Key takeaways
- Agentforce lets you build AI agents that can reason over your Salesforce data and take actions, not just answer questions.
- Pricing is flexible: per conversation, consumption-based Flex Credits, or per-user add-ons for unmetered employee use.
- The best first use cases are narrow, frequent and well documented, such as answering order or appointment questions.
- Agent quality depends on your data, knowledge articles and permissions far more than on prompts.
Salesforce made Agentforce generally available in October 2024 and has expanded it quickly since, most recently with Agentforce 360, announced as generally available in October 2025. For small and mid-size businesses the question is practical: what can it do for us, what will it cost, and how do we start without betting the budget?
What Agentforce is, in plain terms
An Agentforce agent is an AI assistant that understands a request, decides what to do, and carries out actions inside Salesforce and connected systems. Earlier chatbots followed fixed scripts. An agent uses a reasoning engine, Salesforce calls it Atlas, to choose from the topics and actions you define, such as looking up an order, updating a case, booking an appointment or drafting an email.
Agents can be customer-facing, for example on your website, or employee-facing, helping staff inside Salesforce. Both run on your Salesforce data, permissions and automation, and both are governed by the Einstein Trust Layer, which handles protections such as data masking and zero data retention agreements with the AI model providers.
Where it delivers value today
- Customer service: answering order status, appointment, billing and policy questions, and handing off to a person with context when needed.
- Sales development: responding to inbound leads, answering product questions and booking meetings.
- Internal productivity: summarizing accounts and cases, drafting replies and finding information across records and knowledge.
- Operations: triggering routine processes, such as creating follow-up tasks or updating records, from natural-language requests.
How pricing works
Salesforce offers several ways to pay, and the right one depends on how the agent is used. As of this writing, Salesforce’s published list prices include:
| Model | List price | Best for |
|---|---|---|
| Conversations | $2 per conversation | Customer-facing agents with predictable conversation volumes |
| Flex Credits | $500 per 100,000 credits; a standard action uses 20 credits (about $0.10) | Consumption-based use across agents and actions |
| Agentforce add-ons | From $125 per user per month | Unmetered employee use within a specific cloud, such as Sales or Service |
| Agentforce 1 Editions | From $550 per user per month | Organizations wanting the full platform bundled with their edition |
Salesforce pricing changes regularly and contract prices often differ from list prices. Confirm current pricing with Salesforce before budgeting.
A low-risk way to start
- Pick one narrow use case with high volume and clear answers, such as “Where is my order?” or “Can I reschedule?”.
- Fix the knowledge first. Make sure the knowledge articles and data the agent needs are accurate and current.
- Define topics and actions carefully, with clear instructions and guardrails about what the agent must not do.
- Test with real questions taken from past emails, chats and cases, including awkward ones.
- Launch with a human handoff and review conversations weekly in the first month.
- Measure outcomes such as resolution rate, handoff rate, time saved and customer satisfaction, then expand.
Common pitfalls
- Starting with a broad “answer everything” agent instead of a focused one.
- Connecting an agent to messy, duplicated or outdated data.
- Giving the agent’s user more permissions than it needs.
- Skipping testing with realistic, unusual and adversarial questions.
- Treating launch as the finish line rather than the start of tuning.
Do you need Data 360?
Salesforce renamed Data Cloud to Data 360 in October 2025. It unifies data from many sources and can ground agents in richer context. Many first agent projects work well with core Salesforce data and knowledge articles alone. Data 360 becomes valuable when the agent needs data from several systems, such as an ERP, e-commerce platform or data warehouse.
Frequently asked questions
Is Agentforce only for large enterprises?
No. Consumption-based pricing makes it possible to start small with a single use case, which suits small and mid-size businesses.
Will an Agentforce agent replace our support team?
In practice, agents handle routine, repetitive requests and hand complex or sensitive ones to people with full context, which frees the team for higher-value work.
Is our data used to train the AI models?
Salesforce’s Einstein Trust Layer includes zero data retention agreements with its model providers. Review Salesforce’s current trust documentation for the details that apply to your contract.
