Salesforce just shipped seven AI agents with job titles instead of prompts, and the sales-specific ones, Hunter and Piper, come with customer performance numbers attached. That is the real news inside this week’s Agentforce expansion: sales AI is moving away from the general-purpose copilot model and toward a catalog of pre-built roles a revenue team can hire into, the same way it hires a person.
From configurable assistant to job-ready hire
For the past two years, agentic AI in sales has mostly meant a flexible assistant that a RevOps team configures, prompts and iterates on. Salesforce’s September 11 announcement reframes that model. Hunter, described as an outbound sales agent, “works a sales pipeline from research to outreach, collaborating with sellers over weeks and months.” Piper, an inbound pipeline generation agent, works across websites and inboxes to qualify inbound leads into pipeline for B2B sales and marketing teams. Both ship as finished roles, not blank templates: skills, data models and actions already mapped to the job, tailored afterward rather than built from scratch.
Salesforce frames the shift as a response to scale. The company says it has logged 7 billion Agentic Work Units across Agentforce and Slack over the past two years, including 3.2 billion in the second quarter alone. That volume, Salesforce argues, is what exposed the limits of the assistant model: agents that can answer a question but cannot own a job. It also extends a pattern this publication has already tracked: a multi-vendor shift toward weeks-long agent autonomy paired with tighter human control, not less of it.
Advertisement
The numbers are the actual pitch
What separates this release from a typical feature announcement is that Salesforce attached customer performance data to specific agents rather than leaving the results to inference. Perk now has 60 percent of its sales pipeline built by Hunter. Asana’s inbound pipeline-gen agent, Piper, is driving four times the conversation volume the site saw before, with a deployment that averaged 45 days. Anthropic resolves 79 percent of the conversations Fin sees without human intervention. Those are not soft claims about potential; they are adoption figures from named accounts, disclosed alongside the launch.
That level of specificity matters because “AI agent” has become one of the most inflated phrases in enterprise sales technology this year. A vendor naming the exact share of pipeline a named customer credits to a named agent is a different claim than a roadmap slide. It also sets a new disclosure bar: competitors pitching agentic sales tools without comparable customer-attributed numbers, including the newer wave of AI SDR vendors now pricing by the lead instead of the seat, will increasingly look like they are selling concept, not deployment.
Built to run past a single conversation
The other structural change is a new “long-horizon runtime” underneath Hunter, meant to let an agent pursue a goal across days or weeks instead of resolving one interaction and stopping. Salesforce’s example: a seller asks Hunter to rescue at-risk deals before quarter end, and Hunter turns that into a plan, works the plan as new information arrives, and escalates for approval only where guardrails require it. Three capabilities make that possible: memory that persists context between sessions, durable execution that keeps a plan running and lets an agent resume after interruption, and dynamic steering that adjusts behavior based on a seller’s own feedback.
This is a meaningful technical claim, not just a marketing one. Most sales AI tools sold as “agents” today are closer to sophisticated single-turn automations: draft this email, score this lead, summarize this call. An agent that has to hold a goal open for weeks, incorporate new deal information, and know when to stop and ask a human is a genuinely harder engineering problem, and Salesforce is explicit that Hunter is the first agent on this runtime, with the rest of the portfolio migrating to it over time.
The same pattern is showing up down-market
The job-shaped, not prompt-shaped, packaging is not confined to enterprise CRM. Nimble, a small-business CRM vendor, launched AI Sequences on September 9, a feature that turns a single description of a desired outcome into a full multi-channel follow-up: the automated emails, the human touches, the timing and the exit conditions, written in the rep’s own voice. Jon Ferrara, Nimble’s founder and CEO, put the framing in almost identical terms to Salesforce’s own pitch: “Everyone tells small businesses to follow up. Almost nobody makes it easy. You describe what you want to happen, and Nimble writes the play that gets you there, in your voice, so it reads handwritten, not generated.” Where Salesforce is building named, persistent agents for enterprise pipelines, Nimble is compressing the same idea, output shaped like a completed job rather than a raw capability, into a single prompt for a one-to-five-person sales team. The shape of the shift is the same at both ends of the market.
What it means for the sales leader
For a RevOps or sales leader evaluating agentic tools over the next two quarters, the practical test is shifting from “can this AI do a task” to “can I hire this as a role, and what did it produce for someone like me.” That changes the diligence questions: ask a vendor for the same kind of customer-attributed adoption number Salesforce published for Hunter and Piper, not a capability demo. Ask whether the agent can hold a multi-week objective without losing context, or whether it resets with every interaction. And watch guardrail design closely: Hunter’s design explicitly separates the tasks it can complete autonomously from the ones that require seller approval, which is a governance decision every buyer will have to make explicit for their own pipeline, whichever vendor they choose.
The near-term risk is dilution: as “agent” packaging becomes the expected shape of a sales AI product, vendors without real long-horizon execution or without a customer willing to attach a number to their results will still borrow the language. The disclosure Salesforce set here, specific companies, specific percentages, specific deployment timelines, is the standard buyers should now hold every agentic pitch to, including Salesforce’s own next release.
Source: Salesforce
