The Real Story Behind Claudeforce: The Unbundling of Enterprise Software

How AI companies are exploiting enterprise software’s Cognitive Tax blind spot to redistribute value across the software ecosystem.

TL;DR

1. The Announcement Is Not the Real Story

On August 26, 2026, Salesforce and Anthropic announced Claudeforce, an expanded partnership that brings Salesforce data, workflows, business logic, actions, and governance into Claude.

The immediate story is compelling. Salesforce users can interact with live customer and revenue information through Claude.

But the more consequential story is captured in Marc Benioff’s description of the partnership:

“Here, the UI is the AI.”

Salesforce has traditionally delivered its user experience and procedural intelligence within the same application.

With Claudeforce, Claude can own the user experience and orchestration. Users express intent, receive information, review recommendations, and initiate actions through Claude, while Salesforce retains the application-specific procedures, business logic, rules, controls, and transaction execution as the governed enterprise application underneath.

Salesforce argues that this model increases the value and use of its platform. Patrick Stokes, Salesforce’s President of Applications and Marketing, told CIO that customers using Salesforce through an agentic interface are using the platform more than before.

That may be true operationally. Salesforce could process more queries, updates, and transactions. But users may increasingly associate the experience, intelligence, and resulting productivity with Claude rather than Salesforce.

This creates a new question for Salesforce and every enterprise application provider:

When AI owns the user experience and orchestrates the workflow, where does the enterprise application create value, and who captures it?

That is the real story behind Claudeforce. It signals a broader unbundling of enterprise software.

2. How Agentic AI Unbundles Enterprise Software

Two developments are converging to reshape enterprise software: the long-standing, largely unaddressed Cognitive Tax imposed on users and a new architecture that allows AI to assume much of that burden.

In our article, Enterprise Software’s Cognitive Tax, we define Cognitive Tax as the knowledge and effort users need to translate a business requirement into an outcome. Users must know which applications and screens to use, what data to enter, which rules apply, and how to coordinate work across systems. Complaints about insufficient training, poor adoption, unintuitive interfaces, and too many clicks are often symptoms of this underlying burden.

Enterprise applications have traditionally bundled the user experience with application-specific procedures, data, rules, permissions, and transaction execution. Users supplied the cognition and cross-system orchestration needed to connect these capabilities into complete business outcomes.

Agentic AI allows these responsibilities to be unbundled. LLMs interpret user intent. Knowledge representations, including semantic libraries, ontologies, and knowledge graphs, provide the meaning of application data, schemas, and relationships. MCP tools give agents governed access to capabilities embedded within enterprise applications and other services.

With Claudeforce, Claude can own the user experience, cognition, and orchestration. Salesforce remains the governed provider of skills, data, application-specific procedures, business logic, permissions, workflows, and transaction execution.

This unbundling does not eliminate Salesforce. It separates Salesforce’s capabilities from its direct relationship with the user. Claude becomes the intermediary through which users access and combine those capabilities. In this sense, unbundling enables disintermediation: Salesforce continues to perform critical work but may no longer own the user experience or receive full credit for the resulting productivity.

This becomes more consequential because business workflows often span multiple systems. Creating a proposal, for example, may involve CRM, CPQ, inventory in ERP, pricing, credit checks, approvals, and document generation. Each application performs its part while Claude coordinates the complete workflow.

Adding skills from more applications and services creates a flywheel of value accretion in the AI layer. Each new skill expands the workflows Claude can orchestrate. As more workflows begin and end within Claude, usage grows, its relationship with users deepens, and more of the resulting value is attributed to it.

This opportunity is not exclusive to Anthropic. AI companies can move up the enterprise software stack by absorbing the Cognitive Tax and orchestrating capabilities across an expanding ecosystem of applications. Enterprise applications remain essential systems of record and transaction, but the AI layer increasingly owns the experience, orchestration, and perceived value of the outcome.

3. AI Expands Salesforce’s Reach and Reframes Its Value

Claudeforce provides a concrete example of a broader shift. As AI platforms become user experience and orchestration layers, they can expand the reach of Salesforce while changing how value is distributed across its product portfolio.

Headless 360 and Salesforce MCP tools make Salesforce data, permissions, business rules, workflows, and transactions available to Claude and, potentially, other AI-driven experiences.

The strategic pressure comes from the same architecture. When an AI platform owns the user experience and orchestration, users may associate the resulting productivity with that layer while Salesforce performs the governed work underneath.

The impact will vary across the Salesforce portfolio:

