What can Agentforce do for your marketing in Marketing Cloud Next?

AI agents are now built into Salesforce’s newest marketing product. Here is what they actually do, where they help most, and what to sort out before you hand them a campaign.

Almost every marketing platform now says it has AI. Most of the time, that means a button that suggests a subject line. Marketing Cloud Next goes a good deal further. Its AI agents, built on Salesforce’s Agentforce platform, can turn a business goal into a campaign, write the content in several languages, decide which message each customer should receive and when, and answer customers who reply.

That is a real change in how a marketing team works, and it raises sensible questions. What do the agents actually do? Where do they save time, and where do they need watching? What does it take to get value from them, and what does it cost?

This article works through those questions in plain language: what we mean by AI and agents, why it matters that they live inside the Salesforce platform, the agents you can use today (including Campaign Agent, the Data 360 segmentation agent and Agentforce Coworker), the main advantages, the limits to plan for, and a sensible way to start.

First, what do we mean by AI and agents?

“AI” covers quite different things, and it helps to separate them, because they bring different benefits and different risks.

• Generative help. You ask, it writes. In Marketing Cloud Next that covers campaign briefs, first drafts of campaigns, and content such as subject lines, emails and SMS messages. You stay in charge of every step and keep what you like.
• Predictive AI. Scores and forecasts, such as how likely a customer is to buy or the best time to send. Marketing Cloud has offered this kind of AI for years.
• Agents. You give an agent a goal and some limits. It plans the steps, uses the platform’s tools to carry them out (building a segment, drafting content, setting up a journey), checks the results and adjusts. You approve the parts that matter.

Agentforce is Salesforce’s platform for building and running these agents across sales, service and marketing. In Marketing Cloud Next the agents are part of the product rather than an add-on you wire in. Salesforce’s help documentation shows how simple the starting point is: from the Marketing app you create a new campaign, choose to draft it with Agentforce and describe your objective. The agent then produces a draft brief and a campaign preview with a proposed flow and message content for channels such as email and SMS, which you review and refine. Behind the scenes, the agent’s actions run on Salesforce Flow.

Why it matters that the AI lives inside the platform

An AI agent is only as useful as what it knows and what it is allowed to do. This is where Marketing Cloud Next differs from adding a separate AI tool to your marketing stack.

• It works from your real customer data. Salesforce has built Data 360 into Marketing Cloud Next as its foundation. So the agent sees the same unified profiles your teams do: purchases, website activity, service cases, loyalty status and consent. It isn’t guessing from a spreadsheet someone exported last week.
• It follows the same rules as everyone else. Users, permissions, sharing and consent are managed once, on the Salesforce platform, and the agent works within them. Salesforce also routes generative AI requests through its Einstein Trust Layer, which is designed to mask sensitive data and prevent it being kept by the underlying model providers. Your data protection lead will still want to review the details, but the starting point is a governed one.
• It can act, not just suggest. Campaigns, segments, journeys and messages are all Salesforce records and flows. That means an agent can create them directly, and your Salesforce admins can see, check and extend what it built, using tools they already know.

Compare that with a stand-alone AI tool: data is copied out, results are pasted back in, and nobody is quite sure which version of the customer the AI was looking at. Keeping the AI where the data and the work already live removes most of that friction.

The agents you can put to work today

Salesforce has been adding agents to Marketing Cloud Next steadily since its launch in 2025, and the Winter ’27 release, published this month, brought the largest wave so far. Here are the ones most marketing teams will care about.

Two notes on that table. First, availability moves quickly, and what you get depends on your edition, so check the current position for your region and contract before planning around any of these. Second, the agents are designed to work together: the Data 360 agent helps define the audience, Campaign Agent builds and runs the campaign, Personalized Paths shape the journey, conversation agents pick up when a customer replies, and Agentforce Coworker gives your people one place to ask for any of it.

A closer look at Campaign Agent

Campaign Agent is the headline addition, and Salesforce’s own description splits its work into two phases.

Coworking. You give the agent a goal, for example to grow subscriptions. It turns that into a campaign tied to a brief: the goal and KPI target, dates, channels, budget, the audience with who is included and excluded, and a first set of content variants. It then produces the actual messages across email, SMS, RCS and MMS, personalised from your data and checked against your brand guidelines. Salesforce lists English, French and Spanish among the languages, which matters for any organisation working across Europe. Nothing goes live until your team has reviewed and approved it.

Decisioning. Once the campaign is live, the agent keeps working. For each customer it decides which campaign should actually reach them and holds back the ones that don’t fit. It picks the language, content variant, channel and send time. And it keeps adjusting based on results: moving budget, rebalancing channels, pausing an offer that isn’t working, shifting send times, or widening the audience to similar customers.

The part we find most interesting is the first decisioning step. Today, most journeys run in their own lane. The welcome journey, the seasonal promotion and the win-back campaign each follow their own rules, and none of them knows the others are about to message the same person this week. Campaign Agent is designed to arbitrate across all of them in the customer’s favour. For the customer, that means fewer, better-timed messages. For you, it means fewer unsubscribes caused by your own campaigns competing with each other.

