What Is Salesforce Agentforce?

Salesforce Agentforce has quickly become one of the biggest parts of Salesforce’s AI strategy, but if you’ve only heard the name, it can be difficult to work out what Agentforce actually is, what it can do, and where to start. At its simplest, Agentforce is Salesforce’s platform for building and deploying AI agents that can understand requests, make decisions, and take action across your Salesforce environment.

This guide covers everything you need to know about Agentforce, from the basics of how agents work and the different types of agents Salesforce offers, to AgentExchange, Agentforce pricing, security, testing, and implementation. We’ve also included links throughout to our deeper explainers, practical tutorials, and technical guides, so you can dig into the areas that matter most to you. Whether you’re completely new to Agentforce or looking to get more out of an existing implementation, this is the place to start.

What Is Agentforce?

Agentforce is Salesforce’s platform for building autonomous AI agents: digital colleagues that can actually do things inside your CRM, rather than simply telling a human what to do next.

At its simplest, an agent can understand a request, work out what needs to happen, and then take action. That’s the important distinction from many of the conversational AI tools we’ve become used to. An Agentforce agent isn’t just generating a response. Within the boundaries you define, it can make decisions, carry out tasks, and adapt its approach depending on the situation.

You can create an agent from scratch or start with one of Salesforce’s templates. The real power comes from what you give that agent access to, and the guardrails you put around it.

Think about the difference between a satnav that tells you which junction to take and one that can actually reroute your journey, book your parking, and let someone know you’re running late. Salesforce’s earlier AI was closer to the first example. It could recommend what to do next, but a human still had to do it.

Agentforce is designed to take the wheel, making decisions and carrying out actions on your behalf.

Under the hood, there are a few key building blocks. Subagents define the areas of work an agent can handle, such as lead qualification or case escalation. Actions are the tools it can use to get that work done, such as updating a record or sending an email. Then there’s the reasoning engine, which works out which subagent and actions are needed as the conversation develops.

Just as importantly, Agentforce can work with your Salesforce data. This gives agents access to the information they need to provide relevant answers and take appropriate actions, rather than relying solely on what an LLM happens to generate.

That’s the pitch, anyway. Whether every agent delivers on it from day one of a real deployment is another question, and one worth asking before you hand over the keys to anything customer-facing.

The two areas where Salesforce has invested particularly heavily, and where you’re most likely to encounter Agentforce, are sales and service.

Agentforce for Sales

Agentforce for Sales, officially “Agentforce Sales”, is built around a fairly simple idea: a sales rep’s day contains a lot of selling, but also a lot of admin. And the admin is exactly the sort of work an agent should be able to take off their plate.

Two purpose-built Agentforce Sales Agents do much of the heavy lifting:

• Lead Nurturing Agent (formerly SDR) works inbound leads around the clock, answering product questions, handling objections, and booking meetings across channels including email, SMS, and WhatsApp. When the conversation needs a human, it hands the prospect over to a seller.
• Sales Coach lets reps practice against their actual deals, role-playing everything from discovery calls to negotiations before providing feedback. Think of it as a sales rehearsal partner that never gets tired of hearing the same pitch.

There’s also the Sales Workspace, bringing together opportunity recommendations, account research, and post-deal follow-up suggestions in one place. The goal is simple: give reps more time to sell and less time to update fields nobody looks at.

Agentforce for Service

Agentforce for Service, officially “Agentforce Service,” takes a similar approach but perhaps has an even more obvious use case. A lot of customer service is repetitive. Every minute a human spends answering “where’s my order?” is a minute they aren’t spending with the customer who actually needs their expertise.

The Agentforce Service Agent sits at the center of this. It’s a conversational agent designed to resolve routine queries end-to-end through self-service, while escalating to a human when it reaches something it shouldn’t handle alone.

Salesforce claims it can deflect up to 72% of routine inquiries, alongside faster resolution times and improved CSAT. As always, those are Salesforce’s numbers, so treat them as something to test against your own service data rather than taking them as gospel.

For the humans who are still on the front line, Salesforce has also redesigned the Service Console. The Service Assistant agent surfaces case history and next-best actions during a conversation, helping reps resolve cases faster with step-by-step guidance rather than digging for it themselves, while a supervisor-facing Command Center gives managers visibility across human and AI interactions from one place.

