Table of Contents
Salesforce Agentforce 360: Which of the 7 Named Agents to Deploy First
Six of the seven named Agentforce agents Salesforce shipped on September 11 are generally available today, and for most orgs the first one to turn on is either Casey or Piper. If customer-service tickets are your highest-volume, highest-cost repetitive work, Casey pays back first. If your bottleneck is inbound leads sitting unqualified in a web form, Piper does. The call hinges on where your expensive, repeatable volume actually lives, and on a pricing model that bills customer-facing agents at $2 per conversation. Hunter, the outbound sales agent drawing most of the headlines, is still in pilot until November, so it is not the answer to "what do we switch on this quarter." Below: each agent mapped to the function it targets, the conversation math at three volumes, the point where building outside Salesforce starts to win, and what Hunter's runtime changes for teams weighing build against buy.
The seven agents and what each does
Salesforce calls the September 11 release Agentforce 360, and the framing is deliberate. You are not buying a general assistant and reverse-engineering what it can do. You are buying a named agent with a job description, prebuilt skills, and a release date, per the launch post. Here is each one, with the constraint that matters before you sign.
Casey is the help agent. It resolves customer-service issues across voice, SMS, WhatsApp, and web chat, with prebuilt support for FAQs, returns, account management, and human escalation, GA now. Casey is customer-facing, so every resolved conversation draws against the $2 conversation rate or your Flex Credit balance. At a busy desk that meter runs fast, which is the whole reason the pricing section below exists.
Paige takes IT and HR requests across Slack, portals, and the internal tools employees already use, GA now. Paige is employee-facing, which means it can run under the $125-per-user unmetered add-on instead of per-conversation billing. For a high-traffic internal help desk that flips the economics entirely. Autism Queensland reported Paige resolving 70% of administrative requests, a figure from Salesforce's own launch customers, as summarized by Startup Fortune.
Carter is the shopper agent: discover and compare products, get questions answered, convert with in-chat checkout, GA now. It assumes a Salesforce commerce storefront. Hibbett said its Hibbett AI handles 90% of core shopper journeys after a six-week go-live, again a Salesforce launch figure collected by Startup Fortune. If you do not sell through Commerce Cloud, Carter has nothing to attach to.
Marshall runs back-office supply-chain work with deterministic execution and an audit record of every action, GA now. That pairing is the point. Ops teams cannot ship an agent that improvises on a purchase order, so Marshall trades the freewheeling LLM behavior for a traceable, replayable log your auditors can read.
Piper works websites and inboxes to engage, qualify, and convert inbound leads into pipeline, GA now. Asana reported four times the conversation volume with Piper and a 45-day average deploy, per Salesforce's launch data via Startup Fortune. The catch is simple: Piper only earns out if you have steady inbound volume. An outbound-led shop gets little from it until Hunter arrives.
Fin handles customer-experience workflows across channels, GA now. Fin came in through acquisition, and Anthropic says it resolves 79% of its support conversations on its own, reported by Startup Fortune. Watch the overlap with Casey. Both touch the service surface, so running the pair without a clean split of who owns which channel means paying two meters for one job.
Hunter is the outbound sales agent, and it is the exception on every axis. Pilot now, GA November 2026. It works a sales pipeline from research through outreach, collaborating with sellers over weeks and months rather than a single chat. Perk reported 60% of its pipeline built by Hunter in pilot, a Salesforce figure reported by Startup Fortune. Hunter is also the first agent to run on Salesforce's new long-horizon runtime, a runtime built to persist across sessions rather than reset each chat, and it gets its own section below. For a buyer deciding this quarter, the operative fact is short: Hunter is not GA, so if you need a sales agent in production before November, it is not on the table yet.
| Agent | Function | GA status | Best fit org type | Skip if... |
|---|---|---|---|---|
| Casey | Customer service (voice, SMS, WhatsApp, web) | GA now | High-volume support desks on Service Cloud | Ticket volume is low or highly bespoke |
| Paige | IT and HR service (Slack, portals) | GA now | Mid-to-large orgs with heavy internal ticketing | Small headcount or no Slack/ITSM footprint |
| Carter | Commerce, shopper (compare, in-chat checkout) | GA now | Retail and DTC on Commerce Cloud | You do not sell through a Salesforce storefront |
| Marshall | Supply chain (deterministic, audit trail) | GA now | Ops teams needing auditable back-office automation | Your supply chain runs outside Salesforce |
| Piper | Inbound pipeline (qualify, convert leads) | GA now | B2B teams with steady inbound web/inbox volume | Your pipeline is outbound-led |
| Fin | Customer experience across channels | GA now | CX teams wanting cross-channel resolution | Casey already covers your service surface |
| Hunter | Outbound sales (weeks-long pursuit) | Pilot; GA Nov 2026 | Sales orgs on Sales Cloud with multi-week cycles | You need a GA sales agent this quarter |
Pricing: what per-conversation billing looks like at volume
The consumption meter is the number that decides your first deployment, not the feature list. Customer-facing agents bill at $2 per conversation, or you buy Flex Credits at $500 per 100,000 and draw down per action. Run the conversation rate across three realistic desks:
| Monthly conversations | Conversations model at $2 each | Annualized |
|---|---|---|
| 1,000 | $2,000/mo | $24,000/yr |
| 5,000 | $10,000/mo | $120,000/yr |
| 20,000 | $40,000/mo | $480,000/yr |
Source for the rate: Salesforce Agentforce pricing. A mid-size support desk clearing 5,000 conversations a month is looking at $120,000 a year for Casey alone, before you add a second agent. That is the number your CFO will circle, and it is the one that makes the build-versus-buy question below a real question rather than a formality.
