Box Inc. CEO Aaron Levie said the falling cost of frontier AI models could substantially increase the number of tasks companies can economically assign to AI agents. In a post on X on Tuesday, September 22, 2026, Levie highlighted price reductions for Anthropic's Opus 5.5 and OpenAI's GPT-6 Sol and GPT-6 Luna, whose token prices fell 50%. OpenAI introduced the two models with API prices up to 50% below GPT-5.6's promotional pricing, while Anthropic also reduced the cost of its latest model. Levie said the pace of decline in AI's cost per task was unlike that of any other technology and described the resulting expansion in applications as the Jevons paradox applied to agents. He said lower token costs could enable agents to process company data, scan code for security issues, review log data, make decisions and operate in agent swarms across workflows. The developments underscore competition to make AI agents cheaper and more efficient to deploy at scale. Anthropic reported in June that nearly six in 10 survey respondents expected AI to handle more of their work tasks within a year, while more than one-third expected AI to perform most or nearly all of their tasks. Andrej Karpathy, the former Tesla Inc. AI chief and ex-OpenAI researcher, has said AI agents dramatically changed his coding workflow and built Dobby, an assistant for smart-home functions. The content was partially produced with AI tools and reviewed and published by Benzinga editors.