Salesforce: AI Agents as New Colleagues
The competition in the SaaS field for AI dominance is intensifying, with developments accelerating. Analysts refer to this as the third wave of AI. The first was predictive AI, followed by a breakthrough with generative AI, and now the phase of AI agents has begun. Unlike copilots, these are complex, individual AI applications that not only provide answers but also make autonomous decisions and perform tasks. According to a Gartner report, AI agents act autonomously based on knowledge, circumstances, and rules. By 2028, Gartner analysts predict that about one-third of all business applications will include AI agents.
“Digital Employees”
Salesforce CEO Marc Benioff talks about digital employees that will soon be used in all industries worldwide. This bold prediction was prompted by the announcement of Agentforce 2.0, which allows the creation of unlimited digital workers to complement the human workforce. Adam Evans, Executive Vice President and General Manager for AI at Salesforce, states that AI agents are needed to achieve the next level of productivity. Agentforce 2.0 can reduce the reliance on human employees for routine tasks while ensuring continuous efficiency and accuracy in various business areas.
Salesforce highlights its internal use at the helpdesk. Normally, around 32,000 conversations occur weekly with an 83% resolution rate. Before Agentforce, about 10,000 conversations were forwarded to human agents. With Agentforce 2.0, this number has halved, requiring only 5,000 conversations to involve humans, as AI agents can fully resolve the remaining inquiries.
There are many examples of AI agent applications. In the Sales Cloud, an AI agent helps actively follow up on leads, qualify sales potential, and generate targeted pitches. In the Marketing Cloud, AI agents create personalized campaigns based on previous customer interactions, allowing better segmentation and increased effectiveness. A recruitment agent scans resumes to select the most suitable candidates for a job description, informs HR for further selection, and then organizes interviews, preparing the interviewer with appropriate questions.
An example from the Canadian bank RBC involves analyzing how falling interest rates affect a customer’s portfolio. The response is multi-layered, considering different future scenarios, including optimal reallocation strategies.
To perform these tasks, the agent must have predefined skills. Agentforce 2.0 offers an extensive library that accelerates agent creation, including modules for Salesforce CRM, Slack, MuleSoft, and Tableau, as well as others developed by partners and available on AppExchange.
Modern Technology
Agentforce 2.0 is based on four central elements: the Data Cloud, the Analytics Tool Tableau, MuleSoft connectors, and the new semantic analysis engine Atlas-Reasoning. The Data Cloud allows access to all company data without the need for data migration. Tableau Topics and Actions provide data visualization and forecasts for better understanding of agent results. Tableau Semantics offers precise answers with comprehensive context. MuleSoft is an API-based integration platform with many predefined APIs, known as connectors. New APIs can be created in natural language. The key element is the Atlas-Reasoning engine, which uses knowledge graphs and recursive tree structures along with advanced indexing and multi-intent reasoning.
Questions are initially evaluated, and necessary data and metadata are retrieved from the Data Cloud. The response is verified in the Agentic Loop with various tools and sources to achieve a reliable and informed result. The Data Cloud enriches the data fragments retrieved via RAG with metadata from the Salesforce platform. This allows Agentforce 2.0 to better adapt to a company’s specific circumstances and provide precise and relevant answers, including detailed source references.
Slack: The Optimal Host Application for Agentforce 2.0
Agentforce can be accessed in any Slack conversation through the Agent Hub or @-mention and integrated directly into the workflow. Predefined Slack actions like “Create a Canvas” or “Inform the Channel” speed up the customization or creation of new AI agents. Slack CEO Denise Dresser states that Slack is the Salesforce-centered app opened first every morning, making it the natural place for Agentforce. Employees can collaborate with AI agents to access information across all Salesforce apps and external systems, making Slack a central hub for communication, not only between people but now also with AI agents.