The Claudeforce vs Agentforce difference is what ships

If you need an AI system that can participate in paid-media decisions today, the Claudeforce vs. Agentforce question has a fairly practical answer as of August 28, 2026: Agentforce Marketing is the actionable option. Claudeforce is real, but its Marketing, Commerce, Revenue, Tableau, MuleSoft, and Data 360 skills are still labeled “COMING SOON” on Salesforce’s official product page. [1]

That makes “which one uses Claude?” the wrong first question. Both ecosystems now use Claude. The planning question is narrower: which product surface is available, supported, and authorized to affect campaign work—and which one is still a partnership announcement with future skills attached?

Claude is the shared model layer, not the comparison

Claudeforce and Agentforce are not two competing Claude models. Claudeforce is Salesforce and Anthropic’s partnership package, initially presented through “Salesforce in Claude.” The first plugin exposes Salesforce through MCP servers, APIs, and CLI tools and includes 37 prebuilt sales skills built by the two companies. Salesforce says Salesforce in Claude is in a select pilot, with open beta expected in September 2026 and additional skills expected later in 2026. [2]

Agentforce is Salesforce’s native agent platform. Claude is now its default model in several Salesforce surfaces, including Agentforce Coworker, and Claude also serves as a reasoning model in the Atlas Reasoning Engine. Salesforce says Claude runs inside the Salesforce Trust Boundary through Amazon Bedrock. Those details matter for architecture, governance, and model choice; they do not automatically make every Salesforce marketing workflow available through Claude. [2]

The 37-skill figure therefore needs careful handling. It describes the first Claudeforce plugin’s prebuilt sales scope, not 37 live marketing capabilities. A buyer evaluating campaign automation should ask where the skill is listed, whether it is enabled for the org, which permissions it has, what actions it can take, and whether Salesforce documents support for that action.

The availability gap is the actual difference

Salesforce’s Claudeforce page draws the important boundary more clearly than the launch language does. Sales is the skill set associated with the first plugin. Marketing, Commerce, Revenue, Tableau, MuleSoft, and Data 360 are presented with “COMING SOON” labels, and the page does not provide a general-availability date for those capabilities. [1]

That is not a minor packaging distinction. “Coming soon” can mean a credible product direction, but it is not a control a media buyer can test in a production account. It does not establish a permission model, an execution surface, a support commitment, or a budget-review process.

Agentforce Marketing, formerly associated with Marketing Cloud, is further along for this specific job. Salesforce’s marketing pages describe natural-language segment generation, campaign assembly, and paid-media optimization. Its paid-media agents can surface underperforming ads across channels, pause them according to marketer-set thresholds, and provide ROAS-based budget-allocation recommendations. [3]

The distinction between an autonomous action and a recommendation still matters. A pause rule that a marketer has configured is an execution capability. A budget reallocation recommendation is decision support unless the product and account configuration authorize it to carry out the change. Before connecting either behavior to live spend, review the Agentforce paid-media pre-flight checklist for the specific ad accounts, thresholds, approvals, and audit trail involved.

Nothing in the supplied product material supports saying that Claudeforce can run those marketing or campaign workflows today. Its marketing skills remain roadmap items. A custom build might make some underlying Salesforce and Claude components interact, but “possible to build” is not the same as “shipped capability a buyer can select and operate.”

What the paid-media workflow looks like in practice

For a performance team, the useful Agentforce Marketing sequence is concrete:

1. Generate or refine an audience segment from a natural-language request.
2. Assemble campaign components using the available Salesforce marketing context.
3. Monitor paid-media performance across supported channels.
4. Identify ads that meet the marketer’s underperformance criteria.
5. Pause qualifying ads and present budget-allocation recommendations based on ROAS, subject to the configured controls.

That does not eliminate the need for a media buyer. Someone still has to define the threshold, decide which channels and campaigns are in scope, approve the account connection, inspect false positives, and explain a budget change to the client or finance team. It does, however, give the team a documented product surface to evaluate rather than a future skill name.

Salesforce reports figures including 32% marketing ROI, 34% customer-lifetime-value improvement, 32% engagement improvement, and 27% lower customer-acquisition cost as “Salesforce Customer Success Metrics.” These are vendor-labeled customer-success claims, not independent benchmarks, so they should not be treated as proof that an individual paid-media account will achieve those results. [3] The more useful first test is whether the agent performs the documented action under the permissions and thresholds your account actually uses.

Pricing is clearer for Agentforce than for Claudeforce

Salesforce has published a price surface for Agentforce. The relevant list includes Agentforce Foundations at $0, Flex Credits at $500 per 100,000 credits, Conversations at $2 each, Agentforce add-ons at $125 per user per month, Industries add-ons at $150 per user per month, Agentforce 1 from $550 per user per month, and the Agentforce User License at $5 per user per month. [4]

The pricing page also gives worked examples: a two-action use case is shown at approximately $120 per month at 20 requests per day, while a five-action field-service example is approximately $300 per month. Those examples make the consumption model easier to visualize, but they are not a quote for your paid-media workload. The choice between Flex Credits and Conversations, the number of actions, users, and included entitlements all affect the calculation. For a broader ownership model, use the Agentforce TCO comparison as a cost-modeling companion.

Claudeforce pricing is not officially published as of August 28, 2026. A VentureBeat search snippet refers to headless consumption pricing, but that reference has not been verified against a crawlable primary pricing page and should not be used as a number in a budget. The honest status is undisclosed, not “about” any third-party estimate.

Why the launch announcement should not settle the buying decision

Claudeforce was announced on August 26, 2026, alongside Salesforce’s fiscal Q2 FY27 earnings. Salesforce’s stock rose by roughly 14% in extended trading after the announcement, which signals that the partnership attracted substantial market attention. It does not establish that the unlaunched marketing skills are available to a buyer. [5]

MarketScale makes the more useful skeptical point: the count of 37 skills should be examined as a deployment-readiness question—what is shipped, supported, auditable, and usable without a custom build. [6] That is the standard that separates a partnership narrative from an operating decision.

For teams documenting the decision, keep three records separate: the model available in the stack, the agent or plugin enabled in the org, and the action that the system can execute under a named permission. This also prevents Claude’s ad-workflow abilities from being mistaken for a Salesforce marketing-agent entitlement; the model layer and the workflow product are related, but they are not interchangeable.

The decision boundary for late Q3 2026

If the requirement is to generate segments, assemble campaigns, identify underperforming paid ads, pause them under marketer-defined thresholds, or support ROAS-based budget decisions now, Agentforce Marketing is the route supported by the current product pages. Confirm your edition, channel coverage, permissions, approval settings, and pricing configuration before connecting it to live spend.

If the requirement is to evaluate Salesforce in Claude, the current scope is a select-pilot experience centered on the first plugin’s sales skills, with open beta expected in September 2026. Marketing, Commerce, Revenue, Tableau, MuleSoft, and Data 360 should remain on the watchlist until Salesforce changes their status and documents access. [1][2]

References

1. Claudeforce — Salesforce
2. Salesforce and Anthropic Announce Claudeforce: The #1 AI Meets the #1 AI CRM — Salesforce, August 26, 2026
3. Marketing — Salesforce; Agentic Marketing — Salesforce
4. Agentforce Pricing — Salesforce
5. Salesforce and Anthropic Announce Claudeforce in Q2’27 Earnings — Salesforce Ben, August 26, 2026
6. Salesforce’s Claudeforce Lets Claude Run CRM Tasks — MarketScale