On September 11, 2026, Salesforce introduced seven named, job-ready Agentforce agents, each built for one role: Casey for customer service, Paige for IT and HR requests, Carter for shopping and checkout, Hunter for outbound sales, Marshall for supply chain and back-office work, Piper for inbound pipeline and Fin for complex customer experience. Six were generally available at launch; Hunter was in pilot with general availability planned for November 2026. They arrive pre-built for their job, but they still run on your data, permissions and business rules, so readiness work decides whether they succeed.
What does each agent do?
| Agent | Job | Where it works | Status at launch |
|---|---|---|---|
| Casey | Help agent for customer service: FAQs, returns, account management and handoff to a person | Voice, SMS, WhatsApp, web chat | Generally available |
| Paige | IT and HR service agent for employee requests | Slack, portals and the tools employees already use | Generally available |
| Carter | Shopper agent: product discovery, comparison and in-chat checkout | Commerce storefronts, with in-chat checkout | Generally available |
| Hunter | Outbound sales agent that works a pipeline from research to outreach over weeks and months | Alongside sellers | Pilot; GA planned November 2026 |
| Marshall | Supply chain agent for end-to-end back-office processes, with deterministic execution and an audit record | Back-office systems | Generally available |
| Piper | Inbound pipeline agent that engages, qualifies and converts leads | Websites and inboxes | Generally available |
| Fin | Customer agent for complex customer experience workflows, powered by Operator and the Fin Apex models | Every customer channel | Generally available |
Salesforce presents these as purpose-built agents for specific jobs. You can still build custom agents; the named agents are a faster starting point for common jobs.
What new technology sits behind them?
- Long-horizon runtime: agents can pursue a goal over days or weeks, with memory across sessions, durable execution that recovers when a step fails, and steering from user feedback. Hunter, which works pipelines over weeks and months, is the clearest example.
- Multi-agent orchestration: a single front-door agent understands the request, breaks it into tasks and coordinates specialized agents. Salesforce says the architecture is A2A-compliant and works with orchestration on AWS, Azure and Google.
- AI Skills: repeatable work captured as reusable, governed instructions that any agent can run.
- Agent Optimizer (generally available October 2026): reads production sessions, finds recurring failure patterns and ranks them by impact, then helps refine agents.
- Agent Script: an open-source language that combines deterministic rules with AI reasoning, which matters for processes like Marshall's that must follow approved paths.
Which agent should you pilot first?
Pick the job where volume is high, the right answer is well documented and a mistake is easy to catch. For most companies that points to one of three:
- Casey or Fin, if service volume is high and your knowledge base is current. Start with the five or ten questions that make up most contacts.
- Paige, if IT and HR tickets pile up with repeat questions and your policies already live in one place.
- Piper, if inbound leads wait too long for a first response and your qualification rules are written down.
Hunter, Carter and Marshall usually come later. Hunter needs clean account and contact data and agreement on how agents and sellers share accounts. Carter needs commerce data and product content in good shape. Marshall needs the back-office process mapped and approved before an agent executes it.
What should you check before turning one on?
- Data: the records and knowledge the agent will use are current, deduplicated and owned.
- Permissions: the agent's access follows least privilege, and customer-facing agents can only reach what customers should see.
- Escalation: when the agent hands off to a person, which queue receives it and what context travels with it.
- Testing: real conversations, edge cases and attempts to push the agent off topic, before customers see it.
- Cost: how actions are metered in your contract and how many actions a typical conversation uses.
- Measurement: a baseline for the metric you expect to move, such as first-response time or resolution rate.
How do they relate to Agentforce Coworker?
Coworker is the employee-facing assistant in the Salesforce search bar. It can call on other agents during a conversation, and it supports teaching AI Skills. The job-ready agents do focused work on their own channels. In practice, employees ask Coworker, and Coworker or the orchestration layer brings in the specialist.
