Salesforce Agentforce's Named AI Workforce Arrives
Salesforce delivered a concrete answer to a question every enterprise AI team has been asking: instead of configuring a general-purpose agent from scratch, what if you could hire one that already knows the job? On September 11, 2026, one day before Dreamforce opened in San Francisco, Salesforce announced seven named, role-specific Agentforce agents and a new long-horizon runtime designed to let them work toward goals across days and weeks, not just complete a single interaction.
The announcement is backed by real deployment data. As of Q2 FY27, Agentforce has delivered 7 billion Agentic Work Units across Agentforce and Slack, with 3.2 billion in Q2 alone, growing 97% quarter-over-quarter. Agentforce ARR exceeded $1.5 billion, up 240% year-over-year. These are not pilot numbers. Salesforce’s Q2 earnings make the scale of production deployment visible.
Meet the Named Agents
Salesforce’s shift here is deliberate. Rather than selling a platform that enterprise teams configure from scratch, each agent arrives with the skills, data models, and guardrails the role requires. Companies can rename any agent to match their own branding, but the underlying job definition is already built.
| Agent | Role | Status |
|---|---|---|
| Hunter | Outbound sales: prospecting, outreach, pipeline, proposals | Pilot; GA November 2026 |
| Piper | Inbound pipeline gen: qualify and convert inbound leads | GA now |
| Casey | Customer service: voice, SMS, WhatsApp, web chat | GA now |
| Marshall | Supply chain and back-office operations | GA now |
| Fin | Complex CX workflows across all channels | GA now |
| Paige | Employee IT and HR requests | GA now |
| Carter | Commerce and shopping journeys | GA now |
Fin arrives with a significant installed base. Salesforce completed its acquisition of Intercom on September 10, 2026, the day before this announcement, bringing Intercom’s more than 30,000 enterprise customers into the Salesforce ecosystem. Fin Apex, a set of custom models trained specifically for customer experience, powers the agent.
The Long-Horizon Runtime: What Changes
Every earlier generation of AI agents was fundamentally conversational. A user started a session, the agent responded, and when the session ended the agent forgot everything. Even “agentic” platforms that could execute multi-step workflows were still transactional: one request, one result.
Salesforce’s long-horizon runtime breaks that pattern. Hunter is the first agent to run on it, and the design makes the ambition concrete.
A sales manager might ask Hunter to rescue at-risk deals before end of quarter. Rather than generating a report, Hunter turns that into a persistent goal. It builds a plan, identifies the deals, determines what context and tools each step needs, begins working them autonomously within the guardrails the team set, and surfaces the decisions that require human approval. Memory preserves progress across sessions so a seller who checks in two weeks later finds the work continuing, not reset.
Three components make this possible:
Memory. Context, preferences, and progress persist across sessions. Hunter does not re-read the CRM from scratch every time it acts; it carries state.
Durable execution. Plans continue running even after a conversation closes. If circumstances change, the agent can resume or course-correct without waiting for a user to re-engage.
Dynamic steering. The agent can update its approach when the user redirects it, without losing prior context.
Hunter is the first agent on this runtime. Salesforce’s roadmap extends it to more agents and to custom agents that customers build themselves with Agentforce.
Real Numbers from Named Deployments
Salesforce’s announcement included customer results that are worth examining carefully, because they illustrate the difference between generic deployment metrics and function-specific outcomes.
- Perk: 60% of the company’s sales pipeline is now built by Hunter.
- Asana: Piper drives 4x the conversation volume on Asana’s website. Average deployment time: 45 days.
- Engine: 50% of chat inquiries are fully resolved by its Casey-equivalent help agent.
- Autism Queensland: 70% of administrative requests are resolved by its Paige-equivalent employee service agent.
- Hibbett AI: 90% of core shopper journeys handled autonomously. Deployment time: six weeks.
- Anthropic: 79% of customer conversations handled by Fin are resolved autonomously.
The deployment times matter as much as the resolution rates. The traditional friction in enterprise AI deployment is not capability; it is configuration time, integration work, and organizational adoption. Salesforce is claiming 45 days for Piper because it arrives pre-configured for the role.
Multi-Agent Orchestration and Agent Optimizer
The portfolio announcement also introduced two platform-level capabilities that are less visible but strategically important for enterprise AI teams planning at scale.
Multi-Agent Orchestration routes work across the named agents automatically. When a job crosses multiple functions, such as a deal that requires both Hunter for outreach and Casey for customer support, orchestration keeps agents operating as one coordinated team without requiring manual hand-offs. This is generally available now.
Agent Optimizer is a meta-agent that works alongside teams across the agent lifecycle: helping build and refine agents, test performance, and analyze session traces to identify what to improve. It launches in pilot in October 2026 with general availability to follow.
Together, these capabilities represent Salesforce’s answer to a second-order problem in enterprise agent deployment. The first problem was getting one agent to work. The second problem, which every enterprise team running dozens of agents now faces, is managing the agent workforce itself.
What This Means for Enterprise GTM and Operations
The shift from generic AI assistants to named, role-specific agents has three practical implications for enterprise teams.
First, time-to-value collapses. A team deploying Piper does not start with a blank prompt and spend weeks writing agent instructions. They configure a system that already understands what inbound pipeline generation requires. The 45-day average deployment time is a structural advantage over build-from-scratch approaches.
Second, governance becomes tractable. The Trusted Enterprise AI Harness, announced alongside the named agents, is the governance layer that makes the agent portfolio governable at enterprise scale. Each named agent runs inside the permissions, data access rules, and audit controls a company already has configured in Salesforce. The agent workforce is visible in the AI Control Plane.
Third, the role model for AI deployment shifts. Instead of thinking about AI as a capability to acquire and configure, the named agent model treats AI deployment as workforce planning. Hunter has a job description. Piper has a ramp time. Fin has a specialization. The organizational vocabulary for managing an AI workforce is starting to look like the vocabulary for managing a human one.
For GTM and revenue operations teams specifically, Hunter’s design is worth studying. Salesforce built it as a sales harness: it learns the company’s GTM playbook, connects to deal data in CRM, carries context through the full revenue cycle from prospecting to proposals, and coordinates the downstream steps that sellers otherwise manage manually. For teams thinking about how to build autonomous GTM systems, Hunter is the first general-availability reference implementation at this scale.
The results from Perk and Asana suggest that enterprises that move fastest on deploying named agents are establishing leads that will compound. When 60% of a sales pipeline is built autonomously, the productivity gap between that team and a team still running manual SDR processes is not easily closed by incremental improvement.
The Integration with the Salesforce-Anthropic Partnership
It is worth noting that Anthropic appears in Salesforce’s own customer results list: 79% of Anthropic’s Fin-handled conversations resolve autonomously. Salesforce and Anthropic formalized a deeper partnership in August 2026 through ClaudeForce, embedding Claude as a model option across Agentforce and Slack. The Trusted Models layer in the Enterprise AI Harness supports model selection based on accuracy, cost, and task requirements, meaning Hunter and Piper can route to Claude, OpenAI, or other models depending on the work.
Dreamforce 2026 runs September 15-17 in San Francisco. Expect further detail on Hunter’s GA timeline, the Agent Optimizer rollout, and the long-horizon runtime roadmap in the sessions that follow.
Sources: Salesforce Agentforce job-ready agents announcement, Salesforce Q2 FY27 earnings press release, Forbes analysis by Tim Keary, Hunter product page