An AI trading agent is a system that observes market conditions, decides what to do, and executes trades — repeatedly, over time, in pursuit of an objective a person has given it. Agentic trading is the approach; the agent is the thing that does it.
The three things an agent does
Perceives: reads prices, the order book, its own position, and whatever else its model uses. Decides: evaluates the state against its mandate and chooses an action — trade, adjust, wait. Acts: places or modifies orders and observes the result.
Then it does it again. The loop is what makes it an agent rather than a one-shot predictor.
Agent versus bot
A bot executes a fixed rule. An agent evaluates conditions and chooses. The practical difference: a bot told to buy every 2% dip will buy through a crash. An agent told to accumulate a position at good prices can conclude that the current prices aren’t good and wait.
Agent versus chatbot
A chatbot generates text. Asked about a trade, it produces an opinion. It doesn’t observe the market in real time, doesn’t hold a position, and doesn’t execute. Many products marketed as trading agents are chatbots with a brokerage link, and the reasoning quality of a language model is not the same as the predictive quality of a trained trading model.
Boundaries
An agent operates inside a mandate — what to trade, how much, over what window, with what hard limits. It’s autonomous within those walls and inert outside them. The person sets the walls; the agent works inside them; neither does the other’s job.