The Agentforce Specialist Certification is tailored for professionals who want to leverage the capabilities of generative AI within the Salesforce platform. Those in this role are key to implementing and managing Einstein AI solutions, optimizing business processes with the help of advanced tools like Copilot Builder, Prompt Builder, and Model Builder.

This certification is particularly suited for Salesforce Administrators, Developers, and Architects, who utilize both standard AI functionalities and custom enhancements to drive innovation within their organizations.

There are 5 areas of knowledge that are covered by the Salesforce Agentforce Specialist certification.

Agentforce Specialist Topic Weighting Chart

The following are the core topic areas of the  Agentforce Specialist certification and what you’re expected to know:

This topic includes the following objectives:

Agents, such as the Agentforce (Default) agent, can leverage standard and custom actions to execute tasks on behalf of Salesforce users. Standard actions are included with agents by default and help users complete common tasks. Custom actions, on the other hand, offer flexibility to tailor the agent’s capabilities to specific business needs. When creating a custom action, an AI specialist can select a Reference Action Type, which can be Apex, Flow, or Prompt Template. Agent actions can be assigned to topics in Agent Builder. Instructions can be specified to define how a custom action should be used during conversations. Properly managing and assigning these actions ensures that agents deliver the most relevant and effective responses to user requests.

The reasoning engine and the large language model (LLM) play a critical role in executing agent actions based on user requests. The reasoning engine is responsible for orchestrating this process by launching topics and actions, ensuring the right tasks are performed to meet the user's request. The LLM identifies the user's intent, determines the best matching actions in the right order, and generates an appropriate response while maintaining the conversation flow. Additionally, session event logs in the Agent Builder can be accessed to debug and analyze the sessions associated with the execution of copilot actions. Einstein Copilot supports OpenAI GPT-4o for planner service calls. Copilot actions support making calls to other predefined LLMs.

Agents, such as the Agentforce Employee agent, can leverage standard and custom actions to execute tasks on behalf of Salesforce users. Standard actions are included with agents by default and help users complete common tasks. Custom actions, on the other hand, offer flexibility to tailor the agent’s capabilities to specific business needs. When creating a custom action, an Agentforce specialist can select a Reference Action Type, which can be Apex, Flow, Prompt Template, or API. Agent actions can be assigned to topics in Agent Builder. Instructions can be specified to define how a custom action should be used during conversations. Properly managing and assigning these actions ensures that agents deliver the most relevant and effective responses to user requests.

Managing Agentforce user security involves configuring permissions and access controls to ensure secure interaction with AI agents in Salesforce. Permissions can be assigned to both agents and users who require access to specific Agentforce functionalities. This is achieved through permission sets, which define access levels based on roles and responsibilities.
To create an Agentforce Employee agent, users must have the Manage AI Agents permission. For specialized agents, additional permission sets such as Manage Agentorce Sales Coach are required. An agent user record must be assigned to an appropriate permission set that contains the Einstein Agent license and the Einstein Agent User profile.
Security best practices for agents include following the principle of least privilege, granting access to the relevant objects, and assigning all the necessary permissions, such as the Run Flows permission that allows executing flows.

         Agentforce offers different types of agents for specific clouds and common use cases. Agentforce Sales Agents enhance sales productivity by assisting sales reps at different stages of the sales cycle. Agentforce SDR (Sales Development Representative) is a sales agent that focuses on the top of the sales funnel with the main goal of nurturing leads. Agentforce Sales Coach supports sales reps at the bottom of the sales funnel, offering personalized coaching, deal-specific feedback, insights, and interactive role-playing to help close deals. The Agentforce Employee Agent supports employees and includes various topics and actions that help sales reps perform tasks related to sales. The Agentforce Service Agent supports customers by processing incoming cases, autonomously resolving common inquiries, and escalating complex or sensitive support requests.

