As Enterprise AI Begins Billing for Work, Accountability for Results Rises

A quiet but significant shift is underway in how enterprises buy and measure value from artificial intelligence. Rather than focusing on traditional usage metrics like cost per token, major enterprise buyers and AI vendors are turning to outcome-based pricing, evaluating artificial intelligence on the tangible work it accomplishes. This move, widely championed by OpenAI’s CFO in early 2024 and echoed in annual projections to investors by August 2026, marks the end of the so-called “tokenmaxxing” era, where maximizing token consumption was once equated with extracting value.

Instead, enterprises have begun routing routine tasks to the most cost-effective models and reserving advanced, costly AI for only those functions where its unique capabilities are justified. As industry observers note, this decoupling of effort from value is overdue; an agent that closes more deals with fewer tokens is more useful than one that maximizes activity but delivers less outcome. Companies like Sierra, which prices its AI customer service agents by resolved conversation rather than by the minute or token, and Salesforce, now reporting Agentic Work Units measured by work completed rather than seats occupied, exemplify the new approach.

This fundamental change is also altering the very structure of enterprise software. As AI agents increasingly deliver results autonomously, the traditional user interface is receding. Application value is now found less in dashboards and menus and more in the “harnesses” of skills, documentation, and workflows that encode best practices and domain expertise into agents themselves. Salesforce’s Claudeforce product, unveiled on August 26, 2026, integrates Anthropic’s Claude as its default AI across Salesforce’s ecosystem, allowing users to complete workflows without ever opening the main app interface. With 37 prebuilt sales skills in its “AIforce” harness, Salesforce reported that the Agentforce platform had surpassed $1.5 billion in annual recurring revenue, growing more than 240 percent year-on-year.

Yet, as these AI harnesses become the engines of departmental productivity, they also inherit the boundaries of traditional business units. Just as no system of record has unified all corporate knowledge, AI agents tend to dominate within domains like sales, HR, or procurement rather than operating seamlessly across the enterprise. This fragmentation persists due to both technical and organizational boundaries, with trust, governance, and auditability locked into each domain-specific system.

Outcome-based AI pricing may also lay the groundwork for platforms to claim a share of the value they create, especially as they move deeper into mission-critical workflows. OpenAI has signaled its intent to pursue licensing and IP-based agreements for applications ranging from scientific research to financial modeling. The further a platform integrates into customers’ processes, the stronger its position to claim a share of any resulting value.

In response, enterprises are increasingly demanding sovereignty over both data and business models, favoring open AI models and on-premises deployments that prevent suppliers from unilaterally repricing or extracting more value. Prosus, for example, has reported inference cost reductions of up to 26 times from running AI workloads on open models. This push is spawning a new class of implementation firms that fine-tune, integrate, and manage AI models within customers’ own infrastructures, a rapidly growing opportunity particularly well-suited to India’s established global capability centers.

With outcome-based pricing, enterprises can now make direct comparisons between AI agents and human workers on the same terms, evaluating productivity by processes completed rather than by effort expended. While the transition is complex and the nature of productivity units continues to evolve, the underlying trend is clear: enterprises are moving to pay for tangible results, and this shift is quickly redrawing the lines of competition and value capture in enterprise AI.