External Interactions
External Interactions refers to the mechanisms, protocols, and interfaces through which an ai-agent or system communicates with, perceives, and acts upon its environment outside its immediate computational context. This includes API calls, tool usage, data ingestion, and human-in-the-loop feedback loops.
Core Components
- Perception: Ingesting external data streams (text, code, sensor data) to update internal state.
- Action: Executing commands via Tools or APIs to modify the external environment.
- Feedback Loops: Mechanisms for receiving results from actions to inform subsequent planning steps.
Agentic AI Architecture Context
Recent frameworks, such as those defined by IBM, emphasize specific terminologies to structure how agents manage these interactions:
- Planning: The agent’s ability to decompose complex goals into sequential steps involving external interactions.
- Tool Use: The specific invocation of external functions or services to perform tasks the model cannot execute internally.
- Memory: Storing interaction history to maintain context across multiple external exchanges.
- Control: Managing the flow of interactions to prevent infinite loops or unsafe external modifications.
- Evaluation: Assessing the outcome of external actions against the original goal.
See IBM Defines Five Key Terms for Agentic AI Architecture for a detailed breakdown of these five key terms as presented by Martin Keen.