An Agentforce agent can be connected to multiple digital channels, such as Slack, Email, and Messaging. A connection can be set up between an agent and a digital channel by adding a connection in the Connections panel in Agentforce Builder. An agent surface acts as a bridge between an agent and a channel. It includes instructions and adaptive response formats.
An Agentforce Service Agent can be connected to a digital customer channel, such as a messaging channel exposed on an Experience Cloud site. By setting up Omni-Channel, a Messaging Channel, and Omni-Channel Flows, messaging requests can be routed efficiently to an agent. Omni-Channel Flows help manage inbound and outbound routing of conversations, while an Embedded Service Deployment allows an agent to engage with customers through an Experience Cloud site. Context variables can be used to personalize interactions, and progress indicators keep customers informed about agent activity. Before going live, testing in a test channel ensures proper message formatting.

This topic includes the following objectives:

Prompt Builder is a tool that allows users to create detailed prompt templates that instruct large language models (LLMs) to produce AI-generated responses. A prompt template consists of a prompt that utilizes various elements like ingredients, guidelines, and grounding.
Prompt Builder offers several types of reusable prompt templates, such as Sales Email, Field Generation, Record Summary, and Flex, that can be configured to meet specific business needs. The Prompt Template Workspace allows users to draft and test these prompt templates.
The Draft with Einstein button allows users to draft emails using a sales email prompt template. Users can click an icon next to a dynamic form field to use a field generation prompt template and populate the field with AI-generated text. In addition, invocable actions and flows provide seamless integration of prompt templates with workflows.

Salesforce users can be allowed to manage and execute prompt templates by assigning the appropriate permission sets. The Prompt Template User permission set can be assigned to users who need to access and run prompt templates. It provides visibility to the generative AI-enabled icon next to a dynamic form field that uses a field generation prompt template.
The Prompt Template Manager permission set can be assigned to users who need to create and manage prompt templates. The Einstein Sales Emails permission set can be assigned to those who need to draft emails using a Sales Email prompt template. Additionally, it is necessary to ensure that users can access all the relevant fields when drafting sales emails.

When creating prompt templates in Prompt Builder, understanding the key considerations and limitations for field generation and flex types ensures accuracy, consistency, and optimal grounding. Record Snapshots use the page layout of the current user for grounding. A Field Generation prompt template can only be added to a Lightning record page that is upgraded to Dynamic Forms. Some related lists, like Activities, User, and Organization, aren’t supported due to relationship and global variable limitations. Prompt templates must also respect numerical limits, such as a maximum of five related list merge fields. Flex templates introduce additional considerations, including integration with flows using the Add Prompt Instructions flow element, support for Apex and LWC through the Connect API. There are also considerations related to the handling of token limits and hallucination errors. Together, these considerations help ensure prompt templates return grounded, predictable, and contextually accurate responses.

When configuring prompt templates in Salesforce, selecting the appropriate grounding technique is essential to ensure accurate and relevant AI-generated content. Grounding techniques connect a prompt template to data sources such as Salesforce records, related lists, Apex classes, and external data. For instance, a record merge field links a template directly to an object field, while a related list merge field is used to ground a template using an object's related list.
A flow merge field can introduce dynamic logic by triggering a flow for a more complex use case. A record snapshot can be used to ground a prompt with data available on the user's page layout for an object. Prompt templates can also utilize Apex merge fields for programmatic use cases and DMO merge fields to use data stored in Data Cloud. Moreover, Retrieval Augmented Generation (RAG) can be used to ground templates with unstructured data, such as knowledge articles, emails, and chat transcripts.

Using Prompt Builder effectively requires understanding how to create, activate, and execute prompt templates. When creating a new prompt template, an AI specialist can specify its type, name, and description. Depending on the selected type, additional fields may be required. The Prompt Template Workspace provides a space to write, preview the resolution and response, and test a prompt. It also allows users to select the model configuration. An edited prompt template can be saved as a new version or new template, but it must be activated to make it available to users. The execution of a prompt templates depends on its type. A sales email prompt template can be run by clicking the Draft with Einstein button. A field generation prompt template can be run by clicking a generative AI-enabled field icon. A flex prompt template can be run using an invocable action, Connect REST API, or Connect in Apex. Consistently generating effective prompts is an iterative process, requiring best practices like testing multiple responses.

