Agentforce is Salesforce’s platform for building and running AI agents that work on your CRM data, handling jobs such as customer service replies, inbound lead follow-up and campaign setup inside Salesforce. This page helps a B2B marketing, sales or development team decide whether Agentforce belongs in the stack, how its pricing model works, and which alternative, such as Microsoft Copilot Studio, to test against it.
What is Salesforce Agentforce?
Salesforce Agentforce is a set of tools for creating AI agents that act inside Salesforce products. An agent gets a role, a list of topics it may handle, written instructions, and actions it can take, such as looking up an order, updating a record or drafting an email. A reasoning engine that Salesforce calls Atlas decides which action to run next, checks the result and replies or hands the case to a person.
Before Agentforce, Salesforce AI features were sold under the Einstein brand, and the earlier Einstein Copilot assistant became part of the Agentforce line. Einstein still names some features, including the Einstein Trust Layer. That layer masks sensitive data, screens outputs for toxic content, keeps an audit trail, and is designed so that outside model providers do not keep your prompts. Check the vendor’s documentation for the current terms on data retention.
Agentforce runs on large language models from several providers, which Salesforce hosts or connects to on your behalf, and you can bring your own model through its model tools. Agents draw on CRM records, knowledge articles and Salesforce’s customer data layer, known as Data Cloud. You build them in a low-code builder inside Setup, and you deploy them to web chat, messaging apps, Slack, email and the Salesforce apps your staff already use.
Models, versions and what each is for
Agentforce is versioned as a platform rather than as a model. Salesforce has shipped several numbered releases since launch, from Agentforce 2.0 to Agentforce 3, and at the time of writing it markets the wider platform as Agentforce 360. Each release added builder features, more prebuilt agents, better testing and more channels. The release names matter less than which agents and editions your org can switch on.
The prebuilt agents map to Salesforce clouds. Service agents answer customer questions from your knowledge base and resolve simple cases. Sales agents include an SDR agent that replies to inbound leads and books meetings, and a coaching agent that runs role-play practice for reps. Marketing agents help build campaign briefs, audience segments and content inside Marketing Cloud. Commerce agents assist shoppers and merchants.
For staff, the Agentforce assistant sits inside Salesforce and Slack, summarizing records and drafting replies. Developers get Agentforce for Developers, a coding assistant for Apex and Lightning work in their editor. Names and packaging change often, so check the vendor’s product pages before you plan a rollout around any one agent.
What it does well for marketing, creative and development teams
Agentforce is strongest where clean CRM data and repeatable conversations meet. Service teams use it to answer common questions at any hour and pass complex cases to a person with a summary attached. Sales teams use the SDR agent to respond to inbound leads quickly, qualify them against set rules and offer meeting times. Because the agent reads the same records as your reps, its answers can reflect account history.
Marketing teams inside Salesforce can use agents to draft campaign briefs, suggest segments and produce first versions of emails and landing page copy, which a marketer then edits and approves. Developers and admins benefit from the low-code builder, which reuses existing Flows, Apex classes and APIs as agent actions, and from a testing center that runs agents against sample conversations before launch.
The weak spots are setup effort and dependence on data quality. A well-behaved agent needs clear topics, careful instructions and tidy records, which takes admin time. It also adds little for teams whose customer data lives outside Salesforce. Creative work beyond first drafts is better handled in dedicated writing and design tools.
Pricing and plans
Agentforce pricing has changed several times, so treat any description as a snapshot. At launch, customer-facing agents were billed per conversation. Salesforce then added a credit model that charges per action an agent takes, plus per-user licenses that cover employee-facing use, and it bundles Agentforce into premium editions of its Sales and Service products. Some editions include a starter allowance so you can test before buying more.
What moves you up a tier is volume, scope and data. Higher-volume service agents use more credits. Wider use across employees favors per-user licensing. Agents that rely on Data Cloud may add their own usage costs, and Agentforce generally needs one of the higher Salesforce editions underneath. Read the current details on the Agentforce pricing page and ask your account team for a written estimate based on your expected volumes, because prices change.
