The Names Are Not a Marketing Flourish
All seven carry human names, and each maps to a real job: Casey is a service rep, Paige is internal IT/HR support, Hunter is sales development, Piper is lead qualification. Customers can rename them to whatever they call those roles internally. This is a positioning statement rather than packaging. What it changes is the shape of the purchasing conversation — "should we buy a module" and "should we add a head" run through different budget pools, different approval chains and different comparison baselines. A feature module gets priced against comparable software; a "role" gets priced against labor cost. The thing for a buyer to watch is precisely the step that conversion skips: software is accepted against a specification, people are managed against performance, and software packaged as a role leaves you to define the acceptance criteria yourself.
The Genuinely New Piece Is the Long-Horizon Runtime
Technically the item worth attention is the runtime behind Hunter: goals that span weeks or months instead of a single session, which Salesforce says it will extend to other agents later. That also explains why Hunter alone is still in pilot — it is the only one that has to solve long-task state. This site covered OpenAI's Agents API public beta on September 11, and what that hosts is exactly session durability, context compaction and recovery from interruption. Both companies are working the same engineering problem and selling it to different buyers: OpenAI packages it as an API for developers, Salesforce packages it as a job title for business units. Whoever first makes "run one task for three weeks without drifting" genuinely solid owns the real barrier in this round. The rest of the release points the same way: Multi-Agent Orchestration is GA, Agent Optimizer and AI Skills in Coworker reach GA in October, and there is an open-source language called Agent Script for combining AI reasoning with fixed business rules — Marshall's emphasis on deterministic execution plus an audit record of every action is built for regulated processes.
No Prices at All, Which Is the Biggest Gap for Buyers
Seven GA agents and three platform capabilities, with pricing disclosed for none of them. For anyone evaluating this, that makes total-cost-of-ownership comparison impossible today. And for a product positioned as a role, the billing model matters more than the unit price: per seat, per resolution, or per the Agentic Work Units Salesforce already counts? The figures it published — 7 billion cumulative, 3.2 billion in Q2 — are most likely the precursor of the billing unit. If that is how it gets charged, your cost floats with how industrious the agent is rather than sitting fixed per head. That is the first question to settle in procurement. The customer-side results need reading by source too: Engine says its help agent fully resolves half of incoming chat inquiries, and Perk says Hunter builds 60% of its sales pipeline. Those are vendor-cited customer claims rather than independent measurement, across very different workloads. Judging it properly still means running a round against your own historical tickets and leads.