Salesforce now ships agents by job title, and for any team that has been waiting for the right moment to put one to work, the starting line has moved a long way closer. On 11 September 2026 the company announced a portfolio of seven prebuilt Agentforce agents, each scoped to a single business function and each given a name: Casey for customer service, Paige for IT and HR requests, Carter for shopping and in-chat checkout, Hunter for outbound sales, Marshall for supply chain and back-office work, Piper for inbound lead generation, and Fin for complex customer queries across every channel. Six of the seven are generally available today. Hunter is in pilot, with general availability planned for November.
The portfolio landed a day after Salesforce closed its acquisition of Fin, formerly Intercom, a roughly 3.6 billion dollar deal that brings in a customer agent platform used by more than 30,000 companies and a published average resolution rate of 76 per cent. Read together, the two announcements say something clear about where Salesforce believes the next wave of value sits. Not in the model, which it addressed separately with Koa, and not in the builder, which administrators have had for two years. It sits in the agent that arrives already scoped to a job, with skills, actions and a data model attached.
That is a genuinely useful shift, because it changes what a first agent project needs to be. Assembling an agent from scratch is a design exercise before it is a configuration exercise: somebody has to decide what the agent is for, which actions it may take, where it stops, and what a good outcome looks like. A prebuilt, job-scoped agent answers a fair portion of that in advance. What remains is the part only your own organisation can do, and it happens to be the part that keeps paying off whichever platform ends up running it.
What is Salesforce Agentforce, and what changed in September?
Agentforce is Salesforce’s platform for building and running AI agents that act on CRM data, taking real actions inside Sales Cloud, Service Cloud and connected systems rather than only drafting text for a person to send. Until September 2026 the usual way in was Agent Builder, where an administrator assembled an agent from topics, instructions and actions. The September announcement adds a second route: seven prebuilt agents, each scoped to a job, each shipping with its own skills, actions and data model. You configure and constrain them rather than assembling them from parts. Casey, Paige, Carter, Marshall, Piper and Fin are generally available now, and Hunter, the outbound sales agent, reaches general availability in November 2026.
The detail worth noting is how narrow each one is. Piper works inbound leads across websites and inboxes, engaging, qualifying and converting. Hunter works the other direction, running research through outreach on what Salesforce describes as a long-horizon runtime that lets it hold a goal across weeks rather than a single session. Marshall orchestrates back-office processes with deterministic execution and keeps an audit record of every action it takes. That narrowness is the feature. An agent with one job has a definable success measure, a clear owner and a short list of systems it needs to reach, which is exactly what makes a first deployment straightforward to scope and to approve.
How do you measure CRM ROI on an agent like this?
Measure a CRM agent the way you would measure a team member: the share of a defined workload it carries end to end, the time it takes to respond, and the quality of what it hands back to a person. Salesforce has published customer figures that give you a usable benchmark to plan against. Engine’s help agent resolves 50 per cent of chat inquiries. Perk built 60 per cent of its sales pipeline through Hunter. Autism Queensland resolves 70 per cent of administrative requests with Paige. Hibbett went live in six weeks and now handles 90 per cent of core shopper journeys. Asana’s Piper drove four times the conversation volume, on a 45-day average deployment. Anthropic resolves 79 per cent of the conversations Fin sees.
Those are vendor-supplied numbers on other companies’ processes, so treat them as a range rather than a forecast, and give yourself the cleanest possible comparison by capturing your own baseline first. Two weeks of data on one measure is usually enough: current containment rate on a support queue, current reply time on inbound forms, current conversion on a named segment. The teams who do that small piece of homework are the ones who can show a result in the first quarter, because they know what the before looked like. It also makes the second deployment an easy conversation, which is where the compounding starts.
The work moves from building to scoping
When the agent arrives prebuilt, the effort shifts to three decisions, and all three belong to the business rather than to a developer. First, the boundary: which queue, segment or product range does this agent own, and which cases are explicitly out of scope. Second, the handover: at what point does it stop and pass to a named person, and what does it pass along with the case. Third, the reach: which systems may it read from and write to, and which actions need a human approval before they commit. Written down properly, those three answers fit on a single page.
