AI agent

An AI agent combines a large language model or other model with access to tools, such as web browsing, code execution, software applications, databases or payment methods. Given a goal, it breaks the task into steps, calls tools, evaluates results and iterates until it finishes or needs human input. Agents can run continuously and coordinate with other agents.

Agents shift AI from answering questions toward performing work, such as software development, research, customer service and operations. They typically consume far more inference per task than simple chat, which increases demand for computing capacity. Risks include errors that compound across steps, security vulnerabilities such as prompt injection, and unclear liability when agents act on a user’s behalf, including spending money.

For allocators, agent adoption affects software business models, labor productivity and data center demand, and it is the basis for agent payment protocols and stablecoin use cases. Example: a coding agent receives a bug report, reads the codebase, writes and tests a fix, and opens a pull request for human review.