The Data 360 segmentation agent: describe the audience, get the segment

Every campaign starts with an audience, and in many teams that is where the queue forms. A marketer knows who they want to reach, but turning that into a segment means knowing which data objects hold which attributes, and often waiting for a data specialist to build it.

The Data 360 segmentation agent, which Salesforce calls the Data 360 Marketing Agent, removes much of that wait. From the Agentforce panel, you describe the audience in plain language, for example “customers in France who bought in the last six months but haven’t opened an email since July”. The agent builds the segment from your description, or adds filters to a segment you already have open on the segment canvas, ready for you to check and activate. It handles the technical part, choosing the right data objects and attributes behind the scenes. Salesforce made it available from August 2026.

Further segment actions have been tested in pilot, aimed at the questions marketers usually take to a data specialist: how much two segments overlap, whether a particular customer is in a segment (and which rule left them out if not), why a segment failed to run, and when best to schedule it. Not all of these are generally available yet, so check what your org has before planning around them.

This matters for Marketing Cloud Next because Data 360 is the audience engine underneath every campaign. More self-service segmentation means marketers spend less time waiting and more time testing ideas. It doesn’t remove the need for someone who understands the data model to review important segments, especially in the early months.

Agentforce Coworker: one teammate in front of all your agents

As the number of agents grows, a new problem appears: knowing which one to use for what. Agentforce Coworker is Salesforce’s answer. It is an AI teammate that sits in the Salesforce search bar and also works where people already spend their day, including Slack, Microsoft Teams, Claude, ChatGPT and mobile. You ask in plain language, and Coworker finds the answer in your business data, triggers a flow, or hands the task to the specialised agent best placed to do it.

Coworker was announced in May 2026, became one of the three surfaces of Salesforce’s new AIforce interface layer at Dreamforce, and is generally available for Agentforce customers on Enterprise, Unlimited and Agentforce 1 editions. It is built on Data 360 and inherits your existing Salesforce set-up, so it follows the same sharing rules and permissions as the person using it.

For a marketing team, it changes how people reach the work. A campaign manager can ask which live campaigns are behind target this week. A sales rep can ask what marketing has sent to an account before a call, without opening the marketing app. And as more of your agents are connected, a marketer can ask for an audience or a campaign draft in the same place, with Coworker working out which agent should handle it. Check which of your agents Coworker can reach in your own org, but the direction is clear: Coworker becomes the front door, and your people no longer need to know which agent sits behind it.

The main advantages

Setting the launch language aside, these are the benefits that tend to matter most once the agents are in use.

• From brief to launch much faster. When Marketing Cloud Next was announced, Salesforce pitched campaign creation in hours rather than weeks. You’ll still need time to review and approve, but a complete first draft (brief, audience, journey and content) arrives almost at once, and your team starts from something to react to rather than a blank page.
• Personalisation without hand-building every path. Pre-built journeys can only allow for so many branches. Agents and Personalized Paths choose the route, content variant, channel and timing for each customer, so you get closer to one-to-one relevance without one-to-one effort.
• Audiences without the queue. Describing an audience in plain language and getting a segment back puts more of the targeting work in marketers’ hands, with data specialists reviewing rather than building every request.
• Fewer competing messages. Because the agent decides across campaigns, a customer gets the one message that fits rather than three that clash. That protects engagement, deliverability and your brand.
• Conversations instead of dead ends. Customers can reply on channels like SMS and WhatsApp and get a useful answer, instead of a message bouncing off a “do-not-reply” address. Marketing becomes a starting point for a conversation, not a broadcast.
• Content in the languages your customers speak. For international organisations, producing the same campaign in several languages usually means several briefs, translators and review rounds. Generating on-brand variants in parallel, with brand checks built in, removes much of that delay.
• Smaller teams, bigger programmes. The agents take on the repetitive mechanics: building segments, assembling variants, adjusting send times. Your marketers spend more of their time on strategy, creative direction and judgement.
• Marketing, sales and service on the same page. The agents work from the same customer record as your other teams. A customer with an open service case can be left out of a promotion automatically, and the Account Discovery Agent gives sales and marketing a shared view of which buying groups are active.
• Optimisation that never stops. Instead of learning what worked from a report after the campaign ends, the agent adjusts while it is running and suggests changes for you to approve.
• One place to ask. Agentforce Coworker gives marketers, sales reps and managers a single place to ask questions and get work done, without needing to know which agent does what.
• Work in the tools you prefer. With Headless Marketing, some teams will be able to brief, check and adjust campaigns from Slack or their usual AI assistant, without opening the marketing app for every task.

What it looks like in practice

Take a fictional subscription business selling in France, Spain and the UK, whose marketing team wants to grow paid subscriptions this quarter.

The usual way. The team writes three briefs, one per market, and sends copy out for translation. Someone builds separate journeys for trial users, lapsed users and newsletter subscribers. A seasonal promotion and the welcome journey both target new trial users in the same week, so some people get both. A month later, a report shows which variant did best, by which time the campaign is over.