That page is worth more than it looks, because it survives the technology. It is the same document whether the agent is Casey on Service Cloud, a Copilot agent in Dynamics 365, or a custom agent inside HubSpot. Teams that have already defined their escalation rules and their approved action lists for one agent find the second one takes a fraction of the time, and the third becomes close to routine. This is the quiet reason prebuilt agents accelerate more than their setup time suggests: they force the scoping conversation early, when it is cheap, instead of halfway through a build.
Why is CRM data quality important for a prebuilt agent?
A prebuilt agent ships with its own skills and logic, but it reads your data, so your data sets the ceiling on what it can do well. Carter can only compare products that are described accurately in the catalogue. Piper can only qualify a lead against criteria that exist as fields rather than as habits in a sales manager’s head. Marshall can only reconcile a back-office process where the reference data agrees across systems. None of that requires a perfect CRM. It requires the specific slice the agent touches to be correct, complete and current, which is a far smaller and far more achievable piece of work than a full data cleanup.
This is the most encouraging part of the whole release for anyone who has struggled to get funding for data work. A job-scoped agent makes the case for you, because it makes the value of clean data visible and immediate rather than theoretical. Tidying one product catalogue improves quoting, reporting and search at the same time, so the effort is worth making even if the agent never goes live. Pick the narrowest agent whose data you already trust, ship that one first, and let the result fund the next round of tidying.
How do named agents improve CRM adoption rates?
CRM adoption improves when people can see what a tool does for them and know who to go to when it does not. Giving an agent a name and a single job helps on both counts. A rep who knows that Piper handles inbound enquiries and hands over a qualified conversation has a clear mental model of the division of labour, which is more than most AI features in CRM manage to provide. It also makes ownership legible inside the business: an agent with a name and a defined job can have a named human owner, a performance review and a backlog of improvements, in the same way any other part of the process does.
The practical move is to introduce the agent as a colleague with a remit rather than as a feature in a release note. Tell the team what it takes on, what it will never do, and how to send something back when the handover is wrong. Feedback loops of that kind turn early scepticism into useful signal, and they surface the missing field or ambiguous rule far faster than a report will. Teams that run their agents this way tend to find adoption looks less like a training problem and more like onboarding, which is a problem most sales and service organisations already know how to solve.
Why use an independent CRM partner when every vendor is shipping the same shape?
Because the preparation is portable and the platform choice should follow the process, not the roadmap. September made the convergence hard to miss. Salesforce released seven job-scoped agents on 11 September. Microsoft took its Dynamics 365 Sales Development agent to general availability on 30 September, quoting and closing low-complexity deals as its own licensed user. HubSpot opened agent building to its own customers over the summer. An independent partner can tell you which of those fits the work you actually do, and can say so without a licence target attached to the answer.
It also means the sequencing advice is the same wherever you land: define the job, clean the slice of data it depends on, write the escalation rule, then choose the agent. Buyers who approach it in that order tend to find the platform decision becomes easier rather than harder, because they know what they are asking the platform to do. The convergence is good news for them. Three vendors moving in one direction means the skills your team builds this quarter keep their value through the next contract cycle and the one after that.
The Sirocco perspective
We work across Salesforce, HubSpot and Microsoft Dynamics 365, and what we like about this release is how much of the hard thinking it brings forward. Our experience of agent projects is that the difficult part was never the build; it was getting a business to agree on where an agent’s job ends and a person’s begins. A prebuilt agent with a name and a remit puts that conversation first, on week one, when it costs a meeting rather than a rebuild. That is the right way round, and it is why we expect the teams starting now to be running two or three agents in production by spring rather than still scoping their first.
Our advice to clients this autumn is deliberately small. Pick the one job where you already trust your data, baseline a single measure for two weeks, and ship a scoped agent against it. The page you write while doing that, the boundary, the handover and the approved actions, is the asset. It travels with you to the next agent, the next platform and the next vendor announcement, and it turns each of those from a project into an increment. The organisations that build that muscle this year will be the ones choosing calmly from a much better set of options next year.
Get in Touch
If you are weighing up which of the seven agents to start with, or whether the equivalent in Dynamics 365 or HubSpot suits your process better, we are happy to talk it through. You can also schedule a consultation to scope a first deployment.
