During a recent fireside chat hosted by Scale AI, Boris Cherny, creator of Anthropic’s Claude Code, addressed escalating corporate concerns over artificial intelligence spending, urging enterprises to balance strict return-on-investment expectations with the operational necessity of open experimentation. Moderated by Meta’s Jesse Chen, the discussion followed public scrutiny from industry executives regarding whether rising token consumption justifies accelerating model costs. Cherny affirmed that ROI accountability remains essential, but cautioned against allowing budget anxieties to prematurely restrict AI adoption at the employee level. Rather than imposing front-end restrictions on token usage, Cherny advocated for granting teams adequate token allowances paired with institutional permission to experiment without fear of financial penalty. He emphasized that backend cost management, including Anthropic’s per-seat budget controls, should handle expenditure limits once high-value internal use cases are identified. This approach, he argued, prevents organizations from missing unconventional innovations, noting that transformative AI applications frequently originate from non-technical staff rather than centralized engineering or product teams. The conversation reflects a wider industry reckoning. OpenAI CEO Sam Altman and other AI leaders have increasingly acknowledged enterprise concerns over AI capital allocation. From Anthropic’s perspective, token economics introduce inherent opportunity costs, as internal usage directly reduces available capacity for external customers. Consequently, Cherny framed internal experimentation itself through an ROI lens, aligning tool adoption with measurable business value and sustainable model economics. As generative models accelerate, Cherny noted that traditional AI productivity metrics are becoming obsolete. Early evaluation frameworks relied heavily on the percentage of code generated by AI, a benchmark that loses relevance once systems automate complete development workflows. Moving forward, he suggested enterprises track engineering velocity acceleration and identify subsequent operational bottlenecks. When code production reaches near-total automation, the primary constraint shifts to ideation capacity. Organizations that remove friction from concept generation and foster cross-departmental AI literacy will capture the highest returns on their model investments. The guidance underscores a strategic pivot for corporate AI strategy: treat token allocation as an exploratory capital expense rather than a fixed utility cost, while implementing scalable governance to optimize long-term value extraction and sustain competitive advantage.