Salesforce Agentforce | Working, Key Features, Agent Types

Agentforce is an AI agent platform offering a range of preconfigured autonomous agents and tools for building or deploying them.¹ These agents autonomously perform complex CRM tasks using data, workflows, and business rules within Salesforce. The company announced Agentforce in September 2024.²^,³

Agentforce has evolved from a few prebuilt agents into a comprehensive platform for building, testing, deploying, and monitoring agents.⁴ In September 2026, Salesforce introduced a portfolio of job-ready agents and a runtime that enables agents to pursue a single goal over extended periods.⁵ Agentforce agents operate using existing customer records and business rules in Salesforce. For example, they can resolve service cases, book meetings with qualified leads, or modify orders autonomously, transferring conversations to a person when requests exceed their scope.¹

This article provides a detailed overview of Salesforce Agentforce, focusing on how it differs from Einstein, how it works, its key features, agent types, use cases, and benefits, and how to decide whether your business needs it.

What is Salesforce Agentforce?

Salesforce Agentforce is the platform for AI agents that work inside a Salesforce org. An agent takes a request from a customer or an employee and completes it on its own, right from reading the question to updating the record.¹

It is more than just a set of ready-made agents. Every agent works on the records, automations, and permissions already in the org.⁴^,⁵

As such, the team does not move its data into a separate AI tool before an agent can use it.

Agentforce History

In October 2025, Salesforce made Agentforce the base of Agentforce 360, which also covers Data 360, the Customer 360 apps, and Slack, and it added the Agentforce name to its main clouds.⁶^,⁷ In September 2026, it took that name off the clouds again.⁷ So, Sales Cloud, Service Cloud, and Marketing Cloud are back, while Agentforce remains the name of the agents.⁷^,⁸

Salesforce points buyers to consulting partners for Agentforce rollouts.¹ In most cases, switching the agent on is the easy part, and preparing the process and the records behind it is the hard part. Vegacompass is one such Salesforce Consulting Partner, a CRM-only firm that sells no licenses, and every agent project we take on starts with that groundwork. To learn more about the clouds, please read our article: What is Salesforce?.

Agentforce vs. Einstein and Traditional Chatbots

Einstein is nothing but the set of AI features built into the Salesforce clouds. It scores leads, predicts outcomes, and drafts summaries for a user, who then decides what to do next.¹ Please note that Einstein Copilot, the earlier assistant, is now part of Agentforce.¹

A traditional chatbot follows a conversation tree written in advance.⁴ As such, it gets stuck on any question it was not built for. For instance, consider a customer who wants to cancel an order only if it has not shipped; the tree has no branch for that.

That's why Agentforce works differently. An agent reads the request, plans the steps, and completes them through its allowed actions, passing the conversation to a person when it should not act alone.¹^,⁴ The difference is simple: Einstein suggests, a chatbot replies, whereas an agent does the work.

Quick Comparison: Einstein vs. Agentforce vs. Traditional Chatbots

	Einstein	Agentforce	Traditional Chatbots
Strength	Predictions and drafts inside the record	Finishes multi-step requests on its own	Predictable answers to known questions
Limitation	A person still acts on each suggestion	Only as good as the process and data behind it	Stuck once a question leaves the script
Best For	Lead scores, summaries, next-step hints	Repeatable, high-volume service and sales work	FAQs and simple routing

How does Salesforce Agentforce work?

An Agentforce agent handles every request in roughly the same order. For example, a customer asks the website chat to move a delivery to Friday.

Below is what happens next:

The Request: The message arrives on a channel, such as the website chat, a portal, WhatsApp, Slack, or a phone line.⁹
Subagent Selection: The agent routes it to its subagent for order changes. Subagents, called topics until April 2026, are the jobs an agent is set up to do.⁹
Grounding: It reads the order record and the delivery rules, so the reply rests on this customer's actual order.¹
Reasoning: The Atlas Reasoning Engine breaks the task into smaller steps and checks each result before moving on.¹
Action: It updates the date through an action, such as a Flow, Apex, or a MuleSoft API, within the rules and permissions set for the agent.¹^,⁵
Handover: If the order has already shipped, it passes the chat to a service rep with the conversation so far.¹

The model behind the reasoning is a choice, too. Salesforce offers Claude and Gemini models inside Agentforce, and it has built Koa, its own CRM reasoning model, with NVIDIA.¹⁰^,¹¹^,¹²

Key Features of Agentforce

Agentforce has a tool for every stage of an agent's life.

