Salesforce Inc. introduced a series of artificial intelligence agents designed to make sales and technical support teams more productive. Most of the new features are generally available today, and the rest will launch by the end of the year. The agents arrive alongside a new version of Agentforce Coworker, an automation tool built into several of the company's cloud services. For businesses that already run Salesforce, this expands the set of tasks that can be handed to software rather than to staff.
What Salesforce released and when
The lineup covers three areas. Three sales-focused agents debuted: Hunter helps business-to-business salespeople find leads and craft outreach emails, and it also automates preparation for presentations and the creation of proposals. Piper and Carter can be embedded in company websites: Piper greets business-to-business prospects and directs them to the most relevant salesperson, while Carter answers consumers' product questions for online retailers and includes an in-chat checkout widget. Customer support teams get two agents, Casey, which automates tasks such as processing product return requests, and Fin, built on technology Salesforce obtained earlier this year through a $3.6 billion acquisition. Internal operations are covered by Paige, which fields employees' HR questions and technical support requests, and Marshall, which automates repetitive tasks for supply chain teams.
Two mechanisms sit behind the agents. The first is a module called the long-horizon runtime, which lets an agent operate over large time scales, since some business-to-business deals take weeks to close. It enables the agent to develop long-term work plans and reuse data across chat sessions. On launch, the runtime is only available in Hunter, and Salesforce plans to integrate it into more of the agents that debuted today further down the road. The second is Agent Script, a programming syntax companies can use to customize the agents. According to Salesforce, it lets developers require that an agent carry out important tasks in a specific way, and such instructions reduce the risk of hallucinations.
The update to Agentforce Coworker adds three capabilities. Multi-Agent Orchestration enables the tool to split a complex task among multiple specialized subagents. Another feature enables it to continuously optimize those subagents' output quality. A capability called AI Skills makes it easier for workers to customize how Agentforce Coworker carries out tasks. The existing tool already lets workers interact with business data such as customer records, and it doubles as a way to manage agents created by colleagues. The new agents therefore arrive inside an environment where agents are already a managed part of the workflow, not a standalone experiment.
What this means for companies adopting AI agents
For a business, the practical change is the range of work that can be delegated. Sales outreach, proposal preparation, website qualification of prospects, product questions with checkout, return processing, HR requests and supply chain routines are all named use cases, and most of them are available now rather than in a pilot program. A small company without a sales operations team can put lead qualification and first-line support into agents and keep specialists for complex cases. A large organization faces a different question: how to distribute responsibility between agents built by different departments, which is what the management features in Agentforce Coworker address.
Several conditions matter before committing. The long-horizon runtime, the component that supports work spanning weeks, is limited to Hunter at launch, so long-running deal workflows in other agents are not covered yet. Fin depends on technology from a $3.6 billion acquisition completed earlier this year, which means its roadmap is tied to that team and product line. Agent Script is a programming syntax, so customization requires developer involvement rather than configuration by a business user. The claim that specific instructions reduce hallucinations is Salesforce's own, and it does not remove the need to verify outputs in regulated or customer-facing processes. Questions worth asking the vendor include which features are generally available today versus due by year's end, what data the agents reuse across sessions, and how subagent output is reviewed.
The marker to watch is the rollout schedule: Salesforce says the remaining features will launch by the end of the year, and that the long-horizon runtime will later be integrated into more of the agents that debuted today. If both happen on that timeline, the ability to run multi-week sales workflows will stop being a single-agent exception and become a standard part of the platform. For buyers, that is the point at which comparing vendors on agent orchestration and long-running task support becomes a routine procurement step rather than an early-adopter experiment.
