Salesforce Made Agents 'Job-Ready': Seven Named Agents and a Runtime That Pursues Goals for Weeks

Salesforce did something quietly significant on September 11, and a lot of the coverage buried it. It gave its AI agents first names and job titles: Casey does help, Paige does IT and HR, Hunter does outbound sales. Then it shipped the engineering bit that actually matters, a runtime that lets an agent pursue a single goal for days or weeks instead of finishing one conversation and forgetting everything. That second part is the story. Naming agents is packaging. Letting one run for a fortnight with memory and a plan is a different product altogether.

The seven agents, in one breath

Here is the roster, with the jobs in Salesforce's own framing. Casey is the help agent, working voice, SMS, WhatsApp and web chat on FAQs, returns and account management. Paige is the IT and HR service agent, living in Slack and employee portals. Carter is the shopper agent, handling product discovery, comparison and in-chat checkout. Hunter is the outbound sales agent, taking a pipeline from research to outreach over weeks. Marshall is the supply chain agent, running back-office processes with an audit record of every action. Piper is the inbound pipeline agent, working websites and inboxes to qualify and convert leads. Fin is the customer agent, on every channel, powered by Operator, a customer operations agent, and Fin Apex, a set of custom models trained for customer experience.

Six of the seven are generally available today. Hunter is the exception: it is in pilot now, with general availability set for November 2026. You can rename any of them, which matters more than it sounds. The names are a marketing layer over what is really a set of packaged, opinionated starting points for common jobs.

The piece that actually changes things

The names are the garnish. The main course is the long-horizon runtime, and Hunter is the first agent to run on it. Salesforce frames it as work that does not finish in one conversation, and it rests on three capabilities. Memory preserves context and progress across sessions, so the agent does not start from zero every time. Durable execution keeps a plan running for weeks without a human babysitting each step. Dynamic steering lets the person redirect the agent mid-flight.

Think about what that means for a sales agent. Today most AI assistants do one turn of work: draft an email, suggest a follow-up, summarise a call. Hunter is meant to hold a full outbound motion in its head, research accounts, send a sequence, read replies, decide who is warm, and keep going over weeks and months while a human steps in only to steer. That is the jump from assistant for a task to coworker with a mandate, and it is the phrase I will keep coming back to.

Why names are more strategic than they look

I was ready to roll my eyes at seven first names. Then I looked at what the names actually do. They package a job into something a buyer can recognise and a budget holder can sign off. Nobody writes a blank cheque for an autonomous agent platform. They write one for a support agent that answers FAQs and hands off the hard stuff to a human. Casey is that sentence compressed into a product.

This is also the no-code agent market maturing in real time. Two years ago, building an agent meant wiring together your own prompts, tools and guardrails from scratch. Now the default is a shelf of prebuilt agents you take off the rack and configure. The DIY era is not over, but the packaged era has arrived, and Salesforce just filled a whole shelf in one afternoon.

The other pieces that shipped alongside

It was not just seven names and a runtime. Three platform features came with dates, and they matter if you are planning around this.

• Multi-Agent Orchestration is now generally available, so several agents can be coordinated rather than run as islands.
• AI Skills for Agentforce Coworker is in pilot, with general availability in October.
• Agent Optimizer is also generally available in October, aimed at tuning agent behaviour over time.

None of these are the headline, but together they tell you the direction: from single agents doing single jobs, to a managed fleet of agents that a team can tune and coordinate.

The proof points are real, and they are theirs

Salesforce's release carries a page of customer results, and I want to be straight about where they come from. These are self-reported, not third-party audits. Engine says its help agent resolves half of chat enquiries. Perk credits Hunter with 60 percent of its sales pipeline. Asana says Piper drives four times the conversation volume, with customers deploying Piper in about 45 days on average. Hibbett says its shopper agent handles 90 percent of core shopper journeys after six weeks. And Salesforce cites Anthropic itself, the rival AI lab: Fin resolves 79 percent of Anthropic's conversations autonomously.

None of that is independently verified, and you should read it accordingly. But the pattern is worth noting: these are all cases where an agent was scoped to a specific job with a clear handoff boundary, not a general-purpose assistant dropped into a department and told to be helpful. The ones that work are narrow.

What this means for no-code builders

Here is the part I care about as someone watching the no-code space. Salesforce is the biggest CRM on the planet, and it has just told the market that agents should be bought as job-shaped products, not built as blank canvases. That pressure flows downhill. If enterprise buyers now expect a named agent for support, a named agent for pipeline and a named agent for the back office, then the no-code platforms selling agent builders have to answer the same question: can you ship me something job-ready, or do I still have to assemble it myself?

I think that is good for builders, mostly. The packaged agents set a floor for what good looks like, and the open platforms get to compete on how much further they let you go when the off-the-shelf agent stops fitting. The risk is the opposite: if the shelf is good enough, the long tail of custom builds shrinks, and a lot of DIY agent work turns into configuration instead of construction. That is a real fork for a lot of no-code agencies, and it is worth naming out loud.

The takeaway

Watch Hunter. Not because outbound sales is the most exciting job on the list, but because it is the first agent that runs on a runtime built for weeks of autonomous work, and its November GA is the real test of whether Salesforce can deliver a goal-pursuing coworker rather than a well-marketed chatbot. If it holds, the question every no-code and ops team should be asking itself gets sharper: which of my processes is a job I can hand to a named agent, and which still needs a human holding the plan? Answering that honestly is worth more than any feature list.