Agentic AI is a class of artificial intelligence systems that work toward a goal by planning a series of steps and carrying them out with limited human direction. An agent is usually built around a large language model that can call outside tools: it searches for data, runs code, reads the result, and decides what to do next until the task is finished or it is stopped.
In finance, agents are used to gather research, summarize filings, write and test strategy code, and monitor positions. Letting an agent send orders is a larger step than classic algorithmic trading, where the rules are fixed in advance and can be tested. An agent may act differently on the same input, can misread data, and can repeat an error quickly. Firms therefore keep agents behind the same pre-trade limits, logs and risk management system checks as any other automated order source, and most keep a person responsible for final approval.