Today, October 12, Salesforce’s Winter ’27 release goes generally available. Buried in that release, for marketing teams on Marketing Cloud Next, is Campaign Agent: an agent that takes a goal, builds a campaign around it, and then keeps deciding which of your live campaigns should reach each contact.
Salesforce’s product page is blunt about the pitch. You set the strategy, the brand guidelines and the guardrails. The agent assembles the audience, the content and the channels. It generates multilingual, multichannel variants. Then, once the campaign is live, it prioritizes and suppresses sends so one campaign wins the contact’s attention and the others wait.
Here is my read. The interesting part is not that an agent can draft an email. The interesting part is that the agent now sits between every campaign you have running and the same inbox. If your suppressions, consent records and frequency caps are scattered, the agent will amplify that mess at machine speed. If they are clean and in one place, you finally get the inbox arbitration marketers have been building by hand for years.
What shipped, and what it costs
Salesforce has been previewing Campaign Agent since Dreamforce. The September 14 product blog from Brett Bothe and Max Kiely walks through the two phases. In coworking, a marketer prompts something like “Build a campaign to grow subscriptions,” and the agent returns a brief with a goal and KPI, live dates, channels, budget, audience inclusion and exclusion criteria, and a first pass at content. Content spans email, RCS, MMS and SMS, in English, French and Spanish, grounded in approved brand assets. Marketers review and approve before anything goes live.
In decisioning, the agent keeps working after launch. It prioritizes which campaign reaches a given customer, assembles the exact language, content variant, channel and send time, and optimizes against performance: shifting budget, pausing an underperforming offer, expanding an audience. Salesforce’s own framing is that marketers “set goals and guardrails” rather than hand-build journeys.
The Pricing page puts Marketing Cloud Next Advanced at $3,250 per org per month, billed annually. The Campaign Agent product FAQ adds the important footnote: limited Campaign Agent capabilities (agentic campaigns and agentic content) come with Marketing Cloud Next Advanced, and “additional agentic campaigns and content generations will be available for an additional cost.” Salesforce has not published that incremental price. Winter ’27 itself is GA today, per Salesforce’s August 31 release announcement and MarTech’s September 4 rundown of the marketing updates.
Account Discovery Agent is listed as GA for October 2026 on the Marketing Cloud “What’s New” page. It surfaces high-potential accounts and buying groups for outreach prioritization. That is the other half of the B2B loop, and it only matters if your CRM account data is clean enough to trust.
Why arbitration is the product
Most “AI campaign” announcements stop at assembly: draft the brief, draft the copy, pick a segment. Campaign Agent’s send prioritization is different. Salesforce says the agent “evaluates competing campaigns against each customer’s intent and context, prioritizing the best fit and suppressing the rest.” Each response feeds the next decision.
That is a job most brands still do with overlapping journeys, exclusion Data Extensions, and a weekly meeting where someone argues that the nurture should yield to the promo. The agent promises to do that continuously. AI Agent Chronicle’s October 11 write-up put the operator implication cleanly: nothing ships without approval, and the suites are competing on whose guardrails hold.
The risk is structural, not theatrical. An agent that suppresses the wrong messages needs to know who must not be contacted. In a lot of Marketing Cloud Engagement accounts, that knowledge is split across exclusion lists, consent fields, journey exits and rules buried in Automations. CGC Agency’s October 6 Dreamforce analysis made the same point for SFMC operators: “an agent is only as good as your data, your consent records and your suppression rules.” Their practical advice was one suppression Data Extension fed by a query, plus a 30-day send-pressure check, before you trust any agent to pick the send.
This matters more in Q4 than it does in a quiet February. Holiday promos, loyalty pushes, product launches and win-backs all pile onto the same contacts in the same three weeks. A human team usually notices when someone gets four messages in two days. An agent optimizing toward a subscription KPI will not notice unless the frequency rule is in the data it reads.
I agree with CGC’s framing. The prompt is the easy part. The ceiling is the product.
What “approve” has to mean
Salesforce is careful to keep marketers in the loop. The product FAQ says Campaign Agent is grounded in approved assets and brand guidelines, and that marketers “remain in the loop with full control to review, refine, and approve content and guardrails before Campaign Agent launches and sends.”
