Key takeaways
- Agentforce is Salesforce’s platform for AI agents that can answer questions and take actions using your CRM data, within rules you set.
- Good first use cases are narrow and high-volume: answering service questions, qualifying leads, or summarising records for staff.
- Pricing comes in several models, including per-user add-ons, Agentforce 1 editions and consumption-based Flex Credits.
- Clean data, clear guardrails and a small pilot matter more than the model itself.
What is Agentforce?
Agentforce is Salesforce’s platform for building and running AI agents. An agent can understand a request in natural language, look up information in Salesforce, and carry out defined actions, such as updating a record, creating a case or booking a meeting. Agents work for customers on your website or messaging channels, or for employees inside Salesforce and Slack.
Salesforce launched Agentforce in late 2024 and has expanded it since. At Dreamforce 2025 it introduced Agentforce 360 as the umbrella for its agent platform, and renamed Data Cloud to Data 360, the data layer agents use to work with unified customer data.
How an Agentforce agent works
Each agent is configured from a few building blocks:
- Topics: the jobs the agent handles, such as “order status” or “appointment booking”.
- Instructions: plain-language rules about how to handle each topic and what to avoid.
- Actions: what the agent can do, built from Flows, Apex, prompt templates or API calls.
- Data: the Salesforce records and knowledge articles the agent can use to ground its answers.
- Guardrails: permissions, escalation rules and the Einstein Trust Layer, which governs how data is handled when it is sent to the AI model.
The practical lesson is that an agent is only as good as the processes and data behind it. If your Flows are unreliable or your knowledge base is out of date, the agent will repeat those problems faster.
Practical use cases
| Area | Example agent task | What it needs |
|---|---|---|
| Customer service | Answer order, booking or policy questions and create a case when it cannot help | Accurate knowledge articles, case routing, escalation rules |
| Sales | Respond to inbound enquiries, qualify leads and book meetings | Lead data, qualification criteria, calendar integration |
| Internal productivity | Summarise an account, draft follow-up emails, prepare meeting notes | Well-structured account and activity data |
| Field and operations | Check appointment status or reschedule visits | Scheduling data and clear rules |
For UAE businesses serving customers in Arabic and English, test both languages early with real customer phrasing, including mixed-language messages, and keep a clear handover to a human.
How Agentforce pricing works
Salesforce has offered several pricing models since launch. Based on published analyses of Salesforce’s pricing in 2026:
- Flex Credits: a consumption model introduced in May 2025. Each standard agent action consumes credits; analysts report a list price of USD 500 per 100,000 credits, which works out to around USD 0.10 per standard action.
- Per-conversation pricing: the original model at launch, at USD 2 per conversation.
- Per-user add-ons: Agentforce add-ons for Sales, Service and other clouds, listed at about USD 125 per user per month for internal use.
- Agentforce 1 editions: top-tier bundles listed from USD 550 per user per month that include Agentforce, Data 360 credits and other features.
Prices and included allowances change, so confirm current terms with Salesforce before you build a business case. The bigger budgeting question is usage: estimate how many conversations or actions you expect each month, and test that estimate during a pilot.
A four-step plan to start
- Pick one narrow use case. Choose a high-volume, low-risk task where answers are already documented.
- Fix the foundations. Clean the relevant data, update knowledge articles and make sure the underlying Flows work reliably.
- Pilot with guardrails. Limit topics and actions, define escalation to a human, and review conversations every week.
- Measure and expand. Track resolution rate, handover rate and customer feedback, then add topics one at a time.
Data and compliance considerations
Before you connect customer data to an AI agent, check which data protection regime applies to your entity, what data the agent can access, and how conversations are stored. In the UAE, mainland companies are covered by the federal PDPL, while DIFC and ADGM entities follow their own rules. Apply least-privilege permissions to agent users, as you would to any staff member.
Getting help
Our Salesforce consulting team can help you choose a first use case, prepare the data and processes, and design a pilot. If you need hands-on build capacity, you can also hire a Salesforce developer to build the Flows and actions behind your agents.
Frequently asked questions
Is Agentforce the same as Einstein?
Einstein is Salesforce’s earlier brand for predictive and generative AI features. Agentforce is its platform for autonomous AI agents that can take actions. Some Einstein capabilities now sit within the Agentforce platform.
Do we need Data 360 to use Agentforce?
Agents can work with standard Salesforce records and knowledge. Data 360 becomes useful when agents need unified data from several systems. Check the requirements of each feature you plan to use.
Can Agentforce answer customers in Arabic?
Language support depends on the features and channels you use. Test Arabic and mixed Arabic-English conversations during your pilot and confirm current language support with Salesforce.
How do we control what an agent can do?
Through the topics, instructions and actions you configure, the permissions of the agent user, and escalation rules that hand conversations to people when needed.
Exploring AI agents in Salesforce?
We can help you pick a first use case and prepare the data and automation it depends on.
Related articles
Sources