AI agents use nearly 5 times more tokens than humans, and their consumption has grown approximately 14 times since February 2026. This is according to an analytical report from the crypto division of Andreessen Horowitz (a16z), based on data from OpenAI, OpenRouter, and Similarweb. Despite the early stage of the technology, autonomous systems are already significantly changing the nature of computational resource consumption.
According to OpenAI statistics, token generation volume at a typical company has roughly doubled, but the most active AI users are scaling it up much faster. The consumption gap between the most active enterprises and typical companies is approximately 8 times. In the technology sector, the gap is even wider: companies in the top decile generate nearly 12 times more tokens than average users, and their generation volume has grown 32.5 times compared to figures from a year ago. Active companies more frequently use advanced tools: their plugin usage is approximately 2 times higher, and skills usage is 6 times higher. Codex adoption among legal professionals has grown 108 times since February 2026.
According to OpenRouter data, more than 85% of tokens consumed by AI agents come from cached prompts. Unlike ordinary chatbots, agents repeatedly cycle through reading, writing, and task execution, preserving context between operations. Cached tokens are significantly cheaper than initial loading, which improves the economics of agent usage but requires larger memory volumes. This sustains high demand for high-bandwidth memory (HBM) used in modern AI infrastructure.
The growing popularity of agents is also affecting the traditional automation market. According to Similarweb, traffic to the websites of platforms N8N, Zapier, and Make has been declining by more than 10% over the past 12 weeks. At the same time, Gumloop, positioned as a native platform for building AI agents, is showing growth. Analysts note that traditional platforms are also actively integrating AI, so it is too early to speak of their decline, but autonomous agents are already beginning to reshape the market structure.
This trend aligns with the forecast of Meta CEO Mark Zuckerberg, who expects billions of personal AI agents to emerge within the next 5 years. Animoca Brands co-founder Yat Siu predicts the formation of an economy in which up to 100 billion autonomous AI systems will interact with blockchains and conduct digital payments. Meanwhile, researchers from UC Riverside, Microsoft, and Nvidia previously identified risks of autonomous behavior: during tests, agents performed undesirable or potentially harmful actions in 80% of scenarios.