The idea isn’t to replace your service team. It’s to stop asking highly paid humans to spend their day doing the part of the job that was never a particularly good use of a human being in the first place.

What Can Agentforce Do?

It’s easy to think of Agentforce as a sales and service product. That’s where Salesforce has focused much of its messaging. But look at what’s actually available, and the use cases stretch well beyond the contact centre.

The Out-of-the-Box Agents

Salesforce offers a growing range of ready-made agents, each designed around a particular job:

• Employee Agent: gives employees access to company knowledge, carries out tasks on their behalf, and helps streamline everyday work.
• Lead Nurturing Agent: formerly known as SDR, this agent engages leads with personalized content, answers common questions, and books meetings, leaving sales reps to step in when there is a real opportunity to close.
• Sales Coach: gives reps personalized, actionable feedback based on their sales pitch or a role-play session. Think of it as a practice partner that never needs a meeting room.
• Service Agent: handles common customer enquiries and passes more complex issues to a human. No rigid decision trees required.
• Service Assistant: helps service reps resolve cases faster by providing case summaries and step-by-step guidance. It works alongside the rep rather than in front of the customer.
• Setup with Agentforce: helps Salesforce Admins with Setup tasks such as managing users, troubleshooting issues, and customizing the org. Essentially, a second pair of hands for the parts of the admin job that rarely make the highlight reel.

Build Your Own Agent

The pre-built agents won’t cover every business process, particularly when you start getting into the quirks of how your organization actually works.

That’s where Agent Builder comes in. It gives you a low-code way to create a custom agent using Subagents, which define the areas the agent can work in, and Actions, which define what it can actually do.

Those Actions can be powered by Salesforce tools such as Flows, Prompt Builder, and Apex, or extended beyond Salesforce using MuleSoft APIs.

The difference is important. You aren’t just building an agent that can answer questions about orders. You can build one that understands your orders, follows your fulfilment rules, and knows when to hand something over to your account manager.

The Reasoning Engine Does the Deciding

Underneath all of this is the Atlas Reasoning Engine.

Atlas is responsible for working out what needs to happen when an agent receives a request. Rather than simply predicting the next response, it can break a request into smaller tasks, work through them, and check the results before taking action.

It’s also what helps ground Agentforce in your Salesforce data, including CRM and Data Cloud data. That’s important because an agent needs to work from what is actually true about your customer or business, rather than producing an answer that simply sounds convincing.

Beyond the Chat Window

Agentforce has also moved well beyond a text box on a website.

• Agentforce Voice brings the same agent capabilities to phone conversations, including real-time transcripts that managers can review.
• MCP (Model Context Protocol) allows Agentforce to connect with tools and systems outside Salesforce using a standardized approach, rather than requiring every integration to be built from scratch.

And with multi-agent orchestration, different agents can work together on the same request. A service agent might identify a billing problem, for example, and pass the relevant work to a finance-focused agent without a human manually connecting the two.

How Many Companies Use Agentforce?

Salesforce says more than 18,000 companies are now using Agentforce, including OpenTable, SharkNinja, Indeed, and Finnair. That’s a Salesforce figure rather than an independently audited number, but it does show how quickly the product has moved beyond the early pilot stage.

Gartner has also named Salesforce a Leader in its 2026 Magic Quadrant for Conversational AI Platforms.

Whether that matters to you probably depends on what you’re trying to do. It could be a useful line in a business case. It doesn’t, by itself, mean you should switch on every agent in your org tomorrow.

Key Components of an Agent

Before you start building an Agentforce agent, it helps to understand what actually makes one work. There are a few key pieces involved: the agent itself, the subagents and actions it can use, the data it can access, the channels it can operate through, and the reasoning engine that decides what happens next. Put those pieces together, and you get something considerably more capable than a chatbot waiting for someone to ask it a question.

Subagents and Actions

Actions are the tools an agent uses to actually do something.

An action might retrieve information from Salesforce, update a record, create a task, draft an email, or carry out some other piece of work. Salesforce provides standard actions for common use cases, but you can also create your own when you need an agent to do something specific to your business.

