On September 11, Salesforce added seven purpose-built AI agents to Agentforce. Each one arrives with a job already defined — customer service, HR, shopping, supply chain — and companies can rename them to fit their own brand. Alongside the agents, Salesforce introduced a new runtime that lets an agent chase a single goal over days and weeks instead of a single conversation.

Seven agents, each shaped around a job

Until now, deploying an agent usually meant starting from a platform and pouring in your own processes and data. These seven move that work forward: each ships with the skills, actions, and data models its job requires. All of them connect to Customer 360, so they can work with the customer context and business processes a company already keeps in Salesforce.

Here is the lineup. Casey is the help agent, answering inquiries across voice, SMS, WhatsApp, and web chat, with prebuilt handling for FAQs, returns, and account management. Paige is the internal service agent, resolving IT and HR requests in Slack, portals, and the tools employees already use. Carter is the shopper agent, covering product discovery and comparison through in-chat checkout.

Hunter is the outbound sales agent, working a pipeline from research to outreach over weeks and months; it is in pilot now, with general availability set for November. Marshall handles supply chain and back-office processes, replacing manual steps with deterministic execution while keeping an audit record of every action. Piper is the inbound agent, engaging and qualifying leads across websites and inboxes, and Fin handles complex customer experience workflows, built on Operator, a customer operations agent, and Fin Apex, a set of custom models trained for customer experience. Six of the seven are generally available today; Hunter is the exception.

Customers can give each agent a name of their own so it reads as part of their brand and workforce. Behavior is governed by Agent Script, Salesforce's open-source language for agent behavior, which combines AI reasoning with deterministic rules to draw a precise line between what an agent may decide alone and what needs human approval.

Chasing work that outlasts a single conversation

The deeper change is the new long-horizon runtime. It targets the work that cannot be finished in one exchange — the weeks of effort behind closing a deal, for example.

Salesforce offers this example: a seller asks Hunter to rescue at-risk deals before the end of the quarter. Hunter turns that into a measurable goal, builds a plan, and starts working — deciding which tasks to complete, which tools and context it needs, and where the guardrails sit between acting autonomously and asking for seller approval. New information reshapes the plan as the work unfolds, but the seller stays in control.

Three mechanisms make that possible. Memory preserves context and progress across sessions. Durable execution keeps plans running over time and lets an agent resume or course-correct as circumstances change. Dynamic steering adapts behavior to an individual user's feedback. Hunter is the only agent running on this runtime today; Salesforce plans to extend it across the portfolio and eventually let customers build long-horizon agents of their own.

Teach them, combine them, fix them

Three platform capabilities landed with the agents. AI Skills in Agentforce Coworker let an employee demonstrate a task once and scale that know-how across the workforce; it is in pilot now, with general availability in October. Multi-Agent Orchestration routes work across agents so they operate as one coordinated team when a job crosses roles, systems, or stages — generally available now.

Agent Optimizer is the third. It works across the agent lifecycle, helping teams build and refine agents, subagents, and actions, test performance, and analyze session traces to find what needs improvement. An agent looking after agents, in other words, with general availability set for October.

The numbers already on the table

Salesforce published customer results alongside the launch. At Engine, a help agent named Eva fully resolves 50 percent of chat inquiries. At Perk, Hunter builds 60 percent of the sales pipeline. At Autism Queensland, Paige resolves 70 percent of administrative requests, and Hibbett AI went live in six weeks and now handles 90 percent of core shopper journeys.

At Asana, the website agent Piper has driven 4x the conversation volume, with an average deployment time of 45 days. At Anthropic, 79 percent of the conversations Fin sees are resolved autonomously. For scale, Salesforce says it has delivered 7 billion Agentic Work Units across Agentforce and Slack over the past two years, with 3.2 billion of those concentrated in the most recent quarter.

How to read the figures, and what remains open

Every number here is self-reported by customers and collected by Salesforce. A 50 percent full-resolution rate, for instance, says nothing about the other half — whether hard inquiries move to a human or simply fall through. When evaluating a deployment, the resolution rate is worth reading next to what happened to handling time for the cases that did reach a person.

The October and November availability dates deserve attention too. The announcement itself carries a note that it may reference services or features still in development and unreleased, and that purchase decisions should rest on what is currently available. The long-horizon runtime is a coherent idea, but whether a plan can be held steady across several weeks of real business will only be clear once customers start reporting from production.

Even so, the center of gravity is shifting — from building agents up from a platform, to receiving them with a job already outlined and tailoring them to your business. The lower that barrier gets, the more the real question becomes which part of your own operation you carve out and hand over.

Summary

On September 11, Salesforce added seven AI agents to Agentforce, covering everything from customer service to supply chain. Six are generally available now, and Hunter follows in November after a pilot. A long-horizon runtime that pursues goals across days and weeks, AI Skills for teaching agents new tasks, and Multi-Agent Orchestration for coordinating them arrived at the same time — an announcement that moves the emphasis from building agents to choosing and running them by job.

Note: the thumbnail image is AI-generated and illustrative.