Token-maxing is an AI cost sink – how to use agents without busting your budget

Steve Lucas, CEO of Boomi, warns that IT professionals are rushing into agentic AI without a clear strategy, leading to unsustainable token usage. Tokens are the fundamental units processed by AI models, and agentic AI consumes orders of magnitude more tokens. Lucas reported that Boomi's spending on Anthropic's Claude increased tenfold year over year, which is unsustainable. Snowflake CEO Sridhar Ramaswamy acknowledges rising inference costs but emphasizes the importance of allowing staff to explore agents for innovation. Matt Luizzi of Whoop supports giving employees room to take risks with agents while maintaining guardrails and monitoring. Sriram Sitaraman of Synopsys argues that innovation cannot happen with strict constraints, but setting clear context for agent tasks can limit token consumption and increase value. Francois-Xavier Pierrel of TF1 compares AI addiction to sugar addiction and advises using smaller models for narrow use cases to control costs. All leaders agree that tokenomics – measuring, pricing, and managing token consumption – is essential for sustainable AI deployment.