Learn about the terminology used for describing Agents.

Agentic AI

AI systems designed to autonomously perceive their environment, create plans, make decisions, and execute tasks to achieve specific goals without constant human intervention.

A discrete software entity powered by an AI model that can act independently to accomplish a defined objective or workflow.

A framework where multiple AI agents interact, collaborate, or negotiate with each other to solve complex, multi-layered business problems.

The capability of an AI agent to recognize when it needs external help and autonomously call APIs, run code, or search databases to complete a task.

A design approach where an agent is given a high-level objective (the "what") rather than strict, step-by-step instructions (the "how").

The continuous cycle of perception, planning, action, and evaluation that an AI agent runs through until its goal is accomplished.

The ability of an agent to evaluate its own intermediate outputs, identify errors or inefficiencies, and adjust its strategy mid-process.

Systems that give agents both short-term memory (context retention during a task) and long-term memory (saving past experiences using vector databases).

Strict programmatic boundaries, safety policies, and operational constraints set by a business to limit an agent's authority and prevent harmful actions.

A governance model where an AI agent pauses and waits for explicit human approval or intervention before executing high-risk decisions.

The underlying software that manages communication, task routing, and data sharing between different AI agents and corporate legacy systems.

The process where an AI agent breaks down a massive, complex corporate objective into smaller, sequential, and manageable sub-tasks.

A scale defining how much freedom an agent has, ranging from complete human dependence (low autonomy) to fully independent execution (high autonomy).

The contrast between traditional software that follows rigid "if-then" code (deterministic) and agentic software that dynamically calculates its own path.

A security vulnerability where malicious inputs manipulate an agent's instructions, causing it to misuse its connected tools or leak data.

The strategic allocation of computational resources and costs for an agent, as multi-step autonomous reasoning loops consume high amounts of tokens.

The traditional automation of linear tasks, which is increasingly being replaced by agentic AI to handle non-linear, unpredictable variables.

An iterative process where an AI model generates a draft, critiques its own work, uses tools, and refines the output over multiple cycles.

The specific design framework (like ReAct or Plan-and-Solve) that determines how an agent thinks, reasons, and sequences its actions.

The chain reaction of events or changes caused in corporate systems when an autonomous agent executes an action.

The process of dynamically directing an incoming business request to the most qualified AI agent based on the meaning and context of the text.

An operational risk where a small mistake made by an AI agent in an early step multiplies and causes larger failures down the line.

The foundational, hard-coded guidelines that define an agent's identity, tone, capabilities, permissions, and core mission.

The strategy or logic an autonomous agent uses when it encounters an unusual scenario that was not covered in its standard training.

The return on investment calculated by measuring the efficiency gained when an AI handles end-to-end multi-step workflows instead of single tasks.