Agentforce vs the alternatives
| Tool | Made by | Best for | How you access it | Pricing model in words |
|---|---|---|---|---|
| Agentforce | Salesforce | Service, sales and marketing agents on Salesforce data | Inside Salesforce products and Slack | Usage-based credits plus per-user licenses and editions |
| Copilot Studio | Microsoft | Agents across Microsoft 365 and Dynamics 365 | Microsoft 365 apps, Teams and web channels | Per-user licenses plus usage-based billing |
| Breeze agents | HubSpot | Mid-market marketing, sales and service on HubSpot | Inside HubSpot | Included in paid hubs with credit-based usage |
| ServiceNow AI agents | ServiceNow | IT and employee service workflows | Inside the ServiceNow platform | Enterprise contracts with add-on packages |
| Custom agent on a framework | Your own team | Unusual workflows across several systems | Your own apps and APIs | Model API usage plus hosting and build time |
How to read this table: it shows where each option fits and how it is packaged, not which is best, because the right choice mostly follows where your customer data already lives.
The most common comparison is Agentforce vs Copilot. If your CRM is Salesforce, Agentforce has the shorter path, since it reads records, permissions and Flows natively. If your company runs on Microsoft 365 and Dynamics 365, Copilot Studio fits better, and it reaches documents, email and Teams with less integration work. Teams with both often use Copilot for internal productivity and Agentforce for customer-facing agents. Our Microsoft Copilot page covers the Microsoft side.
For mid-market teams on HubSpot, Breeze agents are the natural comparison, with lighter setup and fewer enterprise controls. When a workflow crosses many systems or needs unusual steps, a custom build gives more control. The guide what are AI agents explains that route, and Manus AI shows what a general task agent looks like outside any CRM.
Where it fits in agency work
For a B2B agency, Agentforce work starts with the client’s process, not the tool. A strategist maps the conversations worth automating, such as inbound demo requests or common support questions, and defines what a good outcome looks like. An admin or developer then builds topics, instructions and actions in a sandbox, using the client’s own knowledge articles and Flows.
Testing is where most of the effort goes. The team runs the agent against realistic conversations, reviews every answer, tightens instructions and sets clear handoff rules to a person. After launch, someone reviews conversation logs and credit usage on a regular schedule. The client’s team owns the agent’s answers and the decision to widen its scope.
When an agent has to reach beyond Salesforce, or the workflow needs custom logic, an AI agents development company can design and build it alongside the CRM work. For a broader view of the tools teams combine with it, see the best AI tools for business roundup and the AI and automation section.
What’s new
In July 2024, Salesforce AI Research introduced xLAM-1B, a small model it nicknamed the Tiny Giant, trained for function calling: choosing the right tool or API and filling in its inputs correctly. Salesforce reported that it beat much larger models from other labs on a function-calling benchmark. Function calling is the step that lets an agent act rather than only answer.
In July 2024, the same team described APIGen, a pipeline that generates and verifies function-calling training data, and released a dataset built with it for other researchers. Small, specialized models like this point toward agents that run faster and cheaper on narrow tasks. Salesforce presents its action-model research on the Salesforce AI Research site, but it does not document clearly which models run inside Agentforce.
In July 2024, Salesforce sold its assistant as Einstein Copilot, and the Agentforce name had not yet launched. The research of that period, focused on models that call tools reliably, is the groundwork for the agent platform that followed.
This section is refreshed when the vendor ships something that changes the picture; the date at the top of the post is the last check.
What to check before you adopt it
- Edition requirements: confirm which Salesforce editions and add-ons your org needs before Agentforce can be switched on.
- Data handling: review the Trust Layer settings, which model providers process your data, and where that processing happens.
- Admin controls: check permissions, topic limits, handoff rules and audit logs, and decide who may change an agent in production.
- Testing: run every agent in a sandbox against real conversation samples before it talks to customers.
- Cost monitoring: set up reports on credit or conversation usage so a busy month does not surprise the budget.
- Regional and language availability: confirm that the channels, languages and data residency options you need are offered in your markets.