Writing effective prompts is essential to guiding the large language model (LLM) toward producing accurate, relevant, and consistent responses. Effective prompts are clear and concise, using plain natural language that avoids unnecessary jargon. Maintaining style and consistency ensures predictable outputs, while adding context helps the model understand the user’s intent, often through role-based instructions that define a character and goal. Iterative testing and feedback help refine prompts for improved results. To strengthen clarity, prompts should include direct instructions separated from context using triple quotes (“””). Structuring prompts into sections such as Role, Task, Context, and Constraints provides a framework for consistent and goal-oriented responses. Applying these best practices helps reduce hallucinations, ensures precision, and creates more reliable, business-ready prompt templates in Agentforce.

This topic includes the following objectives:

Agentforce Data Libraries enhance AI features by improving accuracy, adding personalization, and building trust in generative AI responses. A data library acts as a structured repository of knowledge that an Agentforce agent can use to provide precise and contextually relevant answers. Data libraries can be sourced from the Salesforce Knowledge base or uploaded files (such as text, HTML, and PDFs), ensuring agents have access to reliable information.
Key features include grounding AI responses with domain-specific knowledge, chunking data for efficient retrieval, and indexing for organized searches. Data libraries also support retrievers, which fetch relevant information dynamically. A data library can be configured to use an organization’s Knowledge base as its data source by selecting Identifying Fields and Content Fields. Specific files can also be uploaded and used as the source of a data library. A data library can be assigned to an agent in the Agent Builder.

Agentforce Data Libraries can store unstructured or semi-structured data and enhance AI features by improving accuracy, adding personalization, and building trust in generative AI responses. A data library acts as a structured repository of knowledge that an Agentforce agent can use to provide precise and contextually relevant answers. Data libraries can be sourced from the Salesforce Knowledge base or uploaded files (such as text, HTML, and PDFs), ensuring agents have access to reliable information.
Key features include grounding AI responses with domain-specific knowledge, chunking data for efficient retrieval, and indexing for organized searches. Data libraries also support retrievers, which fetch relevant information dynamically. A data library can be configured to use an organization’s Knowledge base as its data source by selecting Identifying Fields and Content Fields. Specific files can also be uploaded and used as the source of a data library. A data library can be assigned to an agent in the Agent Builder.

In Data Cloud, Retrieval Augmented Generation (RAG) can be utilized to ground large language model (LLM) prompts with accurate, current, and pertinent information. By retrieving structured and unstructured data from vector databases, retrievers improve the relevance and value of AI-generated responses. The retrieval process involves indexing data for efficient search, adding retrievers to prompt templates, and configuring retriever settings. Default retrievers are created automatically when search index configurations are created, while custom individual retrievers can be created and customized with filters in Einstein Studio. An ensemble retriever is a collection of individual retrievers. The configuration panel in Prompt Builder allows fine-tuning retriever settings.

In Data Cloud, search indexes can be created to ground Data Cloud search on unstructured and structured data for more accurate and relevant AI-generated content, deeper insights from analytics, and more efficient automation workflows. In Data Cloud, vector or hybrid search indexes can be created. Vector search helps understand semantic similarities and context between embeddings. Hybrid search combines vector search for semantic similarity with keyword search for lexical similarity. A search index configuration can be created to define a search index by navigating to the Search Index tab in the Data Cloud app. Retrieval Augmented Generation (RAG) is fundamental to Data Cloud search and can be utilized to ground large language model (LLM) prompts with accurate, current, and pertinent information stored in Data Cloud. 

