Salesforce has made autonomous AI agents standard across its marketing platform, building them into every tier rather than selling them as a paid add-on, and has shifted how they are paid for from per-seat licences toward per-action consumption. The agents, which Salesforce says can plan campaigns, build audience segments, write copy and launch journeys across channels, now ship as the default in Agentforce Marketing — the suite formerly known as Marketing Cloud.

The shift matters now because marketing managers are being asked to put AI to work without a clear sense of what they should and should not automate, while finance teams scrutinise every renewal. Salesforce has answered the first question by making the agents the default, and reframed the second by tying cost to what those agents do rather than how many people hold a licence. Both decisions land on the marketing manager’s desk.

What Agentforce Marketing actually changes

Agentforce Marketing builds the agents directly into the platform rather than bolting them on, running them natively on Salesforce’s Data 360 customer data platform. Salesforce describes it as giving every customer “autonomous AI agents that act as always-on collaborators”, and positions the agents as able to reason, make decisions and execute tasks across the campaign lifecycle without constant manual oversight.

In practice, that centres on a campaign-creation capability the company has built into the platform’s editions. Salesforce says the agents build segments, draft email copy and launch campaigns from a brief, drawing on unified customer data rather than the siloed records that made personalisation difficult in the older product. The company points to the recruitment site Indeed, which it says now builds segments, writes copy and launches campaigns “in seconds” using the tool — a customer result reported by Salesforce, not independently measured.

The distinction worth holding is between what the platform can do and what it does unsupervised. Salesforce positions the agents as autonomous; the working reality for most teams is an agent that drafts and proposes while a marketer approves. The platform makes that delegation easy. It does not decide how much of it is wise.

How the pricing shifts from seats to actions

Salesforce has changed how the platform is paid for, pricing its agents through Flex Credits, a consumption model in which a standard agent action costs 20 credits — roughly 8p ($0.10) — with credits sold at about £390 ($500) per 100,000, according to pricing the company documented through early 2026. Voice actions cost more, at 30 credits each. An older flat rate of about £1.55 ($2) per conversation remains available for customer-facing agents under one buying structure.

That model decouples cost from headcount. A team no longer pays mainly for the number of people holding a licence; it pays for what its agents do. Marketing Cloud Next, the underlying platform, starts at about £1,180 ($1,500) a month billed annually, with a higher Advanced tier documented at roughly £2,560 ($3,250), but the agent activity sits on top as a variable line.

For a marketing manager forecasting a quarter, that is a meaningful shift. A seat-based bill is predictable; an action-based bill rises with use, and a campaign that leans heavily on autonomous generation and optimisation will consume more credits than one that does not. The upside is that spend maps directly to work performed. The risk is a bill that is harder to predict before the work is done.

Why marketers are right to weigh the autonomy carefully

Gartner gives marketers good reason to weigh the autonomy carefully: the firm has predicted that more than 40% of agentic AI projects will be cancelled by the end of 2027, citing cost, unclear value and weak controls. The warning is a reminder that a polished demo and a dependable production workflow are not the same thing, and that the teams getting value from Agentforce Marketing will be the ones that decide deliberately which parts of the campaign to hand over.

The judgement runs task by task. Generating a first-draft segment or a set of subject-line variants is low-risk work an agent does well and a marketer can check in seconds. Launching a journey to a live audience, or reallocating budget automatically, carries consequences that warrant a human gate. Agentforce Marketing supports both, and its consumption pricing rewards neither caution nor abandon — it simply meters whatever the team chooses to run.

None of this displaces the marketer. It moves the work up a level, from producing the asset to defining the brief, setting the guardrails and judging the output. That is a more demanding job than the one the older tool asked for, not a smaller one.

Salesforce has not published a sunset date for the legacy Marketing Cloud Engagement product, and frames the move to the new platform as a gradual convergence rather than a forced migration. The open question for marketing teams is less whether to adopt agentic tools, which Salesforce has made the default path, than how much of the campaign they are prepared to let run without a hand on the wheel.