Based on the definition of the International Organization for Standardization,5 an AI agent can be broadly defined as an entity that senses percepts (sound, text, image, pressure etc.) using sensors and responds (using effectors) to its environment. AI agents generally have the autonomy (defined as the ability to operate independently and make decisions without constant human intervention) and authority (defined as the granted permissions and access rights to perform specific actions within defined boundaries) to take actions to achieve a set of specified goals, thereby modifying their environment. The core components of an AI agent FIGURE 1: Figure 1 highlights how an agent is made up of several core components, including: –User input: the external (e.g. human, another agent) input that the AI agent receives. This could be instructions such as typing via a chat- based interface, voice-based commands or pre-recorded data. –Environment: the bounds in which the AI agent operates. It serves as the area in which the agent applies its sensors and effectors to percept and modify its surroundings based on the inputs received and the actions decided upon by the control centre. The environment can be physical infrastructure such as the mapped area of an autonomous vehicle or digital infrastructure such as the intranet of a business for a coding agent. –Sensors: mechanisms through which the agent perceives its environment. Sensors can range from physical devices (e.g. cameras or microphones) to digital ones (e.g. queries to databases or web services). –Control centre: typically makes up the core of the AI agent along with the model, such as an LLM. The control centre helps process information, make decisions and plan actions. Based on the capabilities of the AI agent, the control centre involves complex algorithms and models that allow the agent to evaluate different options and choose the best course of action. –Percepts: the data inputs that the AI agent receives about its environment, which could come from various sensors or other data sources. They represent the agents’ perception or understanding of its environment. –Effectors: the tools an agent uses to take actions upon its environment. In physical environments, effectors might include robotic arms or wheels, while in the digital environment, they could be commands sent to other software systems, such as generating a data visualization or executing a workflow. –Actions: represent the alterations made by effectors. In physical environments, actions might be pushing an object, whereas in digital environments they could be linked to updating a database.AI agent Percepts Environment ActionsSensors Control centre EffectorsDigital infrastructureUser input Physical infrastructure Source: World Economic Forum Navigating the AI Frontier: A Primer on the Evolution and Impact of AI Agents 7

Navigating the AI Frontier 2024