Salesforce spent September 11 doing something CRM vendors don't usually do: giving software workers first names. Casey, Paige, Carter, Hunter, Marshall, Piper, and Fin are the seven agents the company just folded into Agentforce, each built to own one job inside a business rather than field generic questions about it. Six of them are live today. The seventh, Hunter, runs on infrastructure Salesforce calls a long-horizon runtime, engineered to let an agent chase a goal for weeks instead of wrapping up in a single session.
- Seven job-specific agents now sit inside Agentforce: Casey (customer service), Paige (IT and HR), Carter (commerce), Hunter (outbound sales), Marshall (supply chain), Piper (inbound pipeline generation), and Fin (complex customer experience).
- Six of the seven are generally available now. Only Hunter remains in pilot.
- Hunter is the first agent running on Salesforce's new long-horizon runtime, built to work a sales pipeline across weeks or months instead of finishing in one chat exchange.
- All seven sit on Salesforce's existing Customer 360 data platform, inheriting a company's permissions and business rules rather than operating as separate, bolted-on bots.
What does each of the seven agents actually do?
Casey works customer service, and the detail Salesforce keeps emphasizing is that it doesn't just draft a reply, it carries a resolution workflow through to done. Paige lives on the employee side, closing out IT and HR requests wherever an employee happens to raise them, a Slack message, an internal portal, a ticket in whatever system already exists, instead of forcing people into a new app to get help. Carter handles commerce. Marshall handles supply chain. Piper's job is to take inbound interest, the kind that used to sit in a queue waiting for a human to triage it, and turn it into a qualified opportunity a sales rep can actually work. Fin picks up the customer experience cases that are too tangled for a simple decision tree.
Related700 AI Agents Hacked Hugging Face. Then They Tried to Erase It.
Then there's Hunter, and Hunter is different in kind, not just in function. It's an outbound sales agent, which sounds mundane next to the others until you notice what Salesforce built underneath it.
Why does Hunter need infrastructure the other six don't?
Outbound sales isn't a single-turn problem. A rep doesn't research a prospect, send one email, and call it finished. They research, they reach out, they wait, they follow up differently depending on what happened, they adjust over weeks as a deal either moves or stalls. Salesforce's bet is that an agent doing that job needs to behave the same way: pick up a goal, keep working it in the background, remember what it already tried, and check back in with a human when the situation actually calls for a decision. That's the job description Salesforce gave Hunter, and it's the reason Hunter shipped separately from the other six, as a pilot running on new plumbing rather than a straightforward extension of what Agentforce already had.
What does "long-horizon runtime" mean once you get past the marketing name?
A typical AI agent, even a capable one, is built around a session. You give it a task, it reasons, it acts, it returns an answer, and its working context is largely gone. That's fine for a support ticket. It falls apart for a sales cycle running eight weeks, because nothing persists between the Tuesday check and the Thursday follow-up unless something holds state across that gap. A long-horizon runtime is Salesforce's answer: infrastructure that keeps a goal alive, tracks what's been tried, and resurfaces the agent's own progress weeks later instead of starting cold each time. It also implies a different relationship with humans. A single-chat agent finishes or fails within minutes, so a human watches it live or reviews it after. An agent working a goal for two months needs planned check-ins built into the architecture, moments where it surfaces what it found and waits for a steer instead of guessing and plowing ahead.
| Long-horizon (Hunter) | Traditional single-session agent | |
|---|---|---|
| Working memory | Persists across days or weeks | Resets when the session ends |
| Unit of work | A goal ("land this account") | A task ("answer this ticket") |
| Human touchpoints | Scheduled check-ins as the goal evolves | One review at the start or end |
| Failure mode | Drifts off-goal if left unchecked too long | Times out or gives an incomplete answer |
Who actually has to deal with this?
Sales and revenue operations teams are the most direct audience, since Hunter targets outbound pipeline work that today runs through SDRs and sequencing tools. The framing matters more broadly for anyone in ops, support, or supply chain at a company already on Salesforce, though, since these aren't standalone chatbots bolted onto a website. They sit on Customer 360 and inherit whatever permissions already govern that data, so adopting one is less like installing new software and more like switching on a feature that already has access to everything the CRM knows. That puts competing agent vendors, and companies selling point solutions for support or sales sequencing, on notice. Salesforce is arguing those categories collapse into agents that already live inside the system of record.
What it means for the market
This lands just ahead of Dreamforce 2026, exactly when Salesforce wants analysts thinking about Agentforce rather than whether CRM growth has slowed. The competitive subtext is Microsoft, pushing Copilot deeper into its productivity stack, and ServiceNow, building its own agent story from the workflow side rather than the data side. Salesforce's pitch is that its edge isn't a smarter model, it's the data underneath: Customer 360 already holds the history, the case records, the pipeline, and stacking agents on that differs from a general-purpose assistant told everything from scratch. Whether that argument holds depends on how Hunter's pilot performs against real quota numbers, not a keynote demo, and Salesforce hasn't published that yet. None of this is a signal to trade the stock on. It's a read on where the company wants to plant its flag in the enterprise agent race, a race with more entrants than just these three.
RelatedClaude voice mode adds Opus, Sonnet, and app control
What happens next
The clearest milestone is Hunter's move from pilot to general availability, which will say a lot about whether the long-horizon runtime holds up outside a controlled rollout. Dreamforce 2026 is the likely venue, along with whatever follows these seven agents, since a named, job-specific agent is clearly a format Salesforce plans to reuse. Watch for concrete results from Hunter's pilot, and for rival vendors answering with their own persistent, goal-tracking agents.
Our take
The naming is a marketing choice, obviously, but it points at something real underneath. Giving an agent a name and a job title says it's meant to replace a role, not just answer a question, which is a bigger claim than most agent launches actually make. The long-horizon runtime is the part worth watching, because state persistence and goal tracking across weeks are harder engineering problems than another chatbot skin, and they fail in less visible ways too: an agent drifting off-goal for three weeks before anyone notices is a different risk than a chatbot giving a bad answer in real time. Salesforce keeping Hunter in pilot while shipping the other six as GA suggests it knows that, and that's a more honest rollout than pushing all seven out at once and hoping the hardest one works.
- OfficialSalesforce Newsroom: Agentforce job-ready AI agents the original announcement naming all seven agents and the long-horizon runtime
- AnalysisFuturum Group: Salesforce's job-ready agents target enterprise AI's biggest gap independent analysis of the launch and its enterprise implications
- Analysisppc.land: Salesforce agents gain a runtime that pursues goals over weeks, not chats detail on the long-horizon runtime mechanics
Original analysis by GenZTech Team. Sources linked above.