Quick answer: Common types include reactive, model-based, goal-based, utility-based and learning agents, plus multi-agent systems. The right choice depends on the goal, context, actions, risk and oversight — not on selecting the most sophisticated label.
Reactive and model-based agents
Reactive agents respond to the current state using rules, making them predictable in simple environments. Model-based agents maintain a representation of context so they can track stages and previous information.
Goal-based and utility-based agents
Goal-based agents choose steps that move a process towards an outcome. Utility-based agents compare alternatives using criteria such as quality, time, cost and risk.
Poor objectives or scoring criteria produce poor decisions. Reaching a goal never justifies exceeding permissions.
Learning agents
Learning agents use data or feedback to improve part of their performance. In business, learning should not mean freely rewriting production rules after every conversation. Important changes need validation.
Multi-agent systems
These divide responsibilities among specialised agents: one coordinates, another retrieves information and another prepares an action. Modularity can improve, but cost, latency and troubleshooting complexity increase.
How to choose
Evaluate the goal, variability, memory, tools, impact of error, audit needs and maintenance capability. Start with the least autonomy that creates value, moving from recommendations to reversible actions and only then to low-risk automation.
Frequently asked questions
Which type is the most intelligent?
There is no universal winner. The best agent is the one suited to the outcome and required level of control.
What is a multi-agent system?
It is an architecture where specialised agents collaborate and a workflow coordinates their responsibilities.
Must I choose a category before development?
Define the process, data and risk first; the appropriate architecture follows from those requirements.