Chapter 1: What is Agentic AI?
Agentic AI refers to artificial intelligence systems that can pursue goals autonomously — making decisions, taking actions, and adapting their approach based on outcomes. Unlike traditional AI that simply responds to prompts, agentic systems can plan, execute multi-step tasks, use tools, and operate with varying degrees of independence.
The term "agent" comes from the concept of agency — the capacity to act independently in the world. An AI agent isn't just answering questions; it's doing things on your behalf.
A Simple Distinction
- Traditional AI: "Here's an answer to your question"
- Agentic AI: "I'll figure out what needs to be done, do it, check if it worked, and adjust if necessary"
Why Now?
Large language models (LLMs) have reached a capability threshold where they can reliably reason about problems, break down complex tasks, decide which tools to use, and recover from errors. This makes truly autonomous behaviour possible for the first time at scale.
Several converging factors have made agentic AI practical:
- Model capability: Modern LLMs can follow complex instructions, reason through multi-step problems, and generate appropriate tool calls
- Tool integration: Standardised approaches for connecting AI to external systems, APIs, and data sources
- Infrastructure maturity: Platforms and frameworks that handle the orchestration complexity
- Cost reduction: Declining inference costs make autonomous operation economically viable
What Agents Can Do
Agentic systems can:
- Plan: Break down complex goals into achievable steps
- Execute: Take actions using tools and integrations
- Observe: Monitor the results of their actions
- Adapt: Adjust their approach based on outcomes
- Learn: Improve over time through experience (in some implementations)
- Collaborate: Work with other agents and with humans
What Agents Are Not
It's equally important to understand what agentic AI isn't:
- Not magic: Agents are constrained by their training, instructions, and available tools
- Not infallible: They make mistakes, especially on novel or edge-case situations
- Not truly autonomous: They operate within boundaries set by humans and require oversight
- Not a replacement for judgment: Complex, high-stakes decisions still need human involvement
The power of agentic AI lies not in replacing human intelligence, but in handling routine, repeatable tasks at scale — freeing humans to focus on work that requires creativity, judgment, and relationship-building.
The Spectrum of AI Capabilities
Agentic AI sits on a spectrum of AI capabilities:
| Type | Description | Example |
|---|---|---|
| Generative AI | Creates content in response to prompts | ChatGPT answering questions |
| Assistive AI | Helps humans complete tasks faster | Copilot suggesting code |
| Agentic AI | Pursues goals autonomously within boundaries | AI that manages your inbox |
| Autonomous AI | Operates with minimal human oversight | Self-driving vehicles |
Most practical business applications today fall in the assistive-to-agentic range. Fully autonomous systems remain rare and are typically limited to well-defined, bounded domains.
The Pragmatix AI Capability Model
To understand where agentic AI fits in the broader AI landscape, we developed the Pragmatix AI Capability Model (pX-AICM). It organises AI into four domains:
- Perceiving — How AI takes in the world (computer vision, NLP, sensors)
- Thinking — How AI processes, split into Reasoning (symbolic) and Machine Learning (statistical)
- Acting — How AI produces outputs and takes action (generative AI, automation, agentic AI)
- Operating — How AI runs in production (platforms, governance, security)
Agentic AI sits within the Acting domain — AI that doesn't just generate content but pursues goals, uses tools, and takes action.
An agent draws on all four domains: it perceives through NLP and vision, thinks through reasoning and ML, acts autonomously, and must be operated safely at scale.
→ See Addendum: Pragmatix AI Capability Model (pX-AICM)
Adapted from BCG Robotaxonomy (2023), extended to include operational governance.
Looking Ahead
In the next chapter, we'll explore the autonomy spectrum in more detail, introducing the Pragmatix Agentic AI Maturity Model — a framework for assessing where your organisation is today and where you might want to go.