Salesforce Agentforce now includes seven named, role-specific agents for service, sales, commerce, employee support, supply chain and customer experience. Announced on September 11, the portfolio combines several generally available products with a long-horizon sales agent that remains in pilot.
- Casey, Paige, Carter, Marshall, Piper and Fin are generally available, while Hunter targets November availability.
- Hunter is the first agent to use a runtime designed to pursue a goal across days or weeks.
- Buyers still need to separate Salesforce-reported customer outcomes from independently replicated performance.
Everyone else is reporting seven new agents; we are explaining the packaging decision behind them. Salesforce is trying to shorten the distance between buying an agent platform and deploying a working business process by supplying role-specific skills, actions and data models rather than another blank builder.
Salesforce Agentforce starts with defined jobs
The portfolio assigns a distinct job to each agent. Casey handles customer-service requests across voice, SMS, WhatsApp and web chat. Paige addresses employee IT and human-resources requests. Carter supports product discovery and in-chat checkout, while Marshall coordinates supply-chain and back-office processes with deterministic execution and an action record.
The remaining agents extend the portfolio into revenue workflows. Hunter researches accounts, creates outreach and supports a sales opportunity over time. Piper engages and qualifies inbound leads, and Fin resolves customer-experience workflows across channels. Customers can rename and tailor the agents, but the starting point is a preconfigured operating role rather than an empty prompt.
| Agent | Primary job | Status |
|---|---|---|
| Casey | Customer service | Generally available |
| Paige | IT and HR service | Generally available |
| Carter | Shopping and checkout | Generally available |
| Hunter | Outbound sales | Pilot; GA targeted for November 2026 |
| Marshall | Supply chain and back office | Generally available |
| Piper | Inbound pipeline generation | Generally available |
| Fin | Customer experience | Generally available |
Long-horizon work changes the control problem
Hunter introduces the most consequential technical change. Salesforce says its long-horizon runtime preserves memory across sessions, keeps a plan running through durable execution and adjusts behaviour through dynamic steering. The example in the announcement is a seller asking Hunter to rescue at-risk deals before quarter-end, a goal that can require changing information and repeated actions over several weeks.
That longer operating window increases the value of checkpoints. A buyer should identify which actions can run automatically, which require seller approval and what evidence is retained when the agent changes its plan. The question is not only whether Hunter can continue working, but whether an administrator can understand where it is, why it acted and how to stop or redirect it.
Orchestration and optimisation sit around the agents
Salesforce also made Multi-Agent Orchestration generally available. It routes work among specialised agents when a job crosses roles or systems. AI Skills for Agentforce Coworker are scheduled for October general availability and are intended to let employees teach a repeatable task. Agent Optimizer, also targeted for October, is designed to test agents and analyse session traces for improvements.
Futurum analyst Keith Kirkpatrick characterised preconfiguration as the central enterprise value proposition because it can compress deployment time. SiliconANGLE separately highlighted the same operational split: most of the new functions are available now, while Hunter and some supporting tools arrive later. Both sources treated Salesforce’s customer metrics as company-reported results rather than independent benchmarks.
The launch cites resolution or pipeline figures from Engine, Perk, Autism Queensland, Hibbett, Asana and Anthropic. Those examples show the use cases Salesforce wants buyers to examine, but they are not controlled comparisons. Procurement teams should request workload definitions, baseline labour, escalation rates, error categories and the period over which each result was measured.
What enterprise buyers should test first
A sensible pilot starts with one bounded job and a known source of truth. Teams can measure completion rate, human handoffs, permission failures, reversals and the cost per accepted outcome. They should also test how the agent behaves when customer data is incomplete, an approval arrives late or two systems disagree.
The packaging resembles a wider shift toward putting assistants inside established workflows. That is visible in Google’s Gemini Windows app and in Radisys’s telecom voice-AI tools. Salesforce’s bet is that a CRM already holding permissions, customer context and business rules gives prebuilt agents a faster path into production.
In plain terms: Salesforce Agentforce is becoming a catalogue of defined digital jobs backed by orchestration and monitoring tools. That can reduce setup work, but a named agent is still software that must be tested against a company’s data, controls and service standards before it is trusted with longer-running work.
Frequently asked questions
Which Salesforce Agentforce agents are available now?
Salesforce says Casey, Paige, Carter, Marshall, Piper and Fin are generally available. Hunter is in pilot with general availability targeted for November 2026.
What is the Agentforce long-horizon runtime?
It is a runtime designed to preserve context and continue a goal across days or weeks using memory, durable execution and dynamic steering. Hunter is the first announced agent using it.
Can companies customise the new agents?
Yes. Salesforce says customers can tailor the agents to their processes and rename them, while Agent Script combines AI reasoning with deterministic rules and permissions.
Primary source: Salesforce announcement.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
