Agentforce

basicspreading

Agentforce is Salesforce's enterprise platform, enabling companies to build and operate AI agents that independently handle customer service, sales, and other tasks.

In plain terms

Agentforce is a product of the American software company Salesforce. In Salesforce systems, more than one hundred thousand companies worldwide keep track of their customers, orders, and incoming complaints. Agentforce installs AI agents on this data asset: programs that not only answer the question but also handle the matter.

Imagine it as a new colleague who, on their first day, gets access to the part of the customer database relevant to them, receives the rules, and a list of what they may and may not do. They work day and night, talking to thousands of customers at once, and if they get stuck, they hand the case over to a human.

Old chatbots followed pre-written scripts and easily hit dead ends. The Agentforce agent uses a language model to decide itself what steps are needed, for example, checking the order, verifying the warranty, and then initiating the replacement. The company does not need to develop its own AI for this, but must configure the ready-made system, connect it to its data, and test it.

How it works

The company sets up the agent: it defines its role (for example, customer service representative), the topics it can handle, and the operations it can perform, such as initiating a refund or booking an appointment.
When an inquiry arrives via chat, email, or phone, the agent uses the language model to interpret what the customer wants.
The agent retrieves the necessary data from company systems: who the customer is, what they bought, and what the policy says.
The system's reasoning engine (Atlas Reasoning Engine) breaks tasks into steps and executes them using permitted operations.
Built-in safeguards filter sensitive data and prohibited topics; if the matter is too complex or risky, the agent hands it over to a human operator.

An everyday example

At 10 p.m., you type into a web shop chat window because you received the wrong shoes. You do not need to click through menus: the agent finds your order, checks if an exchange is due, sends the return label, and confirms the new shipment. If a Salesforce system runs in the background, you likely spoke with an Agentforce agent.

Who is affected

entrepreneur, executiveIf your company uses Salesforce, this is the most obvious path to introducing AI agents, but you must calculate the costs and return on investment yourself.

employeeIn customer service and sales, agents take over some routine tasks, while your work shifts toward more complex cases and supervising the agents. → Work & HR

marketerAgents can prepare campaigns and filter leads, changing how a potential customer reaches the salesperson. → Marketing

everyoneAs a customer, you are increasingly answered by an agent where a human or a dumb chatbot used to sit.

A common misconception

MythAgentforce is a smarter chatbot that only needs to be turned on, after which it takes over customer service.

In realityAgentforce is a platform on which a company has to configure, connect to its data and test its own agents. They are only as good as the data and rules behind them are well organized. Complex cases still require a human.

Limits and risks

An agent can also make mistakes: it may confidently give a wrong answer or perform an incorrect operation, so human oversight is required for steps involving money or access permissions.
Disordered, incomplete customer data leads to poor decisions, so most of the implementation is actually data cleaning.
The pricing varies and often changes: it can be usage-based (according to conversations or operations performed), per-user, or flat-rate. In the usage-based version, costs are difficult to plan and can rise quickly with high traffic.
The system is tightly tied to the Salesforce world; once you build your processes on it, switching providers later is difficult and expensive.

Similar, but not the same

Chatbot – A traditional chatbot provides pre-written answers, while an Agentforce agent plans the steps itself and executes operations within the company's systems.
Microsoft Copilot – Copilot is primarily an assistant that sits beside the worker in office programs, while Agentforce agents typically act on their own, handling matters directly with customers.
Einstein Copilot – This was Salesforce's earlier AI assistant that responded to requests, replaced by the more autonomous Agentforce.

Related concepts

Broader

AI-agent, Agentplatform, Enterprise AI, Customer Relationship Management (CRM) system

Narrower

Customer service agent, Sales agent, Atlas Reasoning Engine

On the edge

Salesforce, Microsoft Copilot Studio, Chatbot, Large language model, Human oversight

A brief history

Salesforce introduced Agentforce in September 2024 at its annual Dreamforce conference, and it became generally available in October 2024. Agentforce 2.0 was released in December 2024, Agentforce 3 in summer 2025, and in autumn 2025 the company reorganized its entire product portfolio around agents under the name Agentforce 360.

Where things stand

At the end of 2026, Agentforce is one of the best-known enterprise agent platforms, central to Salesforce's strategy, with thousands of corporate clients. However, most deployments are still limited to a narrow area, mainly customer service, and views on the return on investment are mixed.

Status: 2026 November

In the news

Appears in 1 Agent Age stories; the curve shows weekly occurrence in the news magazine.

AI agents in HR: from customer service to military HR, while many firms stay stuck in pilotsAug 4, 2026 · Work & HR

Also known as

Agentforce · Salesforce Agentforce · Agentforce 360 · Agentforce 2.0 · Agentforce 3 · AgentForce · Salesforce AI-agent · Salesforce agent

Sources

Salesforce – Agentforce (official product page)

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.