Agentforce's · MindBlaze

Salesforce shipped two marketing agents in Winter ‘27, and they ask for very different amounts of trust. The Content Agent writes a draft and waits for you to approve it. The Marketing Goals Agent takes a goal, a budget, and an audience, proposes a strategy, and then runs the whole campaign and keeps changing it on its own.

One is an assistant. The other is a delegate. Treating them as the same kind of thing is the mistake I expect to see most this release.

If you only adopt one first, make it the Content Agent. Its failure mode is a bad draft you catch, not a budget spent on a plan you never fully read. The Goals Agent is the more valuable of the two and the one worth the care, because autonomy over spend and channel is exactly where an agent earns money or loses it.

These are not the Campaign Creation Agent that arrived with Marketing Cloud Next earlier in 2026. That one builds a campaign skeleton from a prompt. These two go further: one generates the actual creative, the other runs the campaign after launch. Both require qualified Marketing Cloud editions with the right add-ons, so verify whether your edition even sees them.

The Content Agent: a drafting tool that respects your brand book

The Content Agent’s job is narrow and useful. You describe a campaign objective in plain language, and it generates on-brand email and SMS content, pulling from your brand guidelines, campaign briefs, and media assets stored in Salesforce or uploaded as documents. It supports AI-generated and your own imagery, lets you edit inline, and publishes straight to Salesforce CMS.

The phrase doing the work there is on-brand. A generic model writes fluent marketing copy that sounds like every other brand’s fluent marketing copy. The Content Agent only beats a raw chat window if you feed it a real brand book: tone rules, banned phrases, the way you write a subject line, the assets you are cleared to use.

The quality of the draft tracks how structured that grounding is, the same truth that governs every grounded agent. Thin brand guidelines produce thin output with your logo on it.

What it does not do is remove the reviewer. Email and SMS are regulated surfaces. Consent, unsubscribe handling, claims substantiation, and the rules of CAN-SPAM and its equivalents are not things you delegate to a draft generator.

So the Content Agent shortens the distance from brief to a sendable draft, and the person who signs off that the draft is legal, accurate, and on-message is still a person. Used that way, inline editing and CMS publishing make it a real accelerator for a team that drafts a lot of variants.

Because it runs on Agentforce, every generation is an action on the Flex Credit meter. Drafting ten subject-line variants is ten actions, not one, so regenerating until something clicks has a running cost. Get the brief and the brand grounding good enough that the first or second draft is usable, rather than brute-forcing quality with volume.

The Marketing Goals Agent: autonomy you approve once

The Goals Agent is the ambitious one. You give it three inputs: a goal, a budget, and an eligible audience. It returns a recommended strategy for you to review and approve.

Once you activate it, it operates autonomously inside your guardrails, continuously choosing the campaign, channel, content, and timing for each individual customer and adjusting toward the outcome you named. Salesforce calls these self-optimizing campaigns, which is a fair description of the mechanism.

The shift in what a marketer approves is the real story. The old approval was a campaign: these emails, this audience, this send time, launched once. The Goals Agent’s approval is a policy: this objective, this spending limit, this eligible population, run continuously, with the specific decisions delegated. You are not approving the sends. You are approving the rules under which the agent makes them.

Powerful, and where the care goes. Three guardrails decide whether you get a payoff or a slow leak:

The budget is the hard stop, so set it as one. The budget you give the agent is the ceiling on autonomous spend, and the single number that bounds the blast radius of a bad strategy. Set it to what you will let an agent spend without a human re-reading the plan, not an aspirational annual figure.
The eligible audience is a safety control first, a target second. The population you make eligible is the population the agent can act on. Exclude the segments where a wrong message is expensive, regulated industries, recent complainants, high-value accounts you handle by hand, and you keep the autonomy pointed at traffic where a mistake is cheap.
The goal has to be the real goal. An agent optimizing toward a proxy metric finds the cheap route to that proxy. Tell it to maximise opens and it optimises for clickbait subject lines. Name the outcome you want, then watch for the agent satisfying the letter of the metric while missing the point, the marketing version of reward hacking.

