The enterprise software landscape is currently navigating a period of profound transition, characterized by a growing tension between massive investment in artificial intelligence and the tangible business "yield" resulting from those expenditures. Bill Patterson, Executive Vice President of Corporate Strategy at Salesforce, recently addressed this disconnect, highlighting a sentiment increasingly common among Chief Financial Officers (CFOs) who find themselves paying for advanced AI capabilities that remain underutilized or fail to deliver measurable returns. This "yield gap" has prompted a strategic pivot within Salesforce and the broader industry, moving away from the consumption-based "tokenomics" of the early generative AI era toward a model focused on autonomous agents and outcome-based value.
The Yield Gap: When Usage Fails to Generate Value
Since the public debut of sophisticated large language models (LLMs) in late 2022, enterprise adoption has been characterized by rapid experimentation. However, Patterson notes that while adoption rates have been impressive, the transition from "use" to "yield" remains the primary hurdle for the modern enterprise. Organizations have successfully integrated AI to generate marketing content, draft emails, and assist in software development, yet these activities often fail to translate into departmental or enterprise-level breakthroughs.
According to recent industry data from Gartner, while over 80% of enterprises have used generative AI APIs or deployed GenAI-enabled applications in production environments by 2024, many are struggling to move beyond the "pilot" phase. The challenge lies in the lack of clear practitioner guidance. For many companies, the initial excitement of AI has been replaced by the realization that generating more code or more content does not inherently lead to a more efficient or profitable business unit. Patterson argues that the industry is currently witnessing a "sense-making" moment where the focus must shift from how much AI is being used to how that usage solves specific business problems.
The Crisis of Tokenomics and the "Expensive Google Search"
A central theme in the current AI discourse is the concept of "tokenomics"—the economic model based on the consumption of tokens (units of text or data processed by an LLM). In the early stages of the GenAI boom, companies were encouraged to track their token "burn" as a metric of progress. Patterson critiques this approach, suggesting that high token consumption often masks a lack of strategic utility. In many instances, employees have used expensive enterprise AI tools for tasks that could have been handled by a standard search engine, resulting in what Patterson describes as a "very expensive Google search."
The financial implications of this model are significant. For a mid-to-large-scale enterprise, unmanaged token consumption can lead to six-figure monthly invoices with little to show in terms of increased revenue or cost savings. Furthermore, the quality of AI output remains a point of contention. In software development, for example, studies have indicated that while AI-assisted developers may write more code, the resulting "code churn" or technical debt can actually slow down product shipping cycles. This realization is forcing a re-evaluation of the "token-maxxing" philosophy, where the goal was simply to maximize the volume of AI interactions.
Chronology of the AI Evolution: From Co-pilots to Agents
The trajectory of AI in the enterprise has moved at an unprecedented pace, compressed into a timeline of less than three years.
- The Post-Pandemic Labor Crisis (2021-2022): Following the global pandemic, companies faced a significant shortage of human labor. This created a demand for "digital labor" that could handle the increasing volume of customer interactions and data management.
- The Co-pilot Era (2023-2024): The industry entered the "Co-pilot" phase, where AI acted as an assistant to human workers. This required a "human-in-the-loop" for every action, often resulting in a bottleneck where the human had to review and edit every AI-generated suggestion.
- The Agentic Turn (Present-2025): The industry is now moving toward "Agentic" AI—autonomous entities capable of executing multi-step processes with minimal human intervention. Unlike Co-pilots, agents like Salesforce’s "Piper" (a Sales Development Representative agent) are designed to qualify leads, negotiate within set parameters, and take actions across different software systems to close a process.
Patterson suggests that the Co-pilot era was a necessary but brief evolutionary step. The current shift toward agents represents a move toward "Digital Labor" that can operate at scale, helping companies catch up with customer demands that have outpaced human capacity.
Strategic Pivot: Outcome-Based Pricing and Headless Platforms
To address the CFO’s concerns regarding "shelfware"—software that is paid for but not used—Salesforce is advocating for a shift in business models. The traditional per-seat licensing model is being supplemented or replaced by "pay-as-you-go" frameworks. This ensures that enterprises only pay for the value they extract, rather than the potential capacity of the software.
Furthermore, Salesforce is opening its platform in a "headless" manner. This allows for greater interaction between data and process experiences, inviting more innovation from third-party developers and internal IT teams. By decoupling the user interface from the underlying logic and data, Salesforce aims to make its software more ubiquitous, allowing AI agents to trigger actions within the CRM from various external environments.
This strategy is a direct response to the "token movement." By focusing on business outcomes—such as whether a sales lead was closed or a customer service issue was resolved—Salesforce seeks to distance itself from LLM providers who monetize based on the raw volume of data processed.
Redefining Sales and Service Through Agency
The introduction of autonomous agents necessitates a fundamental rethinking of core business processes. In the traditional sales model, human intervention often introduced delays; a negotiation could take days as emails were exchanged and approvals sought. In the agentic model, negotiation can be reduced to a set of prompts with predefined acceptable ranges. An AI agent can negotiate on behalf of a salesperson in real-time, drastically accelerating the sales cycle.
Similarly, in customer service, the industry is moving away from rigid "decision trees" (where a chatbot could only offer a few predetermined options) toward a world of "agency." Modern agents can understand the nuances of a customer’s identity, perform complex actions across multiple systems, and make decisions that are beneficial for both the customer and the business. This transition is as much a "change management" exercise as it is a technological leap, requiring organizations to trust AI to handle "entitlements" and "adherence" autonomously.
The "SaaSpocalypse" and the Myth of Vibe Coding
Despite the rapid changes in the industry, Patterson remains dismissive of the so-called "SaaSpocalypse"—the theory that generative AI will allow companies to easily build their own custom software, rendering traditional Software-as-a-Service (SaaS) providers obsolete. This trend, sometimes referred to as "vibe coding," suggests that developers can use AI to generate entire bespoke CRM or ERP systems with minimal effort.
Patterson characterizes this narrative as "cognitive warfare" and "utter nonsense." He argues that while AI can generate code snippets, it cannot replicate the ten years of institutional knowledge, security protocols, data integration, and compliance frameworks built into a platform like Salesforce. The binary proposition—building a custom CRM versus buying an established one—ignores the immense complexity of maintaining enterprise-grade software. For Salesforce, the focus remains on whether the service helps customers grow or save money, rather than engaging in speculative debates about the death of the SaaS model.
Broader Implications and Future Outlook
As the enterprise moves into 2025 and 2026, the primary metric for AI success will be "proof." The industry is exiting the era of hype and entering an era of validation. For organizations, this means moving beyond small-scale pilots and "hitting the throttle" on use cases that have demonstrated clear ROI.
The implications for the global workforce are profound. The rise of "Digital Labor" is not necessarily a precursor to mass unemployment but rather a necessary response to a structural labor deficit. By automating the "interrupts" in human-led processes, companies can increase their overall output without a linear increase in headcount.
However, the path forward requires caution. Patterson acknowledges that many companies are still hesitant, waiting for validated proof before making large-scale commitments. The success of the next phase of enterprise AI will depend on the ability of tech providers to move away from the "burning of tokens" and toward the delivery of "yield." Those who can demonstrate that their AI agents are closing leads, resolving tickets, and streamlining operations will likely lead the next decade of digital transformation, while those stuck in the "tokenomics" mindset may find their budgets under increasing scrutiny from skeptical CFOs.
