Salesforce launches AI agents for business workflows

Salesforce has introduced a portfolio of job-ready AI agents for sales, service, commerce, employee support and back-office work. The tools are designed to handle defined business tasks using data and processes already held in Salesforce.

The launch includes agents named Casey for customer service, Paige for IT and HR support, Carter for shopping assistance, Hunter for outbound sales, Marshall for supply chain work and Piper for inbound lead generation. Salesforce also highlighted Fin, a customer service agent, as part of the wider group of tools now available through Agentforce.

The agents are connected to Salesforce's Customer 360 environment and can be adapted to fit customers' internal rules, permissions and security settings. Users can also rename them to match their own brands and workforce structures.

Salesforce is pitching the software as a faster alternative to building AI agents from scratch. Each comes with pre-set skills, actions and data models tied to a specific role, while still allowing customers to tailor the systems to their own processes.

Casey, Paige, Carter, Marshall and Piper are generally available now. Hunter is in pilot and due for general availability later this year.

The release follows work with customers across thousands of deployments. Salesforce said Agentforce and Slack together delivered 7 billion Agentic Work Units, including 3.2 billion in one quarter, as companies shifted from AI tools that answer questions to systems expected to complete operational tasks.

Longer tasks

A central part of the update is a new long-horizon runtime, which Salesforce said lets an agent pursue a goal over days or weeks rather than in a single interaction. The first agent using that runtime is Hunter, the outbound sales tool.

In Salesforce's example, a sales worker could ask Hunter to recover deals at risk before the end of a quarter. The software would then set a goal, build a plan, decide which tasks to complete, and determine when to act on its own and when to ask for approval.

The system relies on three features: memory to preserve context across sessions, durable execution to keep plans running over time, and dynamic steering to adapt behaviour based on user feedback. More agents in the portfolio will move to the same runtime over time, and customers will also be able to build their own long-horizon agents.

Workplace tools

Alongside the new agents, Salesforce outlined additional Agentforce functions for employees who use AI in day-to-day work. Agentforce Coworker is designed as an assistant that works across the surfaces where staff already operate, using business context held in Salesforce.

Salesforce also introduced AI Skills in Agentforce Coworker, which lets employees teach the assistant how to complete a task once and then reuse that process across teams and interfaces. Multi-Agent Orchestration is intended to route work between different agents when a task crosses business functions or systems.

Another addition, Agent Optimizer, is designed to help teams build, test and refine agents by analysing session traces and identifying areas for improvement. Multi-Agent Orchestration is generally available now, while AI Skills and Agent Optimizer are due later this year.

Customer examples

Salesforce attached a series of customer metrics to support the rollout. It said Engine's help agent resolves half of its chat enquiries without human intervention, while Perk's outbound sales agent is responsible for 60% of its sales pipeline.

Autism Queensland resolves 70% of administrative requests through its employee service agent, and Hibbett AI handles 90% of core shopper journeys after going live in six weeks. Salesforce also said Asana's website agent is driving four times the previous conversation volume, while 79% of the conversations seen by Anthropic's Fin deployment are resolved autonomously.

The launch reflects a broader shift in corporate AI spending toward systems tied to specific workflows rather than general chat interfaces. Businesses are increasingly focused on whether AI can complete repeatable work in sales, support, HR and operations, particularly where software can act within company rules and leave an audit trail.

Salesforce said Marshall, its supply chain agent, is aimed at that back-office need by automating manual processes with deterministic execution and recording each action. It also said Agent Script, its open-source language for agent behaviour, gives customers control over how reasoning systems combine with fixed rules when making decisions.

"Every company is racing to turn AI into real business results. But the hardest part is no longer getting AI to answer a question. It's getting AI to do the work that matters across every business function," Salesforce said.