Salesforce’s 7 Job-Ready Agentforce AI Agents: What’s New

Salesforce’s latest Agentforce release is built around a more ambitious promise for enterprise AI: agents should not simply answer a question or complete a single workflow step. They should be able to advance a defined business objective over days or weeks.

On September 11, 2026, Salesforce introduced seven named, “job-ready” agents for sales, service, commerce, employee support, supply chain, and customer experience. The company positioned them as prebuilt and customizable agents connected to its Customer 360 environment, rather than systems every customer must assemble from the ground up. 10

The major change: work that persists beyond a chat

The centerpiece is a new long-horizon runtime, first used by the outbound-sales agent Hunter. Salesforce says the runtime is designed for goals that outlast a single conversation or task, such as progressing an at-risk deal over an extended sales cycle. 47

Its three stated capabilities are:

• Memory: retaining relevant context and progress across sessions.
• Durable execution: allowing plans to continue, resume, or be adjusted as circumstances change.
• Dynamic steering: enabling a user to redirect an agent’s behavior with feedback and instructions while work is underway. 7

That distinction matters because a multi-step business process can change after the initial prompt: a prospect may respond, an account may become ineligible, or a seller may change the priority. Salesforce’s pitch is that the agent should retain the operating context and adapt rather than restart from scratch. Hunter is the first named agent on this runtime, with more portfolio and customer-built agents expected to follow. 7

The seven Agentforce agents

Salesforce assigned each agent a defined business role. Six were generally available at announcement; Hunter was in pilot.

• Casey is a customer-service agent for work such as answering questions, returns, account management, and escalation. It operates across voice, SMS, WhatsApp, and web chat. 45
• Paige is aimed at employee IT and HR service, supporting requests through Slack, portals, and existing employee tools. 5
• Carter is a commerce agent for product discovery, comparison, customer questions, and in-chat checkout. 5
• Marshall is intended for supply-chain and back-office processes, including process orchestration, deterministic automation, and auditability. 5
• Piper is an inbound pipeline-generation agent that engages, qualifies, and converts B2B prospects through websites and inboxes. 45
• Fin is a customer-experience agent for complex, multichannel customer workflows. 5
• Hunter is an outbound-sales agent that researches accounts, conducts outreach, and works alongside sellers on a pipeline over weeks or months. It entered pilot at launch, with general availability planned for November 2026. 45

The practical difference across this portfolio is not merely the agent names; it is the attempt to package agent behavior around a job function, its data, its systems of record, and the channels where work happens.

What Salesforce says the usage numbers show

Salesforce reported 7 billion cumulative Agentic Work Units (AWUs) across Agentforce and Slack, including 3.2 billion in fiscal Q2 2027. Salesforce defines an AWU as one discrete task accomplished by an AI agent. 1721

The company’s previous reported totals were more than 2.4 billion AWUs at the end of fiscal 2026 and 3.8 billion through fiscal Q1 2027, before reaching 7 billion in Q2. 192217

Those figures indicate volume of agent activity according to Salesforce’s own measurement framework. They should not be read as a direct, independently verified measure of productivity, revenue, or return on investment for every deployment.

Early customer results cited at launch

Salesforce highlighted several early deployment claims:

• At Anthropic, Fin reportedly autonomously resolves 79% of the conversations it handles.
• Engine said its help agent, Eva, fully resolves 50% of chat inquiries.
• Perk reported that Hunter builds 60% of its sales pipeline.
• Hibbett AI reportedly handles 90% of core shopper journeys and was deployed in six weeks. 512

These are vendor-cited customer results, so they are useful examples of the intended use cases rather than universal benchmarks. Results will depend on the process being automated, data quality, channel design, and the level of human oversight.

Why the timing mattered

Salesforce announced the portfolio four days before Dreamforce 2026, held September 15–17 in San Francisco. 1040 The timing put a concrete product portfolio behind the company’s broader “agentic enterprise” message: enterprise AI should be measured less by its ability to generate an answer and more by whether it can reliably move business work forward.

For buyers, the important question is therefore not whether an agent can produce a convincing response. It is whether it can maintain context, follow permissions and business rules, handle exceptions, and make sustained progress on a real business objective.