What is Agentive AI, Agentic AI and AI Agents?

Contents

• Introduction
• AI Agents
• Agentive AI
• Agentic AI
• AI Assistants & CoPilots & Assistive AI
• Etymology of the words
• Summary

Introduction

Hey, with all these terms flowing around for similar things, it is sometimes hard to pin down the meaning;
AI Agents
Agentic AI
Agentive AI
AI Assistant
Assistive AI
CoPilots

AI agents are software that can autonomously consider the task in hand, and the environment around that task, having agency in making decisions, and taking actions to achieve that task, with minimal human intervention and on behalf of the human.
Unlike traditional software, agent AI can learn, adapt, and operate pro-actively to complete tasks or solve problems in complex, dynamic environments.

In my experience these terms get applied interchangeably. I don't think people know there are subtle differences between them, and frankly, most are not too bothered either!
For those that need to know...

See below how we can progress from from manual task handling => assisted processes => agentive processes, that lay on the fully automated end of the scale. The key here is that the techology is taking input/help from the human until the point at which it is given agency in solving the problem itself. Then the task is automated.

AI Agents

(broadest term)
These are software systems with an AI engine, that is built to complete specific tasks. It does this by breaking the task down into actionable tasks, monitoring progress, and utilising other AI agents or software APIs to complete the task in hand.

In business, they may be used in workflow automation, where an AI Agent manages simple tasks such as routing emails, categorizing documents, or even monitoring inventory levels and generating purchase orders for restocking.

Personally I feel "AI Agents" is the broadest term that spans over the other terms, and can be used in conversation in substitute to all the other definitions below.

Agentive AI

(instructed/directed how to perform task, less agency of its own in how to complete the task)
AI systems that play an active role in executing specific actions as directed by a user or system, focusing on task performance rather than initiating actions on its own. For instance, a chatbot acting on user commands is agentive, as it’s performing a function based on explicit prompts.
Notice agentive AI has less automomy and collaboration in problem solving than the Agentic AI, it receives user focus to move it from task to task to acheive the over all goal.

Agentic AI

(very autonomous, decides itself how to solve problem)
Software system with an AI engine that can act with a large degree of autonomy, taking pro-active steps without human input to accomplish a task, based on observed needs or objectives. They operate with a high degree of initiative aiming to pro-actively support the system users.
Examples would be SMS chat bots having autonomous conversations with customers to resolve customer service issues.
They are systems that show autonomy, acting pro-actively without needing continuous human input. The AI can initiate actions to fulfill specific goals based on its own "decisions" or internal triggers.
Agentic systems are more fire and forget, tell it to go solve this problem and action it.

AI Assistants & CoPilots & Assistive AI - classifications of AI that interact with humans

These AI Assistants and copilots are AI systems that are designed for a human to off load some of the task that the human is completing, handing tasks, or providing extra information, or offering guidance.
They sit side by side the human in the task and the human retains control. They all represent forms of Assistive AI.

Examples are, AI Assistants that correct software code, suggesting improvements, in the editor as someone works. Alexa, Siri and Cortana are all other well known examples.
They are very limited in the range of tasks they can do and are reactive rather than proactive. Compared to higher levels of autonomy found in Agentic AI. They are very much directly supporting the human with discrete micro tasks, but are steered by the human throughout.
CoPilots are a marketing spin on AI Assistants, making them sound more cool and easy to sell subscriptions for. "CoPilot" has now become a widely recognized term, especially with tools like GitHub Copilot.

Should "Agentic" even be a thing etymology of the words

After we have just established the differences, there is a discussion among the linguists out there about the fact that agentic is not even an correct word.
Christopher in the post shown below argues that "Agent" comes from Latin, so the correct suffix should ideally be also latin:

• "-ive" (Latin origin) → "agentive"
• "-ic" (Greek origin) → "agentic"
  for consistency it should be "agentive" over "agentic"

Although he is being a bit playful and pedantic, I agree that I prefer agentive as the correct word over agentic.

Linked in post.

Summary

In essence,
Agentic AI: is autonomous and proactive, possibly coordinating many AI agents to complete complicated tasks.
Agentive AI: is directed and responsive, more limited in task scope single task focused.

Agentic: tends to be back office getting the job done "end to end", in the dark.
Agentive: tends to be up front in the office, with user interface interaction.

Assitive AI helps users complete tasks, where as AI agents do tasks on behalf of the user.

With the proliferation of generative AI, and growth in platforms such as Make.com, Zapier, Automate, N8n that make building AI Agents so easy, 2024 has been the year of AI agents as AI breaks out of AI assistants and grows into AI agents.

This basic overview I hope helped introduce some of the concepts and terminology that you will encounter when getting started with AI. Is is part of a series of posts on AI concepts.