Now the flat-fee path. The $125-per-user-per-month add-on buys unmetered Agentforce usage for licensed employees, which is why it fits employee-facing agents like Paige and does not apply to customer-facing conversation volume. The crossover is arithmetic: $125 divided by $2 is 62.5 conversations. So a single employee-facing user who triggers more than roughly 63 agent conversations a month is cheaper on the unmetered flat fee than on metered conversations. Below that, metered wins. This is the one place the two SKUs are directly comparable, and it only holds for internal, employee-facing use.
Above these, Agentforce 1 Editions start at $550 per user per month and bundle 2.5 million Flex Credits per org per year, which is the shape most large Sales or Service Cloud shops will actually land on once usage stabilizes. Ignore the 7 billion Agentic Work Units Salesforce keeps citing while you model this. The AWU is an internal adoption metric, not a line item you pay, and diginomica notes it has not made its way into customer-facing pricing. Model the conversation rate and the flat fee. Those are the numbers on your invoice.
Buy prebuilt vs build custom on Agentforce
The build-versus-buy answer forks hard on one fact: are you already on Sales Cloud or Service Cloud?
If you are, prebuilt wins by default, and it is not close. The platform overhead, the data model, the identity, the CRM records the agent needs to read and write, is already sunk cost. Casey ships with the service data model wired in. Building an equivalent LLM-backed support agent from scratch means rebuilding that grounding, and a $120,000-a-year Casey bill at 5,000 conversations still undercuts the loaded cost of two engineers maintaining a custom harness plus its own model spend and on-call rotation. Deploy the prebuilt agent, spend your engineering budget on the skills that are specific to your business.
If you are not on Salesforce, the math inverts. Adopting Agentforce means adopting the platform underneath it, and that overhead swamps the per-conversation savings for a single use case. This is where teams reach for something lighter. A workflow-automation layer like n8n or Make lets you wire an agent loop over your existing systems without buying a CRM to host it, and you own the retry logic, the secrets, and the execution graph outright. The tradeoff is real: you inherit the orchestration problem Salesforce is selling you out of. We priced that decision in detail in our build vs buy guide for agent orchestration, and the short version holds here. Managed platforms are the right default while you are still learning what the agent should do; you own the harness once behavior is understood and you start hitting the walls the managed platform builds around you.
What breaks first, on the buy side: the moment you need behavior Salesforce's prebuilt skills do not cover, you are back to building on Agentforce anyway, now inside a more opinionated box. Scope the prebuilt agent to the 80% it handles out of the box, and budget separately for the custom skills you know you will need.
What Hunter's long-horizon runtime changes
Most enterprise agents today, including the current wave of Microsoft Copilot Studio agents and anything you build on n8n, run on a single-session or single-run model. The agent wakes up, does a task, and forgets. Hunter is the first agent on a runtime built to do the opposite, per Salesforce's launch post. Three capabilities sit underneath it. Memory preserves context and progress across sessions, so the work does not reset when a conversation ends. Durable execution keeps a plan running over days and weeks and lets the agent resume or course-correct as circumstances change. Dynamic steering adapts behavior to a specific user's feedback mid-flight.
Put together, that is a different unit of work. You hand Hunter an objective ("rescue my at-risk deals before quarter-end"), and it turns that into a plan, decides which tasks and tools the plan needs, and keeps working it as the buyer goes quiet and comes back, as Salesforce describes the flow. Startup Fortune's framing is the useful skeptic's read: months of persistent memory and durable execution are the claim, and if Hunter delivers on it the job-title branding is earned; if it does not, the branding looks thin fast.
For a platform team, this is the signal worth tracking regardless of whether you buy Hunter. Salesforce says more agents will move onto the long-horizon runtime over time, and that customers will eventually build their own long-horizon agents on it. The build-your-own-harness teams reading this are already solving the same problem with durable-execution engines. Hunter is the managed answer to the question you would otherwise answer with Temporal-style infrastructure. The one thing not to do is buy the September GA agents expecting Hunter's persistence. Casey, Paige, and the rest are single-job agents today. The weeks-long goal pursuit is Hunter's, and it is pilot-only until November.
What ships in October
Two platform capabilities land next month, per Salesforce. AI Skills in Agentforce Coworker lets employees teach an agent to complete a task once, then scale that know-how across the workforce; it is in pilot now with GA in October 2026. Agent Optimizer works across the agent lifecycle to build, refine, and test agents and analyze session traces for what to improve, GA October 2026. Multi-Agent Orchestration, which routes work across agents so they operate as one team when a job crosses roles or systems, is already GA now. If your first deployment is a single agent, none of this blocks you today; it matters once you run two or more.
What we would deploy first, by buyer type
For a small ops team already on Salesforce: start with one agent, not a portfolio. Pick Casey if support tickets are your volume, Paige if internal IT and HR requests are, and run the conversation math at your real ticket count before you commit. At low volume, metered pricing keeps the bill honest.
For a mid-market Sales Cloud customer: Piper now, Hunter later. Turn on Piper to qualify inbound while it is GA, and run the Hunter pilot in parallel so you are ready to promote it the day it reaches GA in November rather than starting the evaluation then. Hold off buying an Agentforce 1 Edition until your Piper conversation volume tells you where usage settles.
For an enterprise already running a custom agent harness: do not rip it out. Treat the September GA agents as the buy side of a hybrid, keep your durable-execution stack for the behavior Salesforce's prebuilt skills do not cover, and watch the long-horizon runtime as the benchmark your own harness now has to beat on memory and resumability. The real decision: does Salesforce's runtime close the gap fast enough that maintaining your own stops being worth the on-call?