Customer service reps spend hours copying data between systems. Sales teams let hot leads go cold because follow-ups slip through the cracks. Salesforce thinks the fix isn’t another chatbot — it’s an AI that can actually take action. That’s the promise of Agentforce, the company’s new platform for building and deploying autonomous AI agents. Announced at Dreamforce 2024, Agentforce aims to move beyond conversational AI to something more useful: agents that reason, plan, and execute tasks inside your CRM.
What Exactly Is Salesforce Agentforce?
At its core, Agentforce is a set of tools for creating AI agents that live inside Salesforce. Unlike the Einstein Copilot you might have used, which waits for prompts, Agentforce agents are proactive. They can be triggered by events like a new case, a lead that goes quiet, or an order that’s stuck, and then work through a series of steps to resolve the issue. The system is built on the Atlas reasoning engine, which breaks down a goal into subtasks, decides which actions to take, and uses your company’s data to ground its responses. That data comes from Data Cloud, Salesforce’s unified data platform, so agents have a full picture of the customer. As Salesforce puts it, these are AI agents that do the work, not just the talking.
How Agentforce Differs from Traditional Chatbots
Most chatbots are simple decision trees. They match keywords to canned responses. Even newer LLM-powered bots often only answer questions — they can’t actually do anything. Agentforce changes that by giving agents the ability to interact with Salesforce records, send emails, update fields, and even call external APIs. Here’s what sets it apart:
- Autonomy: Agents can initiate actions without a human prompt, based on rules or triggers.
- Multi-step reasoning: They can handle complex, multi-turn conversations that require checking inventory, processing returns, and issuing refunds.
- Integration: Agents work across Salesforce apps, Slack, and custom channels.
- Grounded in your data: Responses are based on your CRM data, not generic web knowledge.
Consider a return request. A traditional bot might tell the customer the return policy. An Agentforce agent can verify the order, check if the item is eligible, generate a return label, process the refund, and update the inventory — all without a human touching it.
The Technology Under the Hood
Agentforce isn’t just a thin wrapper on a large language model. It’s a full stack:
- Atlas Reasoning Engine: This is the brain. It takes a high-level goal, breaks it into steps, and selects from a library of actions (called ‘skills’) to achieve it. It can also decide when to escalate to a human.
- Data Cloud: Agents need context. Data Cloud harmonizes data from Salesforce and external sources, so an agent knows a customer’s purchase history, support tickets, and email engagement.
- Einstein Trust Layer: Security is a big concern with AI. The Trust Layer masks sensitive data, prevents it from being stored by the LLM, and audits every interaction.
- Agent Builder: A low-code studio for creating and customizing agents. Admins can define topics, actions, and guardrails without writing code.
Security around AI agents is evolving fast. As Okta’s CEO recently noted, the next frontier of security is AI agent identity — meaning we’ll need ways to authenticate and authorize non-human actors just like we do for employees.
Real-World Use Cases
Salesforce has been rolling out Agentforce across its clouds. Here are some practical applications:
- Service: The Service Agent can handle common support requests, like password resets, order status, and returns. Salesforce says its own support team used Agentforce to resolve about 1 million conversations, with a significant portion handled entirely by AI.
- Sales: The Sales Agent can qualify leads, schedule meetings, and update opportunities. It can also draft personalized follow-up emails based on recent interactions.
- Marketing: Agents can segment audiences, trigger campaigns, and even personalize content at scale.
- Commerce: Shopping assistants can guide customers through product selection, answer questions, and complete purchases.
- Slack: Agentforce is deeply integrated with Slack, following Salesforce’s AI-heavy makeover for Slack with 30 new features. Employees can ask an agent to pull up a case, update a record, or summarize a conversation without leaving Slack.
Pricing and Availability
Agentforce became generally available in October 2024. Pricing is consumption-based: $2 per conversation, with volume discounts. A ‘conversation’ is defined as a session between a user and an agent, whether it’s a chat, email, or other interaction. Salesforce also offers Flex Credits, which can be used across Agentforce and other AI features. The platform is available in Enterprise, Performance, Unlimited, and Einstein 1 Editions, though some features require add-ons. For businesses already on Salesforce, the barrier to entry is low — you can start with pre-built agents and customize later.
What This Means for Businesses and Admins
The rise of autonomous agents changes the role of Salesforce admins and developers. Instead of just building reports and flows, they’ll be designing agent behavior, writing prompts, and setting guardrails. Data quality becomes even more critical — agents are only as good as the data they’re grounded in. Companies will also need to think about governance: who can create agents? What actions are they allowed to take? How do you audit their decisions? These are new questions, and Salesforce is building tools to answer them, but the policies are still catching up.
Getting Started with Agentforce
If you’re curious about Agentforce, you don’t need to boil the ocean. Start small:
- Identify a high-volume, repetitive task — like password resets or lead qualification.
- Clean up the relevant data in Salesforce so agents have accurate information.
- Use Agent Builder to create a simple agent with a few actions.
- Test with a small group of users, gather feedback, and iterate.
- Monitor performance and costs closely, since pricing is per conversation.
Salesforce offers Trailhead modules and a free trial for Agentforce, so you can experiment without a big commitment. The platform is still evolving, but the direction is clear: AI that doesn’t just talk, but actually does the work. For teams buried in repetitive tasks, that’s a welcome shift.