Salesforce Adds AI Agents for Sales, Support and Internal Workflows

The release moves Salesforce’s agent strategy from a general work assistant toward specialized roles, with tools to divide complex work and constrain how agents carry out sensitive tasks.

By 3 min read
Salesforce Adds AI Agents for Sales, Support and Internal Workflows
Salesforce Adds AI Agents for Sales, Support and Internal Workflows

Listen to this story

The audio brief

About 1:34
0:001:34
Read transcript
Salesforce is expanding Agentforce from a general-purpose work assistant into a lineup of specialized AI agents—and giving them a way to work together on complex jobs. The new roles cover the full enterprise: Hunter handles business-to-business sales, from finding leads and drafting outreach to preparing proposals. Piper greets prospects on company websites, while Carter answers shoppers’ product questions and supports in-chat checkout. In service, Casey can process returns, and Fin handles broader support work. Paige answers employees’ human-resources and technical questions, while Marshall targets repetitive supply-chain tasks. The more important shift may be behind those individual agents. Agentforce Coworker’s new Multi-Agent Orchestration feature can divide a complicated request among specialized subagents. Salesforce says AI Skills will make that behavior easier to tailor, while Agent Script lets developers set explicit rules for sensitive tasks—boundaries intended to reduce made-up model responses. Hunter is also the first agent using Salesforce’s long-horizon runtime. The company says it can preserve plans and data across conversations, which matters when a sales cycle runs for weeks rather than minutes. Salesforce plans to extend that capability to more agents. Most of the release became generally available on September 11. Fin uses technology from Salesforce’s earlier three-point-six-billion-dollar acquisition, made in 2026, and some remaining features are due by year-end. The key constraint is whether Salesforce can broaden coordination and long-term memory without losing the task-specific controls its agents depend on.

Story brief

3 key points

Salesforce is turning Agentforce into a coordinated workforce rather than a collection of department-specific bots. Its new agents cover sales, web prospects, shopping, support, HR, IT and supply-chain tasks, while Multi-Agent Orchestration lets Agentforce Coworker delegate complex work to specialized subagents. Most capabilities became generally available September 11, but the long-horizon runtime—currently limited...

  1. 01

    Hunter is the first agent with Salesforce’s long-horizon runtime, preserving plans and data across conversations for sales cycles lasting weeks.

  2. 02

    Agentforce Coworker can split complex tasks across specialized subagents; AI Skills and Agent Script add customization and behavioral controls.

  3. 03

    Fin incorporates technology from Salesforce’s earlier $3.6 billion acquisition, linking the launch to the company’s support-agent strategy.

Salesforce has introduced a new set of AI agents for sales, customer support, internal help desks and supply-chain work, while upgrading Agentforce Coworker to coordinate specialized agents on more complex tasks. Most of the release became generally available September 11; the remaining features are scheduled for release by year’s end.

The launch is built around a simple division of labor: give different business teams agents designed for the work already in front of them. Hunter targets business-to-business sales work, helping people find leads, draft outreach, prepare for presentations and create proposals. Piper is intended to greet business prospects on a company website and route them to the relevant salesperson, while Carter answers shoppers’ product questions and includes an in-chat checkout widget.

  • Customer service: Casey can handle tasks including product-return requests, while Fin is designed for similar support work.
  • Employee support: Paige fields HR questions and technical-support requests from workers.
  • Operations: Marshall is aimed at repetitive supply-chain tasks.

The sales agent is also the first home for a capability Salesforce calls a long-horizon runtime. The company says it lets Hunter create longer-term work plans and reuse data across chat sessions, a design meant for sales processes that can stretch across weeks. Salesforce plans to bring that runtime to more of the newly introduced agents later.

That shared interface is Agentforce Coworker, Salesforce’s existing agent within several of its services. Its new Multi-Agent Orchestration feature can split complex work among specialized subagents. Salesforce also says Coworker can continuously optimize those subagents’ output quality, while a feature called AI Skills is intended to let workers more easily tailor how it performs tasks.

Customization is a central part of the release. Developers can use a Salesforce syntax called Agent Script to specify the way an agent must complete important tasks. Salesforce says those instructions can reduce the risk of hallucinations, or made-up model outputs, by placing more explicit boundaries around agent behavior.

Fin adds an acquisition link to the product lineup. Salesforce said the support agent is based on technology it obtained earlier in 2026 through a $3.6 billion acquisition. The company is positioning the new agents and Coworker update as parts of the same release rather than separate products aimed at isolated teams.

For customers, the immediate choice is not only which agent fits a department, but how much work should be handed from one agent to another. Salesforce has made most of the release available now, but the long-horizon runtime remains limited to Hunter and some features are still due before the end of 2026. The next move will show whether the company can extend its coordination and customization layer across the broader agent lineup without losing the task-specific controls it is emphasizing.

Sources

  1. siliconangle.comSalesforce introduces new AI agents to automate sales, support tasks - SiliconANGLE

Loading discussion...