The Definitive Guide To Understanding Agency and the Role of AI Agents
In this article, I explore the evolving terminology in AI, clarifying key terms that are often used interchangeably. I provide a detailed breakdown of their meanings and implications at a technical level, helping to demystify the language surrounding AI and its applications.
Popular Terms
The most popular terms used currently are Autonomous AI, AI Agents, Agents & Agentic AI.
I would like to introduce the term Agentic “X”. Where the X can be replaced with Workflows, Orchestration and more.
Agentic Agents
Stop speaking about Agentic Agents. The phrase agentic agents is redundant and doesn’t make much sense.
The term agentic already implies having agency, which is the capacity to act independently and make choices.
Therefore, calling agents agentic is repetitive because agents, by definition, possess agency. A more appropriate term would be simply AI Agents or specifying the level of agency, such as high-agency agents or autonomous agents, to convey the intended meaning more clearly.
I prefer referring to AI Agents rather than just agents because it clearly identifies them as software-based entities.
Using AI Agents helps avoid confusion with roles like customer service representatives, ensuring the term is understood in the context of artificial intelligence.
Autonomy vs Agency
Is there a difference?
Yes, in AI, there is a difference between autonomy and agency.
Autonomy refers to an AI system’s ability to operate independently without human intervention, making decisions and taking actions based on its programming and learning.
Agency, on the other hand, encompasses the capacity of the AI to not only act independently but also to make context-aware decisions, understanding and responding to its environment in a way that aligns with specific goals or objectives.
While autonomy is about independence, agency involves a deeper layer of purposeful action and decision-making.
Autonomous AI
Autonomous AI refers to systems that can perform tasks and make decisions independently, without continuous human intervention or oversight.
These systems are designed to handle complex, real-world scenarios by adapting to changing environments, learning from experiences, and optimising their actions over time.
Two examples which helped be create a distinction between Autonomous AI and AI Agents are…
Self-Driving Cars
Autonomous AI systems in self-driving vehicles can navigate roads, make real-time decisions and adapt to driving conditions without human intervention.
Autonomous Drones
AI-powered drones can perform tasks like surveying, delivery, surveillance and more, by using onboard sensors and AI algorithms to make decisions about navigation, obstacle avoidance and mission execution without requiring human control.
AI Agents vs Autonous AI
In contrast, AI Agents typically operate within more structured frameworks, often relying on predefined instructions or prompts to carry out specific tasks, usually within a limited scope.
While AI Agents may have autonomy in certain actions, they still rely on human guidance or input for broader decision-making processes.
Autonomous AI, on the other hand, pushes further by operating fully independently, capable of initiating tasks and adjusting strategies without human input.
What is Agency?
In the context of AI, agency refers to the capacity of an AI system to act independently, make decisions, and perform tasks based on its environment and objectives.
It denotes the AI’s ability to execute actions with a level autonomy/agency while considering the context of its operation.
The term agent is used interchangeably with agency, as an AI agent embodies this capacity to act with a level of autonomy and decision-making.
Essentially, an AI Agent is a system that possesses agency, making them functionally the same in AI terminology.
Both terms emphasise the agentic operational nature (with agency) of the AI system in interacting with its environment and carrying out tasks.
AI Agents (Agents)
AI Agents 101: The Basics
An AI Agent a software entity, is a software system primarily powered by one or more language models.
To enable visual capabilities, the agent requires a multi-modal language model or foundation model with integrated vision features.
Task Decomposition
At its core, an AI Agent usually relies on unstructured conversational input, processing unstructured data from user queries.
The output is typically in natural language, utilising the Natural Language Generation (NLG) capabilities of language models.
For example, you could ask an AI Agent, What is the square root of the year of birth of the man commonly regarded as the father of the iPhone?
This complex query, difficult for traditional conversational UIs, is manageable for an AI Agent.
Way of Work
The AI Agent begins by breaking down the compound question into smaller, manageable sub-steps, solving each step sequentially.
These steps are treated as individual actions. The agent uses its language model to decide the next action to take.
After each action, it observes the result and formulates a thought. If the final answer isn’t reached, the AI Agent loops back, selecting another action to progress toward the final answer.
Agents are capable of breaking down complex tasks into smaller steps, making decisions, and iterating until a goal is achieved. They can also integrate various capabilities like vision or speech, depending on the foundation models they use.
Read more about AI Agents here:
Agentic AI / Agentic X
Agentic AI is conceptually distinct from AI Agents in that it does not necessarily refer to a standalone software entity.
Instead, Agentic AI refers to the integration of varying levels of agency into everyday applications, enabling them to take agency to perform tasks autonomously or semi-autonomously (with human intervention).
This agency can be introduced through mechanisms like Agentic Workflows and Agentic Orchestration, where processes are autonomously managed and optimised.
The result is what I refer to as Agentic X, where different applications, each with varying degrees of agency, can operate with increased autonomy.
In this framework, agency is not confined to a singular AI Agent, but distributed across multiple systems, enriching applications with more dynamic and context-aware decision-making capabilities.
Agentic AI extends beyond the traditional idea of an AI Agent (which is typically a discrete system performing specific tasks) to embed agency in a wider range of applications and workflows.
Rather than having a single autonomous entity (like an AI Agent), the agency can be distributed, allowing different applications to function with varying degrees of independence based on the task at hand.
For example, an Agentic Workflow might involve automating complex business processes where decisions are made based on real-time data, while Agentic Orchestration could involve coordinating actions across multiple systems, with different components operating at different levels of autonomy.
Read more about the agentic spectrum here:
Agentic X
The concept of Agentic X allows for a flexible and scalable approach to integrating autonomy across diverse platforms, giving applications the ability to make decisions, adapt, and learn with minimal human intervention.
This could ultimately transform how businesses and users interact with AI by allowing systems to function more intelligently and seamlessly across a range of tasks and domains.
Read more here about the notion of Agentic X:
The Future
I believe in the future, there will be less focus on AI Agents as an entity and ultimate solution but rather focus on Agency and different levels of Agency being imbedded into everyday applications.
This Agency will be transparent and deeply integrated into everyday software, enabling more autonomous and adaptive behaviour across a variety of tasks.
As applications become more context-aware and capable of making decisions on their own, they will handle increasingly complex workflows with minimal human input.
This shift will blur the line between traditional software and intelligent systems, where agency is not confined to isolated entities but is distributed across interconnected components.
With advancements in AI, applications will be able to orchestrate and execute tasks with a higher degree of flexibility and precision, ultimately streamlining operations and enhancing user experiences.
In this future, users will interact with applications that anticipate needs, make informed decisions, and continuously improve their performance without needing constant oversight.
Chief Evangelist @ Kore.ai | I’m passionate about exploring the intersection of AI and language. From Language Models, AI Agents to Agentic Applications, Development Frameworks & Data-Centric Productivity Tools, I share insights and ideas on how these technologies are shaping the future.
