Salesforce Agentforce’s 7 Named AI Agents: What Each One Does and What the Real Results Show
There is a meaningful difference between a company selling you an “AI agent platform” and a company handing you a named agent with a job title, a defined scope, and customer results you can read before you sign anything. On September 11, 2026 — the day before Dreamforce opened in San Francisco — Salesforce crossed that line. Seven agents. Seven business roles. Six available immediately.
This is not a generic AI assistant announcement. These are named, function-specific software workers built on Salesforce’s Agentforce infrastructure, each designed to reduce the time between “we bought an agent platform” and “something is actually running in production.” Whether they deliver on that promise is worth examining carefully — and the early customer data gives you a real starting point.
Why Salesforce Is Naming Its Agents
The naming decision is not cosmetic. When Salesforce calls an agent Casey instead of “Customer Service AI Module,” it is making a product argument: you are buying labor, not software configuration. A named agent implies a defined job, defined outputs, and accountability for results. It also makes adoption easier — enterprise teams talking about Casey replacing a ticket queue is a fundamentally different conversation than teams debating which “AI workflow module” to configure first.
The naming also signals something about how Salesforce expects the enterprise AI market to evolve. The company is betting that the next wave of enterprise AI adoption will not come from companies building custom agents — it will come from companies deploying prebuilt agents the way they currently deploy prebuilt SaaS software. The agent is the product, not the platform you use to build one.
The numbers behind the launch carry some weight. Salesforce reported 7 billion Agentic Work Units delivered across Agentforce and Slack to date, with 3.2 billion in Q2 alone. That is a platform with real production traffic, not a beta with limited exposure. The seven named agents are the next step built on top of that base.
The Seven Agents: What Each One Actually Does
Casey — Customer Service
Casey handles customer service across voice, SMS, WhatsApp, and web chat. Built-in capabilities include FAQ resolution, returns processing, account management, and escalation to human agents when needed. Casey is generally available now. Engine, a Salesforce customer, reports that its help agent Eva — built on Casey — fully resolves half of all chat inquiries without human intervention.
Paige — IT and HR Service
Paige resolves employee requests for IT support and HR information through Slack and internal portals. It connects to the tools employees already use rather than requiring them to navigate a separate system. Autism Queensland reports that Paige resolves 70% of administrative requests automatically — reducing burden on internal support teams without requiring a change to how employees currently ask for help.
Carter — Commerce and Shopping
Carter helps online shoppers compare products, answer detailed product questions, and complete purchases within a single chat window. It works across e-commerce environments handling order management and customer engagement. Hibbett, a US sporting goods retailer, deployed Carter and reports that its agent now covers 90% of core shopper journeys, with deployment taking six weeks from contract to go-live.
Hunter — Outbound Sales (Pilot)
Hunter is the most significant agent in the portfolio and the only one not yet generally available. It is in controlled pilot, with general availability planned for November 2026. Hunter is purpose-built for long-horizon outbound sales work — it researches accounts, creates outreach, and manages a sales opportunity over days and weeks rather than handling isolated tasks. Perk, a Salesforce customer, reports that 60% of its current sales pipeline is built by Hunter. That figure, if it holds at scale, represents a genuinely transformative change in how outbound sales functions operate.
Hunter runs on what Salesforce calls a “long-horizon runtime” — a memory and durable execution system that allows an agent to maintain context and pursue a goal across weeks of activity. This is architecturally different from agents that complete a task and reset. Hunter remembers where a deal stands, what outreach has gone out, and what the next action should be — without a human having to brief it each session.
Marshall — Supply Chain
Marshall orchestrates back-office supply chain processes with deterministic execution and maintains a full audit record of every action it takes. It is built for the class of supply chain workflows where compliance and traceability matter as much as efficiency. Generally available now.
Piper — Inbound Pipeline Generation
Piper works across websites and email inboxes to engage, qualify, and convert inbound leads into sales pipeline for B2B teams. It handles the first-touch and qualification work that typically falls between marketing automation and sales handoff — the gap where leads go cold. Asana’s deployment of Piper drives four times the conversation volume its previous approach did, with a typical Piper deployment taking 45 days from start to live.
Fin — Customer Experience (via Intercom)
Fin is the agent that came with the acquisition. Salesforce completed its purchase of Intercom — renamed Fin — on September 10, 2026, one day before the Agentforce launch. The acquisition was valued at approximately $3.6 billion based on filings from Hercules Capital, an Intercom lender. Fin resolves complex customer experience workflows across every channel. Anthropic itself uses Fin to handle customer support, and Anthropic reports that Fin resolves 79% of the support conversations it encounters without a human stepping in.
Each of the seven agents can be renamed by customers to match their own branding. Casey becomes whatever name fits the company’s customer-facing voice. Salesforce positioned this as an enterprise customization feature, but it also removes the risk of a company’s customers interacting with a named Salesforce product rather than what feels like that company’s own service.
Three Platform Additions That Matter More Than the Agents Themselves
The named agents are the headline. Three platform additions announced alongside them are, in some ways, the more consequential part of the launch for enterprise buyers.
The AI Control Plane is designed to register, govern, and monitor AI agents across an entire organization — including agents not built on Salesforce. This addresses a problem that enterprise AI adoption is already creating: companies deploying dozens of ungoverned agents across different tools and teams, with no central visibility into what those agents are doing, what permissions they hold, or how they are performing. The Control Plane treats agent governance as an organizational infrastructure problem rather than a per-product configuration task.
The Trusted Enterprise AI Harness is a six-pillar governance framework Salesforce is embedding across all Agentforce deployments. It covers data access controls, action permissions, audit trails, escalation rules, performance monitoring, and compliance documentation. This is what enterprise security and compliance teams will actually read before approving a production agent deployment.
