Salesforce Expands Agentforce Coworker Across Workspaces, Shifting the AI Decision From Access to Economic Proof

What changed

On November 4, 2026, Salesforce announced a significant expansion of Agentforce Coworker well beyond the product’s original capabilities. Salesforce had already documented Coworker as a beta available to eligible organizations by June 23, 2026.¹ The November event broadened how the product could be licensed, connected, and used across employees’ existing work environments.²

Salesforce listed a $50-per-user monthly option alongside Google Workspace and Microsoft 365 connectivity, Conversation History, and reusable Skills as available at the time of the announcement.² Independent coverage corroborated the pricing, integrations, and broader workflow-delegation positioning.³⁴⁵ The disclosed price establishes a visible unit for procurement, but it does not establish realized value: Salesforce did not disclose how many of its reported users were paying that price.⁴

The availability details require careful separation. Salesforce said Microsoft Teams support and applications for iOS, Mac, and Windows would become generally available during November 2026; that wording does not establish that every surface was generally available on November 4.² The company also set November targets for an MCP server and a Flex Credits trial. Scheduled tasks and Daily Digest were described as expected in December, so they were roadmap items rather than generally available capabilities at the announcement date.² Reusable Skills also predated the announcement as an active rollout: Salesforce release notes show that the capability had begun rolling out during the week of October 19.⁶ The November disclosure therefore packaged an existing rollout with new commercial and distribution elements.

Salesforce reported that Coworker had surpassed 185,000 users and that usage had increased twenty-onefold over three months.² Those figures should be treated as attributed vendor claims. The retrieved independent reporting confirms that Salesforce made the claims, but it does not independently validate the user count, the growth rate, or associated productivity benefits.⁴⁵ The defensible news is narrower but strategically important: Salesforce moved Coworker toward a named-user commercial model and expanded its reach across productivity, collaboration, and device surfaces.

Why it matters

The CIO decision is no longer simply whether an AI assistant should be enabled inside a single enterprise application. Salesforce’s expansion asks customers to evaluate an agent that can operate through productivity suites, collaboration environments, and native applications while retaining conversation history and drawing on reusable skills.²⁶ That changes the approval question from feature access to operating-model fit: which employees should receive an execution layer, which workflows should it cross, and what evidence justifies a recurring per-user commitment?

A named-user price makes that decision more concrete. At the disclosed $50 per user per month, CIOs can model the subscription expense directly, but the available evidence does not establish how much verified labor, cycle-time, or application cost the product removes.⁴ Reported user growth cannot substitute for an economic case because the disclosed figures do not distinguish paid users at the stated price from other users, nor do they independently demonstrate productivity outcomes.²⁴ The analytical implication is that seat count should become an output of measured workflow value, not the starting assumption for deployment.

Integration breadth also takes on a different meaning. Connections to Google Workspace and Microsoft 365, alongside planned or staged availability in Microsoft Teams and native applications, broaden the locations from which users can invoke Coworker.² That can reduce the friction of moving into a separate CRM interface, but broader access is not itself proof of broader value. CIOs need to distinguish distribution—the number of places where an agent appears—from execution—the number of governed, repeatable tasks it can complete with acceptable outcomes. This distinction is an analytical inference from the announced surfaces and capabilities, not a vendor-reported performance result.

Persistent conversation history and reusable skills make the architecture more interesting. Conversation History can preserve continuity, while Skills can package capabilities for reuse.²⁶ Together, they point toward an agent layer that carries context and standardized actions across work environments. The architectural question thus expands beyond whether the underlying model is capable. CIOs must determine where context persists, which skills are approved, how those skills map to business processes, and whether a capability invoked from one surface behaves consistently when invoked from another.

The staged rollout further changes procurement discipline. Features described as generally available during November or expected in December should not be credited as production value on November 4.² A business case that depends on Teams, native applications, scheduled tasks, or Daily Digest should attach funding and scale decisions to verified availability and observed performance in the customer’s own environment. This prevents roadmap value from being capitalized before the required capability can be tested.

The resulting decision model has four linked dimensions: commercially eligible users, workflows with measurable delegation potential, production-ready integration surfaces, and verified economics. That is materially different from approving an application-specific pilot for broad experimentation. CIOs must decide whether cross-workspace reach improves enough real work to support recurring per-user spending, while avoiding the assumption that a wider interface footprint automatically produces savings.

Frontier take

The decisive enterprise AI battleground is becoming the right to serve as the employee’s cross-workspace execution layer—not the race to add another assistant—and the vendors that win that position will still have to earn it workflow by workflow.

Salesforce’s November expansion supports this assertion because the key new changes are not limited to a new conversational feature. The company paired a per-user SKU with Microsoft 365 and Google Workspace connectivity, conversation continuity, reusable skills, and a staged expansion into Teams and native applications.²⁵ The package is structurally closer to an employee-level software layer than to an isolated CRM enhancement. That characterization is analysis based on the announced distribution, context, and reuse mechanisms; it is not a claim that Salesforce has already achieved universal cross-workspace execution.