That sentence is doing a lot of work. Approve content is one kind of approval. Approve an audience expansion, a budget increase, a channel shift into SMS, or a lookalike of a high-pressure segment is another. If your only approval step is “looks good, send it,” the agent will optimize toward whatever KPI you named, including the ones that quietly burn your list.
Write down, before go-live, which changes need a human every time. A useful starter list: any budget increase over a fixed percentage, any move into SMS or RCS for contacts who have not engaged in 90 days, any audience expansion beyond the original inclusion rule, and any suppression override. Put a named owner on each. The agent can draft. A person still owns the ceiling.
Also decide what happens when two goals collide. Welcome series versus win-back versus a seasonal promo will all want the same high-intent contact. Salesforce says the agent arbitrates in the customer’s favor. Your brand still needs a written rank order for when “customer favor” and “this week’s revenue target” disagree.
Price and packaging reality
Do not plan a rollout from the press alone. Marketing Cloud Next Advanced is the seat of limited Campaign Agent capabilities. Extra agentic campaigns and content generations cost more, with no public rate card for the overage. Existing Engagement, Account Engagement, Personalization and Intelligence customers reach next-gen features through “+ ” editions that sit alongside current products rather than replacing them. That is Salesforce’s packaging line, not a migration promise. Ask your rep what “limited” means in your edition, how many agentic campaigns are included, and what a content generation costs once you exceed the bundle.
If you are still on classic Marketing Cloud Engagement and not on Next, Campaign Agent is not a toggle you flip this afternoon. The Winter ’27 release also ships Adaptive Web Quick Setup, Personalized Paths in Flow, Mobile Flash Sending and Headless Marketing (MCP) as October GA items on the What’s New page. Those are adjacent tools, not substitutes for a clean suppression layer.
What to do this week
Build one suppression list the agent can actually see. Pull consent withdrawals, unsubscribes, held and bounced statuses, legal holds and any hard brand exclusions into a single source. If you are on Engagement today, CGC’s pattern of one SQL-fed Data Extension is a fine start. If you are already on Marketing Cloud Next, map the same idea onto whatever exclusion object your agent reads. The point is one place, not twelve.
Measure send pressure before you hand the agent the keys. Count messages per contact over the last 30 days. Pick a ceiling (eight is a common starting point for B2B; consumer brands often sit lower) and mark everyone above it. Those contacts are the first ones an arbitration agent should protect, and the first ones a messy account will over-message.
Pick one goal for a pilot, not your whole calendar. “Grow subscriptions” or “reactivate 90-day lapsed buyers” is enough. Keep a holdout group the agent cannot touch. Define success before the first send: a primary KPI the agent optimizes, plus a list-health metric it is not allowed to burn (complaint rate, unsubscribe rate, or 30-day send pressure).
Write the approval matrix on one page. Content approval, audience expansion, budget change, channel change, suppression override. Name the owner of each. Share it with marketing ops and legal before anyone connects the agent to a production send.
Ask Salesforce three questions in writing. Which Campaign Agent capabilities are included in your edition this week? What does an additional agentic campaign or content generation cost? Where do suppressions and consent live for the agent’s prioritization step, and can you audit why a contact was suppressed last Tuesday?
Brief the teams that will feel the side effects. Sales needs to know Account Discovery may change which accounts get worked. The contact center needs to know if two-way conversational email (also called out in MarTech’s Winter ’27 marketing list) will land agent replies in their queue. Brand needs to know which assets the agent is allowed to ground on, and what happens when a generated variant drifts off the approved set.
One more check before you scale: ask marketing ops to export last week’s sends for a sample of 200 contacts and mark every collision (two or more campaigns in 48 hours). That spreadsheet is your before picture. After the pilot, run the same sample. If collisions went up, the agent is assembling, not arbitrating. Fix the ceiling before you give it more goals.
Campaign Agent is a real shift in who picks the send. Treat the first week as plumbing, not as a creative showcase. Set the ceiling. Then let the agent work inside it.