Below are the main ones:

Agent Builder: A low-code tool for setting up an agent's subagents, actions, and instructions, starting from a template or a plain-language description.¹^,⁹
Agent Script: An open-source language for the steps that must follow fixed rules, used alongside the AI reasoning.⁵
Testing Center: Generates test cases and runs them in a sandbox, checking which subagent an agent picks, the actions it takes, and its answers.⁴^,⁹
Agentforce Observability: Near real-time views of how each agent performs, with Agent Optimizer suggesting what to fix.¹^,⁵
Multi-Agent Orchestration: Passes work between specialized agents, so a job that crosses teams runs as one.⁵
Agentforce Voice: The same agents on phone calls, reading and updating the CRM record during the call.⁹
Einstein Trust Layer: Masks sensitive data before a prompt reaches the model, keeps no data with the model provider, checks responses for toxic content, and logs each step.⁴^,⁹
MCP Support and AgentExchange: Connections to outside tools through the Model Context Protocol (MCP), and a marketplace, formerly the AppExchange, for ready-made agents and apps.¹^,⁹

Types of Agentforce Agents

Agentforce agents are classified in two ways, and each one answers a different question about the agent.

Below are the two ways Agentforce agents are classified:

By Who It Serves: Describes the audience. A service agent faces customers on the website, a portal, or a messaging app, whereas an employee agent helps the staff inside Salesforce, Slack, or a mobile app.¹³^,¹⁴
By How It Is Built: Refers to the starting point. A prebuilt agent comes ready for a common job, whereas a custom agent is built in Agent Builder for a job that only the company has.¹⁴

The channel is not a separate type, since one agent can run on several channels at once. As such, every agent has a place on both lists.

Salesforce Agentforce Use Cases

Salesforce gives each of its job-ready agents a name and one main job.⁵

Below are the main use cases:

Customer Service: Casey, the help agent, handles customer requests by phone, SMS, WhatsApp, and web chat.⁵ Fin, from Salesforce's acquisition of Fin AI, also answers customers where parts of the service stack sit outside Salesforce.⁸
Sales Development: Piper talks to inbound visitors, nurtures leads, and books meetings, whereas Hunter works outbound, following up with prospects over days and weeks.⁵^,¹⁵
Marketing: Campaign Agent builds, launches, and tunes campaigns toward a goal, within the marketer's guardrails.¹⁶
Commerce: Carter, the shopper agent, helps buyers on the storefront find the right product.⁵
IT, HR, and Operations: Paige handles employee requests such as a help-desk ticket or a benefits question, and Marshall takes on supply chain work.⁵^,¹⁴

Please note that each agent is only as useful as the records behind it. A help agent reading outdated articles gives outdated answers, only faster.

Benefits of Salesforce Agentforce

You may be wondering: an agent adds usage costs and a setup project. So, a fair question arises: where does the payback come from?

The answer is the routine work it takes off the team.

Below are the main benefits:

Reuse of What the Org Already Has: An agent uses the existing Flows and Apex as its actions.⁴ As such, the team extends automations it already runs rather than rebuilding them.
A Clearer View of the Process: Writing an agent's instructions often exposes steps that were never written down, which pays off even before go-live.
The Same Permissions and Audit Trail: The agent works within the permissions set for it, and every step is logged for review.⁴^,⁵
Round-the-Clock Coverage: An order question or an inbound lead gets an answer at 2 a.m. on a Sunday, and the team keeps its day for the exceptions.¹^,¹⁵

How to decide if your business needs Agentforce

An agent pays off where the same request arrives hundreds of times a week, and far less where every request is different.

The decision works as follows:

Who it suits: Salesforce orgs with a high volume of repeatable requests, a written process for handling them, and records clean enough to act on.
Who should look elsewhere: Teams whose requests are rare or unique, where a person is still faster, and companies on another CRM, since Dynamics 365 has Copilot on Dataverse and HubSpot has Breeze agents.
For everyone already on Salesforce: The answer is an audit before the first agent, because an agent acts on whatever the records say.

At Vegacompass, senior consultants run the Salesforce CRM audit, which generally takes two weeks and ends in a written diagnosis with the fixes ranked by value. Where the records need work, CRM Data Cleanup comes before any agent. To learn more about planning agents around your process, please read our AI CRM Consulting page.

FAQs

Q. Is Agentforce a separate product or part of Sales Cloud?

It is a separate agent platform that works inside the clouds, such as Sales Cloud and Service Cloud. Some contracts may still show the older names, such as Agentforce Sales, since the Cloud names only returned in September 2026.⁷

Q. How much does Agentforce cost?

Salesforce offers three models: usage-based Flex Credits, a charge per resolved conversation, and per-user licenses.¹^,¹⁴ Customers on Enterprise Edition and above can start through Salesforce Foundations, which includes a first batch of credits.⁴ Salesforce's pricing page is the only current source.

Q. Does Agentforce need Data 360?

Yes, Data 360 must be switched on, since the data library and the Einstein Trust Layer depend on it.¹⁷ A full Data 360 build, however, is needed only when an agent answers from documents or needs one customer profile across systems.¹⁸

Vegacompass: Salesforce Consulting Partner for Agentforce and CRM AI

Vegacompass has worked in CRM and digital transformation for 12+ years. We help organizations grow through CRM, amplified by AI. We hold the Salesforce Consulting Partner status and are an official Claude partner.

