What does Agentforce actually do?
An Agentforce agent reads a request, works out what is being asked, pulls the right information from your business data and then either answers or does the job, such as updating a record or booking a meeting.
Salesforce defines an AI agent as software that uses generative AI to decide what to do next and how to do it. Unlike a tool that only drafts text from a single prompt, an agent plans, calls the systems it is connected to, and keeps a record of context and history.
Salesforce offers ready-made agents for service, sales, marketing and commerce, and lets you customize them or build new ones. Its own examples include a service agent that troubleshoots an issue from purchase history and company knowledge, and an SDR agent that emails a new lead, answers a pricing question and books a follow-up on a sales rep's calendar.
How is an AI agent different from a chatbot?
A traditional chatbot follows a pre-built conversation tree. An agent reasons about the conversation, builds a plan, looks at the tools it has and picks the best next step.
| Traditional chatbot | Agentforce agent | |
|---|---|---|
| How it decides | Follows a fixed dialog tree | Reasons about the request and plans the steps |
| What it can do | Answer from scripted replies | Answer and take actions (flows, Apex, APIs) |
| Data it uses | Whatever was scripted | Your Salesforce records and knowledge, through Data 360 and data libraries |
| Off-topic or unsafe requests | Often falls through to a fallback | Kept to approved subagents; harmful and prompt-injection attempts are detected |
How does Agentforce work behind the scenes?
Five pieces work together: the Atlas Reasoning Engine plans, subagents and instructions define the jobs, actions do the work, Data 360 and data libraries supply the facts, and the Einstein Trust Layer protects the data along the way.
- Atlas Reasoning Engine. Salesforce calls it the brain behind Agentforce. It breaks a request into smaller tasks, evaluates each step and proposes a plan until the answer or action is complete.
- Hybrid reasoning and Agent Script. Salesforce combines flexible LLM reasoning with deterministic, rules-based steps in the same engine, so a task such as a payment or identity check can be pinned to an exact sequence the agent cannot skip.
- Subagents, instructions and actions. Subagents (earlier Salesforce pages call them topics) define the jobs an agent will and will not do. Plain-language instructions guide each job, and actions are the things it can run, built from your existing flows, prompt templates, Apex and APIs.
- Data 360 and data libraries. Data 360 (formerly Data Cloud) unifies structured and unstructured data so agents can use retrieval-augmented generation. Data libraries ground an agent in knowledge articles or uploaded files.
- Einstein Trust Layer and guardrails. Dynamic grounding, secure data retrieval that respects permissions, zero data retention with the LLM providers, data masking, toxicity detection and an audit trail.
Where do you build and test an agent?
Agents are built in Agentforce Builder, Salesforce's low-code builder, and tested at scale in Agentforce Testing Center before they go live.
Salesforce says admins and developers can use natural language in the builder to define the jobs to be done, instructions, actions and guardrails. Our Agentforce Testing Center guide explains how batch testing works and why it should run in a sandbox.
Where can an Agentforce agent talk to people?
Salesforce lists the CRM itself, your website, WhatsApp and Messenger among the channels, and Agentforce Voice adds phone calls.
Salesforce also describes sales reps working with Agentforce from Slack. See Agentforce Voice for the phone side.
What is Agentforce 360?
Agentforce 360 Platform is the name Salesforce uses for the platform layer: designing, deploying and orchestrating agents on the Salesforce Platform with Data 360, metadata, security and observability built in.
Is Agentforce suitable for a small business?
Salesforce says Agentforce suits businesses of any size, and customers on Enterprise Edition or higher can try it through Salesforce Foundations, an add-on Salesforce offers at no extra charge.
In practice the deciding factor is less your size than the state of your Salesforce data and whether you have one clear, repetitive job for an agent to do. Our readiness checklist covers what to check first, and the pricing guide explains how usage is billed.
Frequently asked questions
What is Salesforce Agentforce in one sentence?
Agentforce is Salesforce's platform for building AI agents that use your Salesforce data to answer questions and take actions, within guardrails you define.
What is the Atlas Reasoning Engine?
It is the reasoning layer Salesforce describes as the brain behind Agentforce. It breaks a request into smaller tasks, gathers the relevant data, proposes a plan and checks each step until the answer or action is complete.
Does Agentforce need Data Cloud?
Salesforce now calls Data Cloud Data 360. Salesforce's Trailhead material shows Data 360 powering retrieval over unified data, and data libraries grounding agents in knowledge articles or files. Which you need depends on your use case; check your org's licenses with Salesforce.
Is Agentforce the same as Einstein?
Einstein is Salesforce's earlier AI brand, and some Agentforce features still carry it, such as the Einstein Trust Layer. Several products that launched as Einstein agents were renamed to Agentforce, for example Einstein SDR Agent became Agentforce SDR.
Need help with Agentforce?
We help small and mid-size businesses choose a first Agentforce use case, prepare their Salesforce data, build and test the agent and train the team. See our Salesforce Agentforce consulting page, or our Salesforce guides and FAQs.