By Jacques Bughin
Historically, enterprise software was priced on a per-seat basis. Today, in the era of generative and, especially, agentic AI, the (ever-decreasing) number of humans involved becomes less relevant. Now, the more relevant metric is the amount of work the software performs. Attention SaaS providers – the time for reinvention is nigh.
In early 2026, the enterprise software industry crossed a threshold that many observers had anticipated but few had fully internalized. Software was no longer merely assisting human workers; it was increasingly executing work itself. As generative AI and agentic systems moved from experimental copilots to semi-autonomous digital operators, the economic foundations of enterprise software began to shift.
This transition is now widely described as “the Great Decoupling”, the moment when the dominant pricing unit of enterprise software—the human seat—began separating from the unit of value delivered by software, which is the work performed.
The shift becomes visible first in financial signals.1 Global IT spending is projected to exceed $6 trillion, with enterprise software surpassing $1.4 trillion,2 yet investors increasingly question whether the seat-based subscription model that powered the SaaS era can remain stable in a world where AI agents execute tasks that previously required large human teams.
The deeper economic shift, however, lies not in technology infrastructure but in where software captures value. For decades, enterprise software competed against other IT budgets. CIOs allocated spending between applications, infrastructure, and services, and the rise of SaaS largely represented a reallocation of technology spending from on-premise systems to cloud platforms. Once software begins executing tasks directly—responding to customers, writing code, qualifying sales leads, processing invoices—it begins competing against labor costs rather than software budgets.
Labor remains the largest cost pool in modern economies. Across advanced economies, compensation of employees represents roughly 60 percent of value added in GDP, according to OECD national accounts.3 This macroeconomic reality fundamentally expands the opportunity for software once automation moves beyond tools and into task execution. Historically, automation estimates were conservative because they focused primarily on routine execution tasks such as manufacturing operations or standardized back-office workflows. Earlier research suggested that roughly 30 percent of work activities could be automated using classical technologies.4 Generative AI expands this frontier dramatically because it automates cognitive work previously considered resistant to automation. Large language models can draft documents, generate code, analyze data, summarize contracts, and conduct customer interactions. Studies from institutions including McKinsey and Goldman Sachs suggest that 25 to 50 percent of work activities may now be affected by AI systems,5 even when automation is partial rather than complete. Once this shift is incorporated into macroeconomic reasoning, the implications become clear. If labor represents 60 percent of GDP, if roughly 30 percent of work activities are affected by AI, and if productivity improvements of around 30 percent are achieved in those activities, the economic value created approaches 5 percent of GDP. Even conservative assumptions, therefore, imply that the value unlocked by AI-driven automation is several times larger than the current global enterprise software market.
This is why the AI transition represents something fundamentally different from previous software cycles. Software is no longer competing primarily for IT budgets. It is competing for labor economics.
From Seats to Work
Despite the dramatic macro implications, the transition is unfolding gradually and unevenly. The software industry is moving through three overlapping stages that progressively separate software value from human seats.
The first stage is assistive AI, where generative systems function as copilots that improve the productivity of human users. In this phase, the traditional SaaS architecture remains intact. Users log into applications, and AI functions operate as productivity enhancements embedded within those applications.
Microsoft’s Copilot ecosystem illustrates this stage well.6 Copilot does not replace Microsoft Office; instead it enhances it. Developers continue using GitHub, designers continue using design tools, marketers continue using collaboration platforms, but AI functions accelerate output. Pricing therefore remains anchored in the seat model, often through AI add-ons or premium tiers. Many SaaS companies adopted similar strategies. Notion introduced AI capabilities as a paid add-on to existing plans, Slack bundled AI features into higher subscription tiers, and GitHub Copilot remains priced per developer seat. The economic logic of this stage is simple: AI increases the value of the seat, but the seat remains the unit of consumption.
However, once AI begins executing tasks independently rather than assisting humans, the seat becomes a poor proxy for value. This leads to the second stage: semi-autonomous workflows, where software executes tasks rather than merely assisting them. Here the market has already begun experimenting with new pricing models that price work rather than access.
AI agents are not priced primarily by the number of employees accessing software. They are priced by the amount of work the software performs.