Our senior consultants plan and build the agents this article describes, and our AI team adds in-house ones such as AI Agent Salesman and Vega Support AI (SimCase) once the process is ready.

We offer the following CRM consulting services:

CRM Strategy (Marketing, Sales and Customer Service)
Revenue Operations Consulting
CRM Audit & Technical Debt
Salesforce Implementation and Migration
CRM Migration
CRM Data Cleanup
AI CRM Consulting

Get in touch

Please tell us where you are with Agentforce, whether you are planning a first agent or rescuing a stalled pilot, along with your seats and integrations.

A senior consultant reads your note and, after a 30-minute call, sends a written view on where to start, yours to keep. An NDA can come first. That consultant then runs the build with our delivery team in Pune, Palo Alto, Austin, and Singapore.

Talk to a Salesforce consultant today to find out what your org needs before its first agent.

References

(2026). Agentforce: The AI Agent Platform. Salesforce. https://www.salesforce.com/agentforce/
(September 12, 2024). Salesforce Unveils Agentforce–What AI Was Meant to Be. Salesforce. https://www.salesforce.com/news/press-releases/2024/09/12/agentforce-announcement/
Mitchell, C. (October 29, 2024). Salesforce Goes Live with Agentforce, Promises to "Redefine What's Possible In Business". CX Today. https://www.cxtoday.com/crm/salesforce-goes-live-with-agentforce-promises-to-redefine-whats-possible-in-business
(2026). How Does Agentforce Work?. Salesforce. https://www.salesforce.com/agentforce/how-it-works/
(September 11, 2026). Salesforce Expands Agentforce With a New Portfolio of AI Agents Built for High-Value Work. Salesforce. https://www.salesforce.com/news/stories/agentforce-job-ready-ai-agents/
(October 13, 2025). Welcome to the Agentic Enterprise: With Agentforce 360, Salesforce Elevates Human Potential in the Age of AI. Salesforce. https://www.salesforce.com/news/press-releases/2025/10/13/agentic-enterprise-announcement/
Dina, C. (September 14, 2026). Salesforce drops 'Agentforce' from product names, The Information reports. The Next Web. https://thenextweb.com/news/salesforce-drops-agentforce-branding-product-names-dreamforce
Mitchell, C. (September 15, 2026). Dreamforce 2026: The Top Announcements, Inc. AIforce, Koa, & New Partnerships. CX Foundation. https://cxfoundation.com/news/dreamforce-announcements-2026
Marshall, C. (September 9, 2026). What Is Salesforce Agentforce?. Salesforce Ben. https://www.salesforceben.com/how-does-salesforces-agentforce-work/
(August 26, 2026). Salesforce and Anthropic Announce Claudeforce: The #1 AI Meets the #1 AI CRM. Salesforce. https://investor.salesforce.com/news/news-details/2026/Salesforce-and-Anthropic-Announce-Claudeforce-The-1-AI-Meets-the-1-AI-CRM/default.aspx
(October 16, 2025). Salesforce and Google Expand Strategic Partnership with Deeper Integrations Across Agentforce 360 and Gemini Enterprise. Salesforce. https://www.salesforce.com/news/press-releases/2025/10/16/google-gemini-agentforce-partnership-expansion-announcement/
(September 24, 2026). Five Big Ideas Coming Out of Dreamforce 2026. Salesforce. https://www.salesforce.com/news/stories/five-dreamforce-2026-takeaways/
(2026). Before You Begin with the Agentforce Mobile SDK. Salesforce Developers. https://developer.salesforce.com/docs/ai/agentforce/guide/agent-sdk-before-you-begin.html
(2026). Get Started with Agentforce for Employees. Trailhead. https://trailhead.salesforce.com/content/learn/modules/agentforce-for-employees-quick-look/get-started-with-agentforce-for-employees
(July 29, 2026). Salesforce launches AI SDR Piper in just 30 days with Qualified. Salesforce. https://www.salesforce.com/salesforce-on-salesforce/agentic-presales-qualified-piper/
(2026). Introducing our biggest Marketing innovations from Dreamforce 2026. Salesforce. https://www.salesforce.com/marketing/whats-new/
(2026). Explore Data 360 and Agentforce. Trailhead. https://trailhead.salesforce.com/content/learn/modules/data-cloud-powered-agentforce/explore-data-cloud-and-agentforce
(2026). Does Agentforce Require Data Cloud? Cost & Readiness Guide. Clientell. https://www.getclientell.com/salesforce-blogs/does-agentforce-require-data-cloud

Guide to Salesforce Agentforce: Working, Agent Types, Key Features, Use Cases, Benefits