With the agents. The marketing lead gives Campaign Agent the goal, the budget and the dates. The agent proposes a brief and an audience, for example active trial users who have used the product several times, excluding anyone with an open service case. When the team wants an extra audience for lapsed trial users, a marketer describes it to the Data 360 agent and a data specialist checks the segment before it is used. Campaign Agent drafts email and SMS variants in English, French and Spanish, checked against the brand guidelines. The brand lead and a native speaker in each market review and approve.

Once live, the agent decides for each person. A customer who complained to service yesterday gets nothing promotional this week. A new trial user part-way through the welcome journey receives the subscription offer instead of the seasonal one, because it’s more relevant to them. Send times shift toward when each person tends to engage. When one offer underperforms, the agent proposes pausing it and moving budget to the stronger one, and the team approves the change. On Monday mornings, the marketing lead asks Coworker in Slack for a summary of how the campaign is doing and which changes are waiting for approval.

The marketing team’s job hasn’t disappeared. It has moved: setting the goal, defining the guardrails, approving the work, and reading what the agent is learning about their customers.

Limits and trade-offs to plan for

The agents can save real time and improve results, but they are not a switch you flip. These are the points we’d want any client to understand before going further.

Your data decides your results. An agent working from incomplete, duplicated or out-of-date profiles will make confident decisions on the wrong information, at scale. The Data 360 groundwork (connecting sources, matching identities, recording consent properly) comes first. If that sounds like a project in itself, it often is, and our Data 360 article explains what it involves.

The licence is not the whole cost. Agent features vary by edition. Salesforce’s pricing pages show Campaign Agent capabilities tied to the Growth and Advanced editions, and its Campaign Agent FAQ notes that the Advanced edition includes a limited amount, with additional agentic campaigns and content generations at extra cost. Agent activity also draws on Flex Credits, Salesforce’s universal credits for Agentforce, and building and publishing segments in Data 360 consumes Data 360 credits however the segment was created. If you already use Marketing Cloud Engagement, Salesforce’s route to these features is renewing into a “+” edition. Model your expected usage before you scale, not after the first invoice.

Guardrails are a job, not a setting. Someone has to write the brand guidelines the agent follows, decide which choices it can make on its own (send times, probably; budget moves, perhaps not), and approve its output, in every language you use. Early on, reviewing agent work takes real time. That time falls as trust builds, but it should be planned.

Much of this is new. Several of the capabilities above only reached general availability in October 2026. As with any new release, base decisions on what is available in your org, region and edition today, not on what has been announced.

Europe adds its own questions. When an agent decides who receives which offer based on their profile, that is profiling under the GDPR, and conversational agents raise transparency questions too: under the EU AI Act, people should generally know when they are talking to an AI. None of this rules agents out, but your data protection lead should be involved early, not asked to sign off at the end.

Measuring gets harder when the campaign changes itself. If the agent keeps adjusting, it can be difficult to say what made the difference. Agree your KPIs in advance and keep a hold-out group that doesn’t receive the agent-run campaign, so you can compare like with like.

Skills shift. Marketers become editors and supervisors of agent work, which is a different skill from building journeys by hand. Your Salesforce admins, who know Flow, become more important to marketing than before. Plan some training in both directions.

Where to start

You don’t need a large programme to find out whether agents work for you. A focused pilot on one goal will tell you most of what you need to know. We approach it in three steps.

Good first candidates are a campaign that should pause for customers with an open service case, a campaign you already run in several languages, two-way SMS or WhatsApp reminders for renewals or appointments, or, for B2B teams, using the Account Discovery Agent to focus outreach on the buying groups that are actually active. Two lighter starting points are letting a small group of marketers build segments with the Data 360 agent while a data specialist reviews them, and switching on Agentforce Coworker for a pilot team so people get used to asking for work in plain language.

Is it for you? Questions to ask before you start

Agents tend to pay off when your customer data is already in good shape in Data 360, you run enough campaigns across enough segments, channels or languages that the mechanics take real effort, and you can say clearly what a better result looks like. If your data needs work first, or you run a handful of simple campaigns, it is usually wiser to fix the foundations or start smaller.

To work out where you stand, answer these honestly.

1. Which one or two goals would we hand to an agent first, and how would we know it worked?
2. Is our customer data in Data 360 complete, consented and reliable enough for an agent to act on?
3. Which decisions can the agent take on its own, and which need a person to sign off?
4. Who owns brand guidelines, tone and approvals for AI-generated content, in each language we use?
5. Which edition are we on, or buying, which agent features does it include, and what will credit usage cost at our volumes?
6. Has our data protection lead reviewed the profiling and AI transparency questions?
7. How will we measure the agent’s impact: which KPIs, and which hold-out or control group?
8. Who will supervise the agents day to day, and what training do they need?

AI in Marketing Cloud Next isn’t a feature you switch on and forget. It changes who does what: the agents take on the mechanics of building, personalising and tuning campaigns, while your team owns the goals, the brand and the judgement calls. For organisations with good data and a clear idea of what they want, that is a genuine step forward. Start with one goal, keep a person in the loop, measure honestly, and let the results set the pace.