In Frontier’s analysis, this shift changes what constitutes defensible differentiation. A standalone assistant can be judged largely on the usefulness of its responses within one application. A cross-workspace agent must be judged on whether it can preserve relevant context, invoke approved capabilities, and complete repeatable work across the environments employees already use. Salesforce’s Conversation History and reusable Skills address parts of that design, while the productivity-suite and application expansion broadens access.²⁶ The retrieved evidence, however, does not independently demonstrate the resulting productivity, savings, or execution quality.

That gap between architectural ambition and verified economics is the central issue for CIOs with agents today. Salesforce’s reported 185,000 users and twenty-onefold usage growth may indicate interest, but they remain vendor-reported figures and do not actually reveal how many users pay the disclosed price.²⁴ They also do not show which workflows were completed, what failure or rework occurred, or whether organizations retired other spending. Adoption metrics can inform market attention; they cannot close an enterprise investment case.

The correct response is neither blanket deployment nor reflexive delay. CIOs must treat the per-user agent as a governed execution portfolio. Each eligible cohort should be connected to a defined set of workflows, each workflow should have an approved skill and integration footprint, and each expansion should depend on measured economic or operational improvement. Cross-workspace access should be treated as a multiplier only after the underlying workflow has demonstrated value.

This approach also handles staged availability without conflating announcement scope with production reality. Microsoft Teams and the native applications were assigned November general-availability timing, while scheduled tasks and Daily Digest were expected in December.² Organizations should therefore test the capabilities that are verifiably available and defer benefit assumptions tied to later features. Reusable Skills, which had already begun rolling out before the November announcement, can be evaluated separately from roadmap-dependent automation.⁶

The long-term strategic change extends beyond Salesforce. Software agents are moving toward persistent context, reusable capabilities, and access from multiple employee environments. In analytical terms, this creates a potential control point between users and enterprise applications. But a control point is valuable only when it produces governed execution at an acceptable cost. Integration breadth earns consideration; demonstrated workflow economics earns scale.

Three moves for CIOs

1. — License by workflow-qualified cohort, not organizational headcount Create a cohort model that links each proposed Coworker seat to named workflows, expected monthly task volume, a baseline measure, and a responsible process owner. Begin with users whose work spans the announced Salesforce and productivity-suite surfaces, then measure completed work, cycle time, rework, and any cost displaced. Keep the $50 monthly price visible in the unit economics and exclude unverified productivity assumptions from the approval case.²⁴

   • Decision trigger: Expand a cohort only when measured workflow value exceeds the recurring seat cost by the organization’s required margin for two consecutive review periods. Reduce or reassign seats when usage lacks attributable completed work, or when rework erases the measured benefit.
   • Why now: The per-user SKU turns broad enablement into a recurring portfolio decision, while the disclosed user figures do not reveal how many users pay that price or independently prove savings.⁴ Cohort-based licensing operationalizes the shift from assistant experimentation to economically accountable execution.

2. — Certify cross-workspace skills as production assets Establish a certification register for reusable Skills that records the process owner, permitted data and systems, approved invocation surfaces, expected output, fallback path, and review status. Test each skill from every intended environment rather than assuming that availability in Salesforce, a productivity suite, Teams, or a native application produces equivalent behavior. Treat Conversation History as part of the workflow design wherever continuity is required.²⁶

   • Decision trigger: Permit production use only after a skill completes the defined workflow consistently across every required, generally available surface. Suspend the skill if a new surface, changed integration, or persistent-context behavior causes outcomes to diverge from the approved process.
   • Why now: Reusable Skills and persistent history move value—and operational dependence—from one-off prompts toward repeatable capabilities.⁶ As access broadens across workspaces, CIOs need a governed unit of execution that can be tested independently of vendor-level adoption claims.

3. — Separate available value from roadmap-dependent value Split the investment case into capabilities verified as available on November 4, capabilities assigned general availability during November, and capabilities expected in December. Fund production use only for the first category until the required later capability is verifiably available and tested. Maintain explicit dependencies for any business case relying on Teams, native applications, scheduled tasks, Daily Digest, the MCP server, or the Flex Credits trial.²

   • Decision trigger: Release the next funding or deployment stage only when the dependent capability is generally available to the organization and passes a production-readiness test tied to the target workflow. Recalculate the business case if availability slips or the delivered behavior does not support the assumed benefit.
   • Why now: The announcement combined immediately listed capabilities, an earlier Skills rollout, November availability targets, and December expectations.²⁶ Separating those states prevents roadmap breadth from being mistaken for current execution capacity and keeps scale decisions aligned with provable economics.

Sources

• Salesforce’s Agentforce Coworker Usage Soars 21x in Three Months ²
• Meet Your Users’ New AI Teammate: Introducing Agentforce Coworker ¹
• Build Reusable Skills for Agentforce Coworker ⁶
• Salesforce Agentforce Coworker Brings Autonomous AI to Teams ³
• Salesforce said Coworker had more than 185,000 users; CIOs still need savings evidence ⁴
• Salesforce Takes AI Work Delegation Beyond the CRM Dashboard ⁵