The most widely cited example is Intercom’s Fin AI agent, which handles customer support conversations autonomously. Instead of charging per user license, Intercom charges approximately $0.99 per resolved customer interaction.7 The genius of this pricing approach lies in its alignment with operational metrics that customer-support leaders already track internally. Most support organizations know their cost per ticket or cost per resolution. When an AI agent resolves a support request for less than a dollar, the economics become immediately legible to CFOs. Zendesk has adopted a similar approach. Rather than replacing its seat-based pricing entirely,8 the company introduced automated resolution pricing in the range of $1.50 to $2 per automated ticket resolution, effectively creating a hybrid structure where human agents remain priced per seat while AI agents are priced per outcome. Salesforce’s Agentforce initiative illustrates the same transition from another angle. Instead of measuring value by CRM users, Agentforce introduces pricing around AI-driven conversations and automated interactions,9 with benchmarks around $2 per conversation. In this case, Salesforce is not abandoning subscription revenue but layering a usage-based “digital labor” metric on top of existing SaaS subscriptions.
These examples demonstrate an emerging pattern. AI agents are not priced primarily by the number of employees accessing software. They are priced by the amount of work the software performs.
This pattern is spreading beyond customer support. In design and productivity software, companies such as Figma have introduced AI credit systems that meter generative design tasks.10 Collaboration platforms like Miro bundle AI credits within seat tiers, effectively linking AI usage to computational work rather than user access. Marketing platforms such as HubSpot have begun introducing AI-driven content and automation credits that scale with the number of tasks executed by AI agents.
Even observability and infrastructure companies illustrate this transition. Datadog, long before the rise of generative AI, had already developed a hybrid pricing model that charges both platform subscriptions and usage-based fees tied to logs, traces, and infrastructure events. In retrospect, this pricing structure resembles the hybrid models that AI companies are now adopting: a stable platform layer combined with metered operational work.
The third stage of the decoupling process is only beginning to emerge but is already visible in the strategies of several AI-native startups. Here, software no longer merely executes workflows; it performs entire roles. Companies such as 11x.ai
market AI agents as digital sales development representatives capable of prospecting,11 qualifying leads, and initiating customer conversations autonomously. Rather than charging per seat or per interaction, these platforms often anchor their pricing relative to the cost of hiring human employees. The value proposition becomes explicit: instead of hiring an additional salesperson or analyst, a company deploys a digital worker. At this point, the economic comparison is no longer between software tools. It is between human labor and software labor.
The Structural Advantage: Data and Action
Pricing innovation attracts attention, but the deeper competitive advantage lies elsewhere. The companies most likely to dominate the agentic AI era are those that control both the data layer and the action layer of enterprise workflows.
Enterprise software historically divided into systems of record and systems of engagement. Systems of record stored authoritative data—customer records in CRM systems, financial transactions in ERP platforms, identity credentials in authentication systems. Systems of engagement provided user interfaces through which employees interacted with that data. Agentic AI introduces a third layer: systems of action.12 A system of action does not merely display data. It performs operations based on that data. An AI support agent resolving a customer issue, an AI finance system reconciling transactions, or an AI coding agent submitting patches to a repository are all examples of systems of action.
The strategic implication is that the most defensible AI products will combine three capabilities simultaneously: ownership of the underlying data model, visibility into operational workflows, and the ability to execute actions within those workflows. Consider the example of customer support again. Intercom’s advantage does not come solely from building a conversational AI interface. Its advantage comes from operating as a system of record for customer conversations and a system of action for resolving those conversations. Because the platform already stores conversation history, customer metadata, and workflow triggers, it possesses the context necessary to deploy autonomous agents effectively.
The same logic applies to CRM platforms. Salesforce’s ability to deploy AI agents within customer interactions depends on its control over the underlying data graph of leads, accounts, opportunities, and communication histories. In other words, the platform’s historical role as a system of record becomes the foundation for its new role as a system of action.
This pattern will likely define the next generation of SaaS competition. The companies that win will not simply offer AI features. They will control the data models and execution environments where AI agents operate.
How SaaS Companies Must Reinvent Themselves
Are SaaS providers facing the inevitable fate of SaasPocalypse in this decoupling of work through agentic AI? At first glance, the shift toward digital labor appears to favor AI-native startups rather than established SaaS companies. After all, startups have historically captured technological transitions by building new architectures without legacy constraints.
Yet history suggests that incumbents often retain powerful advantages during platform transitions. During the early SaaS era, many observers believed that on-premise software companies would dominate cloud computing because they controlled customer relationships and enterprise data.13 Instead, new SaaS platforms such as Salesforce and Workday emerged precisely because incumbents underestimated the architectural shift required.
