Salesforce is expanding its Agentforce platform by rolling out seven preconfigured artificial intelligence agents designed to handle sales, customer service, commerce, employee support and back-office operations. Alongside these specialised releases, the enterprise software provider is introducing underlying technology that allows selected agents to pursue business goals across multiple days or weeks, preserve progress between separate interactions and coordinate activities with other autonomous agents.
Each agent is structured around specific commercial roles, arriving equipped with predefined skills, operational actions and data models that organisations can tailor to their existing business processes. Because these tools integrate directly into Customer 360, they draw seamlessly upon the wealth of enterprise data and established workflows already managed within Salesforce environments. This broader rollout reflects substantial operational experience, as the company has deployed thousands of customer agents over the past two years. Those implementations have generated 7 billion Agentic Work Units across both Agentforce and Slack to date, including 3.2 billion during the second quarter alone.
Seven agents target specific business tasks
The newly introduced tools target specific enterprise requirements across customer contact and internal operations. Casey resolves customer service enquiries across voice, SMS, WhatsApp and web chat, assisting with account management, returns, routine questions and escalation to human staff. Internal workplace support is handled by Paige, which processes human resources and IT requests, while Carter guides online shoppers through product comparisons, detailed enquiries and final purchases within a single chat window. On the operational side, Marshall manages back-office and supply chain workflows under predefined rules while maintaining an audit record of each action, and Piper engages prospective buyers across websites and inboxes to qualify inbound leads for business-to-business sales and marketing teams.
Customer experience across multiple channels is directed by Fin, which pairs the Operator customer operations agent with specialised Fin Apex models. While Fin and five companion agents have achieved general availability, the outbound sales agent Hunter remains in pilot. Scheduled for general availability in November 2026, Hunter is designed to research prospects, conduct outreach campaigns and collaborate with sales staff over weeks or months.
Organisations maintain full administrative control to adapt each agent to their specific environment, including the freedom to modify the system’s name. Salesforce states that all agents function strictly within the customer’s existing business rules, user permissions and established security controls. Underpinning this operational control is Agent Script, an open-source language developed for defining agent behaviour. The language allows companies to combine artificial intelligence reasoning with fixed, predictable rules that dictate precisely how agents evaluate information, make decisions and execute actions.
Salesforce cited several customer deployments to substantiate these operational claims. Early metrics indicate that Engine’s help agent fully resolves 50% of chat enquiries without human intervention, while Perk attributes 60% of its sales pipeline to Hunter, and Autism Queensland uses Paige to resolve 70% of administrative requests.
Enterprise adoption figures show similar momentum across commerce and technical infrastructure. Hibbett AI went live in six weeks and currently manages 90% of its core shopper journeys. Within business communications, Asana’s deployment of Piper increased website-agent conversation volume fourfold, with Salesforce noting that customers deploy Piper in 45 days on average. Across technical deployments, 79% of Anthropic conversations handled by Fin are resolved autonomously.
Agents can continue work across multiple sessions
Agentforce is also gaining a long-horizon runtime engineered specifically for business initiatives that develop over days or weeks. This architectural capability enables an agent to maintain measurable progress towards a broader objective across multiple interactions, continuously refining its plan as new instructions or information emerge.
Hunter represents the first agent built to utilise this persistent runtime. In an enterprise scenario outlined by Salesforce, a sales representative might task the agent with addressing at-risk deals prior to the close of a quarter. To achieve this, Hunter breaks that single objective down into discrete tasks while identifying which tools and records are required. Throughout the process, the agent complies with predefined rules that clarify when it can act independently and when direct seller approval is required.
Three distinct functions support this extended operational process. Memory preserves critical context and task progress across disparate sessions, while durable execution enables interrupted or unfinished plans to resume seamlessly as working circumstances evolve. Complementing these capabilities, dynamic steering actively recalibrates agent behaviour whenever human users provide new feedback and direction.
Salesforce plans to extend this runtime architecture to additional agents within its portfolio. In addition, the company intends to allow enterprise clients to build their own custom Agentforce agents capable of sustaining continuous work across multiple sessions.
Agentforce adds tools for teaching and coordinating agents
Further updates to the Agentforce ecosystem address the ways digital agents learn new tasks, collaborate across teams and undergo iterative refinement once deployed in production. A central addition is AI Skills in Agentforce Coworker, which allows an employee to teach Coworker how to complete a specific assignment once. That standardised workflow can then be reused across the broader workforce and executed through different software interfaces.
AI Skills is currently in pilot, with general availability scheduled for October 2026. Meanwhile, Multi-Agent Orchestration is generally available today, routing work between specialised agents whenever an enterprise task traverses different roles, systems or stages of a customer journey.
To help organisations construct and refine these autonomous systems, Salesforce developed Agent Optimizer. The tool enables teams to build agents, subagents and individual actions, while also offering capabilities to evaluate operational performance. Administrators can inspect and analyse historical records of agent sessions to identify specific areas for improvement ahead of the tool’s general availability in October 2026.
