AI Agents spectrum¶

AI Agents: Current Landscape, Categorisation, and Strategic Deployment

Key themes and important facts regarding AI agents, drawing insights from "The 7 Types of AI Agents." It explores the definition, rapid adoption, various categorisation frameworks, and critical considerations for successful deployment within organisations.

Topics¶

AI Agents spectrum

1. The Agentic Era: A Fundamental Shift in AI Capability¶

The core premise is that AI is entering an "agentic era," where AI systems move beyond mere productivity tools to fundamentally transform the type and amount of work an organisation can accomplish. This shift is characterised by AIs that "do things for me" rather than simply being "AI that I use to do things."

Key Takeaway: Agents represent a higher level of AI capability, shifting from assistive tools to autonomous performers.

2. Rapid Enterprise Adoption: Agents Are Here Now¶

The adoption of AI agents by enterprises is not a future concept but a current reality. Statistical data highlights a significant increase in agent deployments and pilots.

"The percentage of organisations that had at least some agents fully deployed i.e out of and past the pilot stage and tripled from 11% to 33% between Q1 and Q2."
"This followed a jump in pilots from 37 to 65% between Q4 and Q1."
"In net 90% of the organisations that KPMG surveyed said they were past AI agent experimentation and actively into either pilots or deployments."

Key Takeaway: Businesses are rapidly moving beyond AI agent experimentation into active pilots and full deployments, indicating their immediate relevance and impact.

3. Categorisation of AI Agents: Functionality vs. Focus¶

Understanding the diverse nature of AI agents is crucial for strategic deployment. Two primary ways to categorise agents are by their functionality (how they operate) and their focus (the business outcome they aim to achieve).

3.1. Functionality-Based Categorisation (How they operate):¶

This framework describes the underlying mechanisms and complexity of AI agents. Examples include:

Simple Reflex Agents: Operate based on "predefined rules and the immediate data it has access to." They are suitable for "very simple tasks that don't require a ton of training" (e.g., password resets based on keywords, email autoresponders).
Model-Based Reflex Agents: More complex, they "track how the environment evolves allowing the agent to infer unobserved aspects of the current state." They use a "world model to make better decisions" (e.g., network monitoring for anomaly detection).
Goal-Based Agents (Rule-Based Agents): Can "plan sequences of actions to achieve their desired outcomes" by defining a goal state and planning mechanisms (e.g., inventory management systems planning reorder schedules).
Learning Agents: Capable of "improving behavior over time by learning from previous experiences," rather than relying purely on pre-programmed knowledge (e.g., advanced customer service chatbots that improve through interactions).
Utility-Based Agents: Differ from goal-based agents by "explor[ing] and handl[ing] trade-offs between competing goals" (e.g., flight ticket search agents balancing travel time and price).
Hierarchical Agents: Higher-level agents "deconstruct complex tasks into smaller ones and assigns them to lower level agents," allowing for the execution of more complex tasks through coordinated sub-agents.
Multi-Agent Systems: Refer more broadly to "combinations of these various agents which can achieve more complex goals."

Key Takeaway: Functional categories provide insight into the technical capabilities and operational models of different agents, essential for understanding what's "under the hood."

3.2. Focus-Based Categorisation (Business outcome):¶

This framework, derived from "The Information" article, organises agents based on their intended business output and how they are currently being deployed. The seven categories include:

Business Task Agents: Automate "fairly simple but repetitive and common use cases like data entry, document classification, invoice processing."
Conversational Agents: Encompass "external facing customer service as well as internal facing support around IT or HR questions."
Research Agents: "Can go do research," particularly important for general employees as one of the first impactful agentic experiences.
Analytics Agents: "Can analyze structured data to produce graphics, charts or reports."
Developer Agents: A "major major theme," identified as "perhaps the single most significant breakout agent so far."
Domain Specific Agents (Vertical Agents): "Specialized agents that have very specific domain knowledge