and reasons in a constantly improving loop.

The Atlas Reasoning Engine evaluates a user request against all of the topics available to an agent, and then selects the most appropriate one to accomplish the task.

A view of a Topic being configured in Agentforce, with natual language directions for an order management task.
A view of the Atlas reasoning engine interface, in which a new data library is assembled.
A graph depicting how the Atlas Reasoning Engine learns and reasons in a constantly improving loop.
A view of a Topic being configured in Agentforce, with natual language directions for an order management task.

Agentforce takes action across your entire enterprise.

Agentforce is natively integrated with the entire Salesforce Customer 360. From Sales and Service to Commerce and Marketing, agents can use complete customer context from your CRM applications and take action directly within the flow of work for your employees. For example, an agent can use engagement data to identify an opportunity to upsell and generate a personalized email to a prospect.

A chat window between a sales customer and a sales AI agent, discussing pricing.

Agentforce can be embedded across web and mobile chat, email, SMS, and Slack to meet your customers wherever they are. Agents can reply natively, and also seamlessly handoff to human employees across any channel. For example, a customer can log onto your website and have a conversation with an agent to troubleshoot a broken item. The agent replies with best practices and trouble shooting instructions to close out the case.

In a chat window, a sales rep agent gives a sales employee context about an account.

Agents can take advantage of prebuilt Flows to automate business processes. Agents can be configured to run Flows and use their output to help respond to the intial prompt. Agents can also be embedded within a Flow to help automate even more work. For example, an agent can run a flow to automatically elevate the priority of a customer case ticket based on certain criteria.

A view of a workflow setup, and the data types used in it, for a hotel reservation system.
A chat window between a sales customer and a sales AI agent, discussing pricing.
In a chat window, a sales rep agent gives a sales employee context about an account.
A view of a workflow setup, and the data types used in it, for a hotel reservation system.

Agentforce Activator. Confidently deploy agents fast.

Work with the experts from Salesforce Professional Services and Certified Partners to create an agent tailored to your business goals and aligned with your values.

Agentblazer Characters

Join the Agentblazer community.

Connect with Agentblazers from around the world to skill up on AI, discover use cases, hear from product experts, and more. Grow your AI expertise — and your career.

Agentforce Frequently Asked Questions

An AI agent is an intelligent system that can get work done proactively and autonomously, meaning it doesn't require human input. Agents can be triggered by changes in data and automations. They can understand and respond to human conversation dynamically. And they rely on machine learning and natural language processing (NLP) to reason and handle a wide range of tasks, from answering simple questions to resolving complex issues.

Traditional chatbots require pre-determined conversation trees to respond to inquiries. In contrast, agents are dynamic. They can adapt to human language and conversation, and they have the ability to reason. They make sense of the conversation, build a plan to address it, understand the tools available to them, and then take the best course of action.

You can customize an AI agent to fit any use case and any industry. Salesforce offers a range of out-of-the-box agents for service, sales, marketing, and commerce, so you can get started in minutes. These are just some examples of how agents can support your business:

  • Service agents can automatically help a customer troubleshoot an issue with personalized, branded advice based on their purchase history and company knowledge documents
  • SDR agents can send a personalized introductory email to a lead, respond to a question on pricing, and access an account executive’s calendar to schedule a follow-up call
  • Marketing agents can generate campaigns, briefs, audience segments, emails, and end-to-end journeys based on your company's goals
  • Ecommerce agents can manage your websites and generate personalized promotions

The Einstein Trust Layer protects customer data through robust security features and guardrails, like zero data retention, toxicity detection, secure data retrieval, and dynamic grounding. It improves the safety and accuracy of outputs while ensuring the responsible use of AI agents across the Salesforce ecosystem.

For example, the Audit Trail feature provides the data you need to track AI agent actions and outputs, ensuring AI usage complies with your organization’s security, privacy, regulatory, and AI governance policies.

While other agent platforms require complex data integration and custom automation builds, Agentforce is already built into the Salesforce Platform. You can instantly turn your existing workflows, prompt templates, Apex, and APIs into agent actions — with the added power of native tools, like Data Cloud, Slack, and MuleSoft. Agent Builder enables Salesforce admins and developers to use natural language to create jobs to be done, instructions, actions, and guardrails for their agents.

Agentforce is deeply integrated with the Salesforce Platform, which brings together all the data and context agents need from internal and external sources, with the Einstein Trust Layer built in. This enables agents to take accurate, relevant actions, which are faster to implement thanks to our low-code framework. You can reuse Platform tools like flows and Apex, and build your agent on top of them.

Agentforce also connects to all the channels where your employees and customers already interact, including CRM, WhatsApp, Messenger, and your website. There’s no need to go to another application to talk to an agent because they're bringing value to the existing flow of work.

Agentforce is suitable for businesses of any size. Salesforce customers can now try Agentforce for free with Salesforce Foundations, a $0 add-on for Enterprise Edition and higher.

Foundations helps businesses get ready for AI by extending their CRM with additional key features from Agentforce, Sales Cloud, Service Cloud, Marketing Cloud, Commerce Cloud, and Data Cloud. Foundations now includes credits to power the first 1,000 conversations with Agentforce Service Agent, allowing your company to try its capabilities before buying.