Salesforce shipped seven named AI agents inside Agentforce this week, each built to do one job rather than to demonstrate general AI capability: handling customer service, answering employee IT and HR questions, qualifying sales leads, running supply chain tasks, or closing online purchases. Six are generally available now. The seventh, an outbound sales agent called Hunter, arrives in November running on a new runtime built to pursue a goal across days or weeks instead of completing a task in a single session.
What Actually Happened
On September 11, Salesforce introduced Casey, Paige, Carter, Hunter, Marshall, Piper, and Fin as the next generation of Agentforce, according to reporting from CXM World and Hypertext. Casey manages customer service across voice, SMS, WhatsApp, and web chat. Paige handles employee IT and HR requests through Slack and internal portals. Carter helps online shoppers compare products and complete checkout inside a chat interface. Marshall runs back-office supply chain coordination. Piper qualifies inbound sales leads arriving through web forms and email. Fin manages customer experience workflows spanning multiple channels. Hunter, the outbound sales agent, is the first in the lineup built on a runtime designed for multi-day goal pursuit rather than single-session task completion, and becomes generally available in November.
Salesforce is citing scale as evidence of production maturity: coverage from Fiduciary Tech reports the company pointing to roughly 7 billion Agentforce work units processed to date. The framing matters — Salesforce is not presenting these as experimental copilots but as a deployed digital workforce already running at volume.
Automation Organized by Job, Not by Capability
Most agent platforms have sold capability first — retrieval, reasoning, tool-calling — and left the buyer to work out where to point it. This release inverts that. Each agent is defined by the job function it takes on, not the technique underneath it. That is a small naming decision with a real operational consequence: a role-specific agent can be assigned an owner, measured against the same metrics the human role was measured against (handle time, lead conversion, ticket resolution), and budgeted like a job rather than a software license.
The timing reinforces this is a broader shift, not a single vendor's marketing move. One day earlier, OpenAI opened its own Agents API into public beta, built specifically to handle orchestration, long-running sessions, and context management for agents that operate over extended tasks. Two of the industry's largest platforms shipped infrastructure in the same week for agents that run longer and operate with less human checkpoint. That convergence — not either announcement alone — is the actual signal.
It also raises the question the industry has been slower to answer. Gartner has projected that more than 40% of agentic AI projects will be canceled by the end of 2027, attributing the failures not to model capability but to unclear business value, poor governance, and escalating costs. Naming an agent after the job it replaces addresses the "unclear value" half of that equation — the business case is legible before deployment. It does nothing on its own to address governance, particularly once an agent like Hunter is designed to act autonomously across a multi-day window without a natural pause point.
The Enterprise Lens
If your business currently pays staff or an outsourced vendor to handle customer service tickets, employee HR and IT questions, follow-up on outbound sales leads, or supply chain order-checking, these agent categories map directly onto that existing cost, not onto a technology concept you have to translate. That is the practical difference from the last wave of AI tools: this is not "can we build something that helps with support," it is "here is an agent built for the support role specifically."
The useful question for a technology partner is not whether the agent can do the job — vendors will always say yes — but who currently owns that function today, and what the handoff looks like when the agent gets something wrong. A named, role-specific agent still needs a manager. Ask what escalation to a human looks like, how often it happens in comparable deployments, and who is accountable for the outcome when it does. That question matters more for a long-horizon agent like Hunter, operating over days without a checkpoint, than for a single-turn support agent like Casey.
What to Watch
- Whether Hunter's multi-day runtime ships on schedule in November, and what checkpoints or human handoff points Salesforce builds in — this is the leading edge of autonomous, long-horizon agents reaching general availability.
- Whether named, function-specific agents actually reduce cancellation rates against Gartner's forecast that over 40% of agentic AI projects will be scrapped by 2027 — renewal and expansion data from early customers will show this before any vendor announcement does.
- Whether other major platforms follow with agents organized around named job functions rather than general-purpose assistants, which would confirm this is a category-wide shift in how agentic AI is packaged and sold, not one vendor's positioning choice.
Sources