The AI transition, however, differs in one crucial respect. Today’s SaaS incumbents already control the operational layer of enterprise workflows. They own the systems where work already happens. A CRM platform contains the history of customer interactions. A customer-support platform stores conversation data and resolution workflows. A finance platform contains transactional records. These systems represent both systems of record and systems of action, which are precisely the environments where AI agents must operate.
Today’s SaaS incumbents already control the operational layer of enterprise workflows. They own the systems where work already happens.
AI-native entrants face a difficult challenge: they must build powerful models and agent frameworks while simultaneously integrating into enterprise systems that they do not control. SaaS incumbents therefore possess several structural advantages. They control the data graphs that provide context to AI agents. They operate the workflows where tasks occur. They maintain trust relationships with enterprise customers who are cautious about deploying autonomous systems.
Moreover, they possess distribution channels that AI-native startups lack. Enterprise sales forces, partner ecosystems, and installed customer bases allow incumbents to deploy AI capabilities at scale once the technology matures. The transition therefore resembles earlier technology shifts where incumbents initially move slowly but ultimately leverage their structural advantages once the new paradigm stabilizes.
Based on those assets and capabilities, SaaS players must thus quickly launch a reinvention of the classic SaaS playbook. In the AI era, these four levers must be redesigned:
First, product strategy must evolve from building applications that humans operate to building systems that execute workflows autonomously. Instead of focusing primarily on user interfaces, successful platforms must design structured APIs, event architectures, and orchestration layers that allow AI agents to interact with enterprise systems reliably. The best products will increasingly resemble operational platforms rather than standalone applications.
Pricing strategy must shift from seat-based subscriptions toward hybrid models that combine platform subscriptions with usage or outcome metrics. Vendors that cling too rigidly to seat pricing risk losing relevance as AI reduces the number of human users required to operate enterprise workflows.
Distribution strategy also changes. Traditional SaaS companies optimized for human adoption through user interfaces and enterprise sales. In the agentic era, software must also distribute through machine interactions. APIs, automation hooks, and agent frameworks become critical channels through which software is invoked by other systems.
Promotion strategy must evolve from feature marketing toward economic storytelling. Instead of selling productivity improvements to individual users, vendors increasingly sell labor substitution and operational outcomes to executives. The narrative shifts from “this tool helps your team work faster” to “this platform performs work that previously required human teams.”
However, the transition is not trivial. Many SaaS companies will struggle precisely because they misunderstand how deeply these four dimensions must change.
The most common mistake is to treat AI as a feature rather than an architectural shift. Companies that simply embed generative AI within existing applications without redesigning their workflow architectures often fail to unlock meaningful automation.
Another frequent mistake is misaligned pricing. Vendors that attempt to charge purely for AI access without linking pricing to measurable outcomes often encounter resistance from procurement teams that cannot easily justify the expense. A third mistake involves ignoring the importance of data ownership. AI agents require context to function effectively. Platforms that lack control over the underlying data layer struggle to deploy autonomous systems because they cannot access the information required for decision-making.
Finally, many companies underestimate the importance of ecosystem strategy. Agentic AI systems increasingly operate across multiple software platforms.14 Vendors that fail to expose APIs and integration layers risk being bypassed by orchestration platforms that coordinate workflows across multiple systems. In contrast, the companies most likely to succeed are those that redesign their products, pricing models, distribution channels, and market narratives around the concept of software as a digital labor force.
The Future: Digital Labor Platforms
The Great Decoupling does not mean the end of SaaS. Human seats will continue to exist for governance, collaboration, and oversight. Identity systems, permission frameworks, and user interfaces will remain essential parts of enterprise computing. But the center of gravity of software is shifting. Instead of competing for IT budgets, software companies increasingly compete with labor economics. And as generative AI expands automation into cognitive work, through agentic AI,15 the potential economic value available to software becomes dramatically larger than the traditional enterprise software market.
In the coming decade, the most successful enterprise platforms will not simply provide tools for employees .They will operate the infrastructure through which digital workers perform the tasks of modern organizations.
In this perspective, CEOs of SaaS companies must work hard to master three decisions that together determine whether their company evolves into a digital labor platform or gradually becomes infrastructure for someone else’s automation system. The three decisions concern value capture, ecosystem position, and organizational capability. The plan is known; now it is time to deliver.