Subagents (previously called Topics) sit one level above actions. They group together the actions and instructions needed for a particular job.

Take a sales agent, for example. You might have a subagent focused on deal management, with actions that allow it to find opportunities and contacts, prepare a rep for their day, create follow-up tasks, and log calls.

The subagent provides the context and instructions for that particular area of work. The actions are the tools it can use to get the job done.

This distinction becomes particularly useful when you’re building more complex agents. Instead of giving one agent a huge list of things it can potentially do, you can organize its capabilities around specific jobs.

Data

An agent is only as good as the information it can access.

Agentforce is built on the Salesforce Platform, so agents can work with the CRM data you’ve made available to them. But you’re not restricted to standard Salesforce records.

You can also ground agents in sources such as Knowledge articles, Salesforce fields, uploaded files, and web content. Agentforce Data Libraries and the Search the Web action can help bring these sources into an agent’s responses and actions.

For more complicated unstructured data, you can also use Retrieval Augmented Generation (RAG) to retrieve relevant information before the agent responds.

The important thing here is that you control what the agent can access. Giving an agent access to more data isn’t automatically better. The goal is to give it the right information for the job it needs to do.

Connections and Channels

An agent needs somewhere to interact with people, and that’s where channels come in.

A channel is essentially the place where your agent lives. That could be a customer-facing chat experience, an employee interface, a messaging channel, or a voice conversation.

Channels also matter when a conversation needs to be handed over to a human. A customer asking a straightforward question might be handled entirely by an agent, while something sensitive or complicated can be routed to the right person.

When you deploy an agent to a channel, Agentforce creates a connection that manages how the agent operates there.

Connections handle things such as how responses are formatted, how multimedia like images, buttons, links, and videos are presented, and how conversations are routed between the agent and humans.

The useful bit is that you don’t need to rebuild the agent for every channel. You can develop the agent once and then connect it to different experiences, while still configuring each channel to suit the way people use it.

The Reasoning Engine

So you’ve given your agent instructions, tools, and data. How does it actually decide what to do?

That’s the job of the reasoning engine.

When an agent receives a request, the reasoning engine works out which subagent is relevant, considers the instructions and information available to it, and determines which actions need to be taken.

Agentforce uses the Atlas Reasoning Engine, which takes a graph-based approach to reasoning. It uses Agent Script to separate the overall workflow of an agent from its conversational skills.

This creates what Salesforce calls hybrid reasoning. In practice, that means combining the flexibility of an LLM with more predictable, rules-based execution.

That’s an important distinction for Salesforce professionals. You don’t necessarily want an agent improvising its way through every business process. Some decisions need the flexibility of AI, while others need to follow a defined set of rules.

The LLM

The final piece is the large language model (LLM).

This is the part responsible for understanding what a user is asking, communicating naturally, and helping the agent work out what to do.

The reasoning engine calls the LLM throughout an interaction when it needs that capability. Exactly how many times it does this, and how much information each call contains, depends on the complexity of the task and which subagents and actions are involved.

So while it’s tempting to think of an Agentforce agent as simply “an LLM connected to Salesforce”, there’s quite a bit more going on underneath.

The LLM provides the language and reasoning capability. The Atlas engine coordinates that reasoning. Subagents define the areas of work. Actions give the agent its tools. Data gives it the information it needs, and channels determine where people can interact with it.

That’s the basic architecture you need to understand before you start building.

Features of Agentforce

Agentforce isn’t a single product with one interface. It’s a collection of tools, each with a different job. Here’s what you need to know.

Agentforce Builder

Agentforce Builder is where agents are built, tested, and deployed. It’s a low-code environment that connects agents to Salesforce data, Flows, Apex, MuleSoft APIs, and channels including chat, voice, Slack, and self-service portals.

You can describe an agent in natural language, start from a template, or use Agent Script when you need more control. Builder also includes simulation, reasoning traces, and batch testing to help you understand how an agent will behave before it goes live.

Agent Script

Agent Script adds predictability to Agentforce. It lets you define things like variables, if/else logic, handoffs, rules, and sequences of actions, while leaving the LLM to handle areas where more flexibility is useful.

