Salesforce has expanded its Agentforce platform with a portfolio of preconfigured artificial intelligence agents designed to handle tasks across sales, customer service, commerce, employee support, and back-office operations.
The company said the agents connect with Customer 360, allowing them to use customer information and business processes stored in Salesforce. Organizations can customize their names, skills, actions, data models, permissions, and operating rules.
The portfolio includes Casey, a customer-support agent that works across voice, SMS, WhatsApp, and web chat; Paige, an IT and human resources service agent; and Carter, a shopping assistant that supports product discovery, comparisons, and in-chat checkout.
Salesforce also introduced Marshall for supply-chain processes, Piper for qualifying inbound business-to-business leads, and Fin for customer-service workflows.
Hunter, an outbound sales agent, can conduct account research, prepare outreach, and work with sales personnel over extended periods. It is currently in pilot testing and is scheduled for general availability in November 2026. The other named agents are already generally available, according to Salesforce.
Hunter is the first product to use Agentforce’s new “long-horizon runtime,” which is intended to keep an AI agent working toward a goal over several days or weeks instead of limiting it to a single interaction.
The runtime incorporates memory to retain context between sessions, durable execution to keep plans running, and dynamic steering to adjust an agent’s behavior based on user instructions.
For example, a salesperson could direct Hunter to work on deals at risk of falling through before the end of a quarter. Salesforce said the agent could develop a plan, identify the required tasks and information, and determine when it can act independently or must obtain the salesperson’s approval.
The company is also adding AI Skills to Agentforce Coworker, allowing employees to demonstrate a task and reuse the resulting workflow. The feature is in pilot testing and is expected to become generally available in October 2026.
Another addition, Multi-Agent Orchestration, coordinates specialized agents when a workflow crosses different roles, systems, or stages. Salesforce said the capability is now generally available.
Agent Optimizer, meanwhile, is designed to help organizations build and refine agents, test their performance, and analyze session records. It is scheduled for general availability in October.
Salesforce said customers can use Agent Script, its open-source language for defining agent behavior, to combine AI-based reasoning with fixed business rules.
The company cited customer results from existing deployments, including a 50% autonomous resolution rate for Engine’s chat inquiries, 70% of administrative requests handled for Autism Queensland, and 79% of conversations resolved autonomously for Anthropic. These figures were supplied by Salesforce and were not independently verified.
Salesforce said Agentforce and Slack have processed 7 billion “Agentic Work Units,” including 3.2 billion during its second quarter.
