Sutton and Barto : "The learner and decision maker is called the agent. The thing it interacts with, comprising everything outside the agent, is called the environment." These interact continually. The agent receives some representation of the environment's state and on that basis selects an action.

The environment responds to the agent's actions and presents a new state to the agent. The environment also gives rise to rewards, special numerical values that the agent seeks to maximize over time through its choice of actions. In computer use, this includes operating systems, web browsers, applications, and interfaces.

Sutton and Barto : "Observations — signals that depend on its [environment's] state but, like a robot's sensors, provide only partial information about it." In computer use, these might be screenshots, DOM trees, or accessibility information.

Execution environments allow agents to dynamically interact and receive real-time feedback (e.g., live websites, VMs). Non-execution environments evaluate against pre-recorded traces or static snapshots without dynamic interaction.

The environment includes everything an agent can interact with through distinct interfaces and tools. For example: using the browser interface to access Google Maps, the terminal interface to run git commands, or the chat interface to clarify user intent.

The physical or virtual machine where the agent is deployed and operates (e.g., a developer's laptop). The local workspace is the isolated environment specific to the agent within the host (e.g., a containerized Docker environment).