In other words, you can let the agent reason where it makes sense, while putting firm rules around anything that needs to happen in a specific way.

Prompt Builder

Prompt Builder is used to create reusable prompt templates that pull in Salesforce data, including record fields, related lists, Flows, and Apex. These can generate content such as case summaries or field descriptions, and can be used from record pages, Flow, Apex, or as agent actions.

Intelligent Context

Intelligent Context helps agents work with unstructured information such as PDFs, spreadsheets, scanned documents, and flowcharts. It extracts and structures this content, then indexes it through Data 360 so agents can use it as part of their reasoning.

Agentforce Grid

Agentforce Grid brings AI into a spreadsheet-style interface. Each row represents a job, while columns apply actions, calculations, data updates, or AI steps. It’s designed for working with large volumes of data without building a custom process for every batch.

Agentforce Voice

Agentforce Voice brings Agentforce to phone conversations, allowing customers to have natural conversations with an agent rather than navigating a traditional phone menu. Agents can access CRM data, update records, trigger workflows, and hand conversations to humans with the relevant context intact.

Setup With Agentforce

Setup with Agentforce is an AI assistant for Salesforce Admins. It can help with tasks such as managing users, troubleshooting issues, and configuring an org. Unlike other Agentforce tools, it’s automatically available in every org rather than something you build yourself.

Agentforce Labs

Agentforce Labs is Salesforce’s experimental space for Agentforce technology that isn’t quite ready for prime time. It includes early tools and projects such as coding agents, GUI automation, and open-source Agentforce skills. It’s worth keeping an eye on if you want to see where the platform is heading next.

Agentforce Trust and Security

Giving AI access to your CRM is one thing. Giving an autonomous agent the ability to update records, send emails, and interact with customers is another. That makes trust and security a critical part of any Agentforce implementation.

Agent Guardrails

Guardrails define what an agent can and cannot do, including when it should stop and hand a conversation to a human. Salesforce recommends assessing an agent across People, Business, Technology, and Data before deployment, then using those findings to define guardrails, screen inputs, and validate the agent’s behavior after launch.

Data Protection

The Einstein Trust Layer provides the security controls underneath Agentforce. Data is encrypted in transit, at rest, and during use, while sensitive information can be masked before prompts are sent to third-party models. Salesforce also recommends sandboxing or containerizing agents to limit the impact if something goes wrong.

Permissions and Monitoring

Agents should operate within the same permissions as the user they are acting for, including role hierarchies and field-level security. Salesforce also provides audit trails, monitoring, role-based access controls, and tools such as Security Center and Privacy Center to help identify and investigate unusual behavior or potential data exposure.

These controls don’t make Agentforce foolproof. They do mean that deploying an agent securely requires more than switching it on. Permissions, guardrails, data governance, and ongoing monitoring are all part of the implementation.

Testing Agents and Agentforce

Testing a button is easy: click it, check the result, move on. Testing an AI agent is different. The same question can produce different responses, and there isn’t always one “correct” answer.

That’s why testing needs to look at more than whether an agent returned the expected words. You also need to check whether it chose the right subagent, took the right actions, and produced an acceptable outcome.

Agentforce Testing Center

Agentforce Testing Center lets you test agents at scale before they reach production. It can generate synthetic test cases and run them against an agent to evaluate things such as subagent selection, action sequences, and response quality. Salesforce also supports AI-generated test cases, helping teams cover more of the ways customers might interact with an agent.

Testing should be carried out in a sandbox, as agent tests can modify CRM data. You can create and upload test cases using Testing Center’s CSV format, while developers can use Agentforce DX and YAML test specs as part of a more automated testing workflow.

The aim isn’t to prove that an agent will always give exactly the same answer. It’s to build confidence that, across a wide range of inputs and scenarios, it behaves within the boundaries you’ve defined.

AgentExchange

You can use AgentExchange, previously known as AppExchange, to discover AI agents, skills, apps, integrations, Slack solutions, and professional services.

The name change is part of Salesforce’s bigger shift towards the agentic enterprise, but AgentExchange isn’t simply AppExchange with a new name. Salesforce is positioning it as a broader marketplace for the tools and expertise organizations need to build out their Salesforce ecosystem, including pre-built agents and skills for Agentforce.