This topic includes the following objectives:

The Agentforce Testing Center enables efficient testing of agents by allowing a large number of utterances to be evaluated in a single test. This reduces overall testing time and facilitates the quick activation of agents. Users can access the Testing Center in Salesforce Setup and leverage batch testing to execute multiple test cases simultaneously. The testing process involves creating a CSV file with test cases, running the test in a sandbox environment, and analyzing the test results for refinement. The test results display both successful and failed utterances, which helps fine-tune instructions, topics, and actions. Running tests consumes Einstein Requests and Data Cloud credits, and should be conducted in a sandbox to prevent unintended CRM data modifications. The Testing Center is available with the Einstein for Sales, Einstein for Service, or Einstein Platform add-ons and supports up to 10 tests in a 10-hour timeframe and 1,000 test cases per test.

Deploying an Agentforce agent from a sandbox to production environment ensures its availability for end-users. This process involves metadata deployment using Change Sets or Metadata API. All the relevant metadata components, such as GenAiPlanner, Einstein Bot, and Bot Version, must be included in the deployment. Additionally, service agents typically require the Embedded Messaging component for deployment to an Experience Cloud site. After deployment, the agent must be activated in the Agent Builder to make it available to users. Finally, stakeholders and end-users must be informed, with proper training and documentation provided.

To effectively manage and monitor the adoption of an agent, such as the Agentforce (Default) agent, it's important to track user interactions, feedback, and usage analysis. Agentforce Analytics provides dashboards and reports that offer insights into the usage, feedback, and adoption of agents, helping refine their effectiveness. The Utterance Analysis dashboard provides insights into how users engage with Agentforce (Default), their requests, and whether the agent is able to handle those requests. Event Logs in the Agent Builder allow an AI specialist to track detailed logs of user interactions to optimize user experiences and resolve common agent issues.

This topic includes the following objectives:

Model Context Protocol (MCP) is an open standard that defines how AI models connect with external tools, systems, and data. Originally developed by Anthropic, MCP enables Agentforce agents to perform real-world tasks by securely integrating with external applications through standardized interfaces known as MCP servers.
By using MCP, organizations can extend Agentforce functionality without complex custom integrations. Agents can retrieve live data, update records, trigger workflows, or access documents across connected systems, all while maintaining trust, governance, and security through the Einstein Trust Layer.
MCP transforms Agentforce into a connected ecosystem of intelligent, action-oriented agents capable of automating cross-system processes and delivering accurate, context-aware results.

The Agent-to-Agent Protocol (A2A) defines how AI agents within and across organizations communicate, coordinate, and collaborate securely. It provides a standardized and interoperable framework that allows diverse agents, such as sales, service, or partner agents, to exchange messages, delegate tasks, and share results regardless of platform or vendor.
Built on open standards and complementary to the Model Context Protocol (MCP), A2A focuses on horizontal communication between agents, while MCP connects agents to tools and data. Within Agentforce, Platform Events form the transport layer for this collaboration, enabling agents to publish, subscribe, and respond to contextual events in real time. Together, these capabilities foster scalable, event-driven collaboration, allowing agents to plan, orchestrate, and execute complex workflows securely under the Einstein Trust Layer.

The Agent API is a REST API interface that enables external systems, such as web apps, mobile apps, or other agents, to invoke Agentforce agents programmatically. It is best suited for use cases where interaction with agents needs to occur outside traditional Salesforce interfaces. With Agent API, organizations can extend Agentforce across websites, workflows, or custom platforms while maintaining consistent agent intelligence and context. It supports external UIs, headless automation, and agentic ecosystems where agents can collaborate through programmatic calls. Using Agent API involves setting up a connected app, generating an access token, and managing sessions that exchange messages with agents. The API also provides structured endpoints for session lifecycle, streaming responses, and context variables. Overall, it empowers businesses to embed, automate, and interconnect Agentforce agents across diverse digital environments.