Agent Optimizer, reaching general availability in October 2026, automatically tunes agent performance based on production outcomes. Rather than requiring human teams to manually review agent transcripts and adjust configurations, Agent Optimizer identifies patterns in what is working and what is not and proposes or applies adjustments.
For enterprise buyers evaluating the Agentforce launch, the Control Plane and Trusted Harness deserve more attention than the individual agent capabilities. The governance infrastructure is what makes it viable to run agents at scale across an organization — not just in one department’s pilot.
The Claudeforce Connection
The Agentforce launch does not sit in isolation. On August 26, 2026, Salesforce and Anthropic announced an expanded partnership that embedded Claude as the default reasoning model across Agentforce, Slack AI, and Slackbot. That partnership — called Claudeforce in coverage — means the agents launched on September 11 are, by default, reasoning with Claude’s underlying capabilities.
The Salesforce in Claude plugin, which entered open beta in September 2026, gives Claude access to 37 prebuilt sales skills and live CRM data without users leaving the Claude interface. The practical effect: a sales rep can ask Claude to summarize deal health, pull at-risk accounts, and draft outreach — all from within Claude, with data drawn live from Salesforce. No dashboard navigation required.
Salesforce reported that Claude internally drove 8.1 million hours of annualized productivity gains before the external launch. That figure comes from Salesforce’s own deployment of the tools it is now selling. It is a self-reported number and should be read as such — but the scale is consistent with the platform’s reported Agentic Work Unit volumes.
For broader context on what enterprise agentic AI adoption looked like in the weeks before this launch, the enterprise agentic AI August 2026 coverage covers the patterns that the Salesforce launch is now accelerating. The GPT-6 Astra review covers the competing enterprise AI model landscape.
How to Actually Evaluate This for Your Organization
The customer results Salesforce published are specific: Engine resolves 50% of inquiries, Perk builds 60% of pipeline, Autism Queensland resolves 70% of admin requests, Hibbett covers 90% of shopper journeys in six weeks. These numbers are striking, and they are also Salesforce-reported customer results — not independently verified benchmarks. The variance across different organizations, data environments, and use cases will be significant.
The questions worth asking before any Agentforce deployment are practical:
- What data does this agent need to access, and what are the permission controls on that access?
- What happens when the agent encounters a situation outside its defined scope — how does escalation work?
- What does the audit trail look like, and does it satisfy your compliance team’s requirements?
- How long did the reference customers actually spend in implementation, and what internal resources did it require?
- What is the pricing model — Salesforce did not disclose per-agent pricing in the launch announcement.
The Trusted Enterprise AI Harness exists precisely to answer the first three of those questions at a framework level. Whether it answers them satisfactorily for your specific industry and regulatory environment requires direct evaluation.
What This Launch Actually Means
Salesforce is redefining what enterprise software companies sell in the AI era. The traditional SaaS model sells tools and leaves configuration to the customer. The Agentforce model sells agents — named, role-defined, production-ready workers that a company is expected to deploy rather than build.
The commercial evidence that this model is working is real. Agentforce crossed $1.5 billion in ARR. The full Agentforce and Data 360 bundle reached $3.9 billion in ARR, up 210% year-over-year. These are not pilot metrics. They are production deployment numbers from enterprise customers who have evaluated the platform against real workflows and chosen to scale.
Whether the named agent model becomes the dominant enterprise AI deployment pattern — or whether enterprise teams ultimately prefer building their own custom agents on flexible infrastructure — is a question September 2026 opens rather than closes. What Dreamforce made clear is that Salesforce has built a commercially successful answer to the question of how you get an AI agent into production without starting from scratch. The seven agents are the most visible expression of that answer so far.
Track the latest Agentforce developments and enterprise agentic AI news at clawdbot2.in. For the safety dimensions of deploying agents at enterprise scale, our coverage of the agentic AI cybersecurity challenges covers the risks that governance frameworks like the Trusted Harness are designed to address.
FAQs: Salesforce Agentforce Seven Named AI Agents
When did Salesforce launch the seven named Agentforce agents?
Salesforce launched the seven named agents on September 11, 2026, the day before Dreamforce 2026 opened in San Francisco. Six agents — Casey, Paige, Carter, Marshall, Piper, and Fin — are generally available immediately. Hunter, the outbound sales agent, is in controlled pilot with general availability planned for November 2026.
What is Hunter and why is it only in pilot?
Hunter is an outbound sales agent built on a long-horizon runtime that allows it to pursue sales goals across days and weeks — researching accounts, creating outreach, and managing pipeline without resetting between sessions. It is in pilot because long-horizon, multi-week autonomous sales activity carries higher operational risk than the other agents’ use cases. Salesforce is validating production behavior before general release.
How much do the Agentforce named agents cost?
Salesforce did not disclose per-agent pricing in the September 11 launch announcement. Pricing details are available through Salesforce’s sales team. Enterprise buyers should request pricing alongside the Trusted Enterprise AI Harness documentation and the AI Control Plane governance specifications.
What is Fin and how is it different from the other six agents?
Fin is the customer experience agent acquired through Salesforce’s purchase of Intercom, completed on September 10, 2026, for approximately $3.6 billion. Fin runs on its own Operator and Fin Apex custom models trained specifically for customer experience workflows. Anthropic uses Fin for its own customer support and reports an autonomous resolution rate of 79%.
What is the AI Control Plane and why does it matter?
The AI Control Plane is a governance layer that registers, monitors, and manages AI agents across an organization — including agents built on platforms other than Salesforce. It addresses the emerging enterprise problem of ungoverned agent sprawl: dozens of agents running across different tools with no centralized visibility, permission auditing, or performance monitoring. For enterprise security and compliance teams, the Control Plane is often more important than the individual agent capabilities.