IT services companies are moving beyond the initial wave of token maxxing as enterprises scale AI deployments. According to The Times of India, tokens-the units consumed whenever large language models process information-have become the fundamental currency of AI, but after initial experimentation, IT firms are increasingly warning that relentless token consumption without measurable business outcomes could become the industry's next cost problem. The term gained currency in Silicon Valley, where the double 'xx' in 'maxxing' denotes aggressive optimization, but as enterprises move from pilots to large-scale deployments, companies are increasingly focused on linking token usage to business value rather than raw consumption.
Industry executives are emphasizing the shift from input metrics to outcome-based evaluation. Arumugam Kumaradassan, vice-president and head of AI industrialisation and enterprise IT automation at Cognizant, stated that 'tokens are an input to delivery, not a measure of value, and token consumption is simply a cost signal, tracked for discipline, licence governance and capacity planning'. At Happiest Minds, executive vice-chairman Joseph Anantharaju noted that the company is developing capabilities for token metering and optimization as enterprises scale agentic AI deployments, while also evaluating outcome-based commercial models that combine software, agents, platforms and AI consumption. The debate is becoming increasingly relevant as AI projects move from pilots to production, with Salesforce chief digital officer Vala Afshar noting that 'one of the most important currencies in an agentic AI transformation are tokens'.
Technology providers are developing new pricing structures that align with business value creation. Mphasis CEO Nitin Rakesh explained that customers increasingly bear variable token costs as usage expands, stating that 'what we are pricing is the economic outcome. There is a base price that you (client) will pay me, and the rest will be linked to the outcome I can drive'. This shift represents a fundamental change from the traditional cost-control approach to outcome-based evaluation that focuses on reduced defect rates, fewer production incidents, faster release cycles with higher confidence, and compliance coverage that holds up under audit. As reported by NDTV Profit, the focus has moved from 'what did we spend on AI?' to 'what did we get for it?' with FinOps tools unable to answer whether AI output was correct.
Enterprises are building comprehensive verification infrastructure alongside generation capabilities. According to KushoAI, this shift involves systems that verify AI output rather than just generating it, with the developer community recognizing this need through benchmarks like HumanEval, SWE-bench, and APIEval-20. The evaluation layer is now as important as the generation layer, with the same verification approach extending to enterprise workflows where teams use AI for code generation, test writing, and automation while implementing verification passes to ensure generated tests catch real bugs, APIs behave according to specification, and agent changes don't break downstream systems. Companies like Cognizant have implemented metering and value-linkage capabilities that connect token consumption to business workflows and outcomes, as noted by Kumaradassan who emphasized that 'as the industry moves towards outcome-based models, token spending will reveal the cost of achieving those outcomes'.
The next phase of enterprise AI will be defined by organizations that build comprehensive verification infrastructure rather than just generation capabilities. According to KushoAI, every proxy metric eventually gets replaced by the thing it was trying to measure, with story points giving way to shipped products, compute utilization giving way to business outcomes, and tokens giving way to correctness. The author emphasizes that the end game is correctness, not just AI usage, with serious organizations building the infrastructure to verify what AI produces rather than just generating it. This pattern repeats itself every time a new technology enters the enterprise - teams rush to use tools, leadership celebrates rollouts, and success gets measured by activity, but the gap between activity and outcome becomes visible only when it causes real damage. The shift represents a fundamental move from 'what did we spend on AI?' to 'what did we get for it?' with outcome-based pricing models becoming the new standard for AI deployment and consumption.
