Salesforce puts its long-horizon runtime in one of seven new Agentforce agents

A sales operations lead who switches Hunter on next week is taking the long-horizon runtime on trust. Salesforce says the long-horizon runtime lets the agent develop long-term work plans and reuse data across chat sessions ^[5], and it points at business-to-business deals that take weeks to close ^[4]. I'd test that in one territory first, with a rep who will say out loud when the agent forgets what it decided on Tuesday.

The count is three sales agents, two for support, one for the internal help desk and one for supply chain: seven ^[16]. The runtime is in Hunter only at launch, with integration into the other agents promised later ^[6]. Six of the seven therefore arrive without the part Salesforce describes when it talks about work that spans weeks ^[17].

Task-splitting lives somewhere else entirely. Multi-Agent Orchestration is a feature of Agentforce Coworker, which breaks a complex task among specialized subagents and continuously optimizes their output quality ^[14]. Coworker is also where a worker pokes at customer records and manages agents a colleague built ^[15].

Agent Script is the syntax for pinning an agent to a specific procedure, and Salesforce says such instructions reduce the risk of hallucinations ^[13]. Someone has to write them, and that work lands on a developer before Casey handles a real product return request ^[9].

Support leads get a smaller and more awkward decision. Casey automates tasks such as processing returns ^[9], and Fin is designed for similar support use cases, built on technology Salesforce obtained this year in a $3.6 billion acquisition ^[10].

SiliconANGLE reports that most of the new features are generally available today and the rest launch by year's end, and it never lists which items fall into the year-end group ^[2][18]. Anyone planning a Monday rollout has to ask the account team which is which.

Two questions sort the seven for a given team. Does the work continue after the session ends, and is the output visible to someone outside the company? Carter answers a shopper's product questions and takes the payment through an in-chat checkout widget ^[8]. That work is external and it sits inside a single session, so the Agent Script constraints have to exist before it goes live. Piper greets a prospect and routes them to the most relevant salesperson ^[7]: external, one session, cheap to correct. Paige fields employee HR and IT questions ^[11] and Marshall takes repetitive supply chain work ^[12], both internal, both fine to run with a human reviewing the queue. Hunter sits in the awkward corner, multi-week and eventually customer-facing once it produces a proposal ^[3], and it is the only agent where Salesforce says the plan survives the session ^[6].