If you’re a Salesforce Admin or developer, don’t worry. The AppExchange you know hasn’t disappeared. The apps and solutions you’ve always been able to find are still there, now sitting alongside a growing collection of agents and agent-related tools.

In other words, AgentExchange is becoming the place to go when you’re looking for something to extend what Salesforce can do, whether that’s an app, an integration, a service, or an AI agent.

How Does Agentforce Pricing Work?

Agentforce pricing has changed rapidly since launch, moving away from traditional seat-based pricing towards paying for what an agent actually does.

Salesforce initially launched Agentforce at $2 per conversation, but quickly moved to another kind of consumption-based pricing, built around Flex Credits.

Flex Credits

Flex Credits are now the core unit for most Agentforce usage. Packs of 100,000 credits cost $500, with most agent actions costing 20 credits, or around $0.10 per action. Voice actions cost 30 credits.

Customers can buy credits through three models:

• Pre-Purchase: Pay upfront for a set volume, generally the cheapest option for predictable usage.
• Pre-Commit: Commit to a baseline spend to access better rates, then pay for actual monthly usage.
• Pay-as-you-go: Pay monthly for what you actually use, with no upfront commitment.

Salesforce also offers Flex Agreements, which allow customers to shift spending between traditional user licenses and Flex Credits as their mix of human and digital workers changes.

Agentforce per User

Salesforce also offers seat-based options for employees:

• Agentforce User License: Around $5/user/month, with Flex Credits still used for metered activity.
• Agentforce Add-ons: From $125/user/month for unmetered Agentforce usage.
• Agentforce 1 Editions: From $550/user/month, combining the add-on with a pool of Flex Credits.

Pay-per-Resolution

The latest shift is pay-per-resolution, introduced in June 2026 for Help Agent and Agentforce Customer Service Portal.

Rather than charging for every interaction, Salesforce charges when an agent successfully resolves a customer’s issue autonomously. If the customer is escalated to a human or doesn’t get a resolution, there is no charge.

In other words, Salesforce is increasingly moving from paying for activity to paying for outcomes.

What Does This Mean for Your Budget?

The biggest thing to remember is that Agentforce isn’t something you can budget for based on a simple per-user price. Your costs will depend on factors such as agent activity, complexity, and resolution rates.

Salesforce’s Digital Wallet provides usage tracking and spend alerts, but you’ll still need to model expected usage carefully.

And given how frequently Agentforce pricing has changed since launch, it’s worth checking the latest pricing with Salesforce before committing to a contract.

Can I Get Agentforce for Free?

In early 2026, Salesforce announced it was bringing Agentforce capabilities into its SMB-focused CRM editions, including its no-cost offering for small and medium-sized businesses. This means AI and agentic functionality are now being introduced across the Salesforce Suites, including the Free, Starter, and Pro Suites.

You’re restricted to the capabilities Salesforce has prebuilt. If you’re looking to fully leverage Agentforce with customized agents, these free AI features won’t be sufficient, as they don’t include access to tools like Agent Builder or Prompt Builder.

Get Agentforce Certified

To test and prove your Agentforce prowess, consider doing the Agentforce Specialist certification. The Salesforce Agentforce Specialist Certification is designed for professionals who manage or implement Agentforce within their organizations. It validates your ability to design, configure, and optimize AI agents that deliver real business value.

To succeed, you’ll need a solid understanding of how to configure and customize Agentforce tools to meet different business scenarios, along with an appreciation for data quality and ethical AI practices.

You don’t need to be a coder or an AI engineer to pass, but don’t underestimate this exam. It is more challenging than it looks, and hands-on experience with Agentforce is essential if you want to feel confident on exam day.

Agentforce Documentation and Resources

If you’re getting started with Agentforce, Salesforce’s documentation is probably going to become one of your most-used resources. The Agentforce documentation covers everything from the basics of creating and configuring agents to more advanced topics such as subagents, actions, security, testing, and deployment.

Salesforce also has a growing collection of Trailhead modules, help articles, developer documentation, and other resources covering different parts of the Agentforce platform.