The autonomy also means the agent’s decisions draw actions continuously, not once at launch. A self-optimizing campaign is, on the meter, an agent that keeps acting. Fine when the outcome is worth it, but it belongs in your Flex Credit monitoring alongside your service agents, not in a blind spot labelled “marketing.”

Where both agents get their facts

Neither agent is useful on generic knowledge. The Content Agent needs your brand and assets; the Goals Agent needs your audiences, your channels, and what has worked before. That grounding is the project, which is why these agents land inside Marketing Cloud Next rather than as a standalone toy.

The customer data, consent state, engagement history, and audience definitions they reason over come from the platform’s data layer, which for a serious build means a Data 360 foundation underneath.

The implication for sequencing is blunt. If your audience data is messy, your consent state is unreliable, or your brand guidelines live in a PDF nobody has opened since the last rebrand, these agents amplify that mess at machine speed.

Get the grounding right before you hand an agent a budget. An autonomous campaign agent running on bad audience data is not a marketing tool, it is an expensive way to annoy your best customers.

What I would do with them

Adopt the Content Agent now, as a drafting accelerator with a named reviewer and a structured brand book behind it. The risk is low, it pays off immediately for teams producing many variants, and it teaches your team how these agents behave before you give one a budget.

Pilot the Goals Agent deliberately: one objective, a budget small enough that a bad week is a line item and not an incident, and an eligible audience that excludes everyone a wrong message would cost you. Read the strategy it proposes before you approve it, the way you would read a plan from a new hire, and keep reading its decisions for the first few cycles.

The pilot is not there to prove the agent works. The point is to find where your guardrails are too loose before the budget is large enough to matter.

The pattern underneath both is the one that holds for every agent that acts rather than answers. Draft-and-approve is safe and dull, and you should take it everywhere it fits. Approve-a-policy-and-delegate is where the payoff is, and it is only safe when the budget, the audience, and the goal are set like the controls they are.

The marketers who win with the Goals Agent will be the ones who treated the human-in-the-loop decision as the actual design work, not an afterthought bolted on when legal asked.

Understanding the basics

What is the difference between the Agentforce Content Agent and the Marketing Goals Agent?

The Content Agent generates on-brand email and SMS content from a natural-language brief, your brand guidelines, and your assets, then waits for you to edit and approve it before anything sends. The Marketing Goals Agent takes a goal, a budget, and an eligible audience, proposes a strategy for your approval, and then runs the campaign autonomously inside those guardrails, continuously optimizing channel, content, and timing per customer. One drafts with a human in the loop; the other executes a policy you approved once.

Do I need Marketing Cloud Next to use these agents?

Yes. Both agents are part of Marketing Cloud Next, Salesforce’s rebuilt marketing platform also branded Agentforce Marketing, and they require qualified Marketing Cloud editions with the appropriate add-ons. The first step before planning any adoption is confirming your edition and licensing include them, because availability is gated rather than universal across every Marketing Cloud customer.

How much do the marketing agents cost to run?

They are Agentforce agents, so their work meters as actions on the Flex Credit system rather than as a flat seat. Each content generation and each autonomous decision the Goals Agent makes is a billable action, which means a self-optimizing campaign draws credits continuously while it runs rather than once at launch. Budget for them the way you budget any agent: monitor consumption in Digital Wallet, and get the grounding good enough that you are not paying to regenerate weak output.

Can the Marketing Goals Agent spend money without approval?

It spends inside the budget and audience you approve when you activate it, and only there. You approve the strategy and the guardrails once; the agent then makes the individual send and channel decisions autonomously up to the budget ceiling and only against the eligible audience. Which is why the budget and the eligible audience are safety controls first and targets second: they are the hard limits on what the agent can do without coming back to a human.

Planning a Marketing Cloud Next rollout, or deciding how much autonomy to hand a campaign agent before it touches your budget? Talk to us. Scoping agents so they are useful and governed, with the guardrails set like controls rather than afterthoughts, is the work we do.