• Salesforce applications: Sales, Service, Revenue, and Industry Clouds remain authoritative for their data, permissions, business rules, workflows, and transactions. However, users may increasingly interact with these capabilities through an AI layer rather than their native interfaces.
• Agentforce: Agentforce provides a Salesforce-native environment for building, testing, governing, and operating agents. If a customer adopts Claude, LangGraph, or another platform as its enterprise orchestration layer, Agentforce becomes a second agent environment with separate tools, skills, governance, observability, consumption, and technical requirements. Unless Agentforce provides materially differentiated capabilities, customers may prefer to consolidate orchestration within one platform. Customers already committed to Agentforce and Flex Credits may therefore reassess whether maintaining a separate agent platform delivers sufficient incremental value.
• Data 360: Data 360 ingests or federates data from external systems, maps it into Salesforce data models, and provides unified context for Salesforce applications and Agentforce. If customers build their agents and semantic knowledge in an external platform, using Data 360 as an intermediate mapping layer may add licensing, integration, mapping, consumption, and governance overhead. Customers already using Data 360 for identity resolution, segmentation, activation, analytics, or other business requirements may continue to find value in it. Its value is less clear when it is used solely to provide mapped external data to agents operating outside Salesforce.
• MuleSoft: Connectivity, transformation, API management, and governance remain essential for complex cross-system workflows. However, MCP tools, connectors, and AI-driven orchestration may absorb some of the workflow coordination that would otherwise be implemented within an integration platform.
• Tableau: Governed analytics, semantic context, and trusted metrics remain valuable. However, AI can increasingly formulate questions, perform analysis, and dynamically present results, reducing the need for users to navigate predefined dashboards and reports.
• Slack: Slack can become the collaborative environment in which people and agents work together. Its value as a communication and distribution layer may increase even when the intelligence and orchestration operating within it come from an external AI platform.

The broader pattern is that Salesforce products do not benefit equally from AI-driven orchestration. Core applications retain their importance as governed systems of record and transaction. Other products must demonstrate why their capabilities remain necessary when an external AI platform owns the orchestration and user experience.

AI expands Salesforce’s reach while reframing how each part of its portfolio creates and captures value. Claudeforce makes that shift visible.

4. AI Applications Move Up the Enterprise Software Stack

Claudeforce illustrates an opportunity available to any AI application capable of accessing and orchestrating enterprise systems.

When work begins and ends in the AI application, while the underlying applications continue to govern their data, rules, permissions, and transactions, the AI application assumes a more strategic role. It owns the user interaction and becomes the layer through which users experience the combined value of multiple enterprise systems.

More workflows completed through the AI application increase its usage and strengthen its relationship with enterprise users. As organizations accumulate skills, knowledge, governance, and operating practices within that layer, it becomes more deeply embedded in how work gets done.

Claudeforce is one example. The strategic benefit belongs to whichever AI application becomes the layer through which users convert intent into enterprise outcomes.

5. Customers’ Fast Track to Agentic AI: One Workflow at a Time

Claudeforce demonstrates that customers can adopt Agentic AI without replacing their existing application stack or adopting the proprietary AI capabilities offered by each application vendor.

CRM, ERP, analytics, and other systems can continue to govern their data, rules, permissions, and transactions. A separate cognition and orchestration layer can access their capabilities through skills, connectors, APIs, and MCP tools.

The transition can happen incrementally. Customers can begin with one high-value workflow, connect the required systems, and expand as value is proven. They do not need an enterprise-wide migration or a simultaneous overhaul of their applications.

Customers can adopt a packaged AI application or build their own layer using foundation models, domain-specific SLMs, and specialist agents selected for different functions.

Application vendors’ agent platforms become optional components rather than prerequisites for adopting Agentic AI.

This provides customers with a faster, lower-risk path to Agentic AI. They can modernize how work gets done, one workflow at a time, while preserving the systems that run their business.

6. Partners Must Move from Implementing Applications to Orchestrating Outcomes

For Salesforce and other enterprise software partners, this shift changes where customers need help. Application configuration remains necessary, but it no longer defines the complete transformation.

Customers need partners who can identify Cognitive Tax, redesign workflows around intent, expose governed application capabilities through skills and MCP tools, and coordinate work across systems. This requires business process and domain knowledge in addition to platform expertise.

Partners must also become AI-platform neutral. The customer’s orchestration layer may be Claude, another AI application, or a custom architecture combining multiple models. Optimizing each application independently will not address workflows that span the enterprise.

Partners familiar with a customer’s existing systems have an advantage. They understand the data, business rules, integrations, exceptions, and operating practices that the cognition and orchestration layer must absorb.

Our experience across CRM and CPQ reinforces this point. The hardest challenge is not connecting a model or generating a response. It is capturing the knowledge required to interpret intent, select capabilities, apply rules, manage exceptions, and produce a valid business outcome.

The partner opportunity is moving from deploying software to helping customers build a governed digital workforce that operates it. Partners that combine domain knowledge, enterprise applications, AI, and cross-system orchestration will be best positioned for this transition.

7. Enterprise Applications Are Not Disappearing. Their Role Is Changing

Claudeforce is more than a partnership between Salesforce and Anthropic. It marks the unbundling of enterprise software.

Enterprise applications have traditionally combined data, business rules, transactions, workflows, and user experience. Yet users have remained the cognition and orchestration layer, translating desired outcomes into the actions required across these applications.

With AI, users continue to define the desired outcome, but AI assumes more of the cognition and orchestration. The underlying applications remain governed systems of record and transaction.

Value is not disappearing. It is being redistributed. The strategic question is how much of the value will be captured by the AI layer and how much will remain with the enterprise applications.

Enterprise applications remain essential for trusted data, rules, permissions, and execution. AI applications gain strategic importance by owning user interaction, context, and cross-system orchestration.

For customers, the path to Agentic AI does not require replatforming or replacing existing systems. Partners gain an opportunity to move beyond application implementation and help customers orchestrate complete business outcomes.

The transition will be incremental. Some workflows will remain within enterprise applications. Others will begin and end in an AI application or a customer-built orchestration layer.

Claudeforce makes that future tangible. Salesforce does not need to disappear for Claude to move up the enterprise software stack. That is the real story behind Claudeforce.