Salesforce’s latest Agentforce launch is unusually explicit about where enterprise AI is heading. The company is no longer just selling software that helps employees work faster. It is packaging AI around recognisable jobs and selling those agents as “job-ready”.
Casey handles customer support. Paige deals with internal IT and HR requests. Carter helps customers shop. Hunter works on outbound sales, while Marshall is designed for supply-chain and back-office processes. Salesforce’s Agentforce announcement describes them as preconfigured agents that come with the skills, actions and data models needed for particular kinds of work, before being adapted to a company’s own processes, permissions and systems.
The names are a bit cute, but the model behind them is more consequential. Salesforce is packaging software around work that would previously have been divided among employees, applications and conventional automation tools. The sales pitch is shifting from AI as something a worker uses to AI as something capable of taking responsibility for a meaningful chunk of the work itself.
That makes this a labour story as much as a software launch. If companies can buy agents already organised around customer service, sales, administration and other recognisable functions, the question is no longer only how much more productive employees become. It is also which parts of those jobs still need an employee at all.
From answering questions to doing the job
The distinction between an AI assistant and an AI agent has often been harder to spot in practice than technology companies would like. Plenty of supposed agents still behave like chatbots with access to a few more buttons, but Salesforce is trying to move further away from that model with systems designed to keep working after the initial instruction.
Hunter, its outbound sales agent, is the first of the new agents to use what the company calls a “long-horizon runtime”. Instead of completing a request and stopping, it can pursue an objective across days or weeks, remember what has already happened and adjust its plan when new information arrives.
A salesperson could, for example, ask Hunter to work on deals that are at risk before the end of a quarter. The agent can break the objective into tasks, decide which tools and information it needs, continue working as circumstances change and ask for human approval where its permissions require it. Hunter is currently in pilot, with general availability planned for November.
Salesforce is also adding orchestration that allows specialised agents to hand work between one another, while its Agentforce Coworker will let employees teach an agent how to perform a task and then make that skill available more broadly inside an organisation. An Agent Optimizer, expected to become generally available in October, is intended to analyse how agents are performing and help companies improve them. Together, those pieces start to resemble an operating model more than a software feature: an employee teaches the system a process once, and the organisation can reproduce that knowledge without requiring the same employee to perform the process every time.
There are still limits to how confidently any of this should be read. Some capabilities are in pilot or haven’t reached general availability, and Salesforce itself warns that customers shouldn’t base purchasing decisions on unreleased features. Its performance figures are also largely company and customer-reported rather than independent tests. Salesforce nevertheless says travel platform Engine’s help agent resolves half of its chat enquiries without human involvement, while sporting-goods retailer Hibbett says its AI handles 90% of its core shopper journeys.
Salesforce has already tried this on itself
Salesforce doesn’t have to speculate about what happens when agents begin doing work previously handled by employees because it has been running that experiment internally. The company deployed Agentforce on its own customer-support operation in early 2025. Within months, it said the system had handled 2.6 million customer conversations, resolving 63% of them while producing customer-satisfaction scores comparable with those achieved by human support staff. Salesforce later said hundreds of support employees had been moved into other parts of the business while it also reduced the need to backfill some roles. Salesforce has previously discussed how Agentforce is reshaping its own workforce.
That experience helps explain why Salesforce’s sales pitch has changed. Its earlier Agentforce messaging often concentrated on agents helping employees become more productive. The latest announcement describes software in the language of labour: agents are “job-ready”, come with the skills needed for the job and can be taught what an organisation’s best employees know.
It also fits the broader direction Salesforce has been taking with Agentforce. Reframed recently looked at how Claudeforce lets people work with Salesforce data and processes without necessarily using Salesforce’s traditional interface. The application itself is becoming less important as AI systems gain the ability to reach into company data and act on it directly. The new agents take that another step because, instead of asking which software an employee needs to do a job, an employer can increasingly ask how much of the job needs an employee at all.
South Africa has more at stake than most markets
That question lands differently in a country where the official unemployment rate reached 33.6% in the second quarter of 2026, with 8.5 million people unemployed, according to Statistics South Africa’s latest Quarterly Labour Force Survey. It becomes particularly relevant when looking at the kinds of jobs Agentforce is targeting because customer support, sales administration, internal service desks and back-office processing overlap with areas where South Africa has deliberately built employment capacity.
The country’s Global Business Services sector created 26,346 new jobs serving international markets in 2025, according to industry body BPESA. About 90% of those jobs went to young people, and the sector has set a target of creating hundreds of thousands more jobs by 2030. BPESA’s latest employment figures put some scale around what could change if international companies decide more of this work can be handled by software.
It would be too simple to conclude that every AI customer-service agent removes a South African customer-service job. Companies can use automation to increase capacity without cutting staff, handle demand that previously went unanswered or move employees into work requiring more judgement. Salesforce says it has done all of those things itself. The harder problem is what happens to the entry point into those professions when the simpler work begins disappearing first.
Many jobs develop expertise by giving people relatively straightforward work before handing them more complicated problems. A support employee learns the product by answering common queries before dealing with difficult customers. A salesperson learns how leads behave by researching prospects and following them up. An administrator becomes useful partly because years of ordinary processes teach them where the exceptions are. Those are precisely the tasks companies are becoming increasingly capable of handing to software.
Reframed has already seen a version of this emerging in the local employment market, where AI skills are spreading into jobs that wouldn’t traditionally have been considered technology roles. In cybersecurity, the same shift creates another problem: employers increasingly want people capable of supervising AI systems while the junior work that traditionally helped people acquire that experience is changing. The experience gap is becoming part of the AI skills gap. Salesforce’s latest products suggest that the same problem may spread well beyond highly technical occupations.
When the software arrives already trained for the role
There is a genuine business case behind what Salesforce is building. Companies don’t particularly want AI for the pleasure of owning AI. They want shorter waiting times, fewer repetitive processes, more sales opportunities followed up and lower operating costs. If agents can reliably provide those things, employers will use them. What is changing is the size of the piece of work that can be handed over.
Earlier workplace software automated steps. Generative AI started automating pieces of knowledge work. Salesforce now wants businesses to buy agents organised around recognisable jobs, capable of retaining context, pursuing objectives over time and cooperating with other agents. That doesn’t mean Casey replaces a customer-service department or Hunter replaces a sales team, but it does make the boundary between software and labour less obvious than it used to be.
For a South African economy desperate to create more routes into formal employment, productivity alone isn’t enough to settle whether that is progress. The gains will depend partly on what happens to the people whose first opportunity used to be doing the simpler work. Salesforce calling its software “job-ready” is unusually candid because employers can increasingly buy technology that arrives already arranged around a role. South Africa now has to work out how people become job-ready when some of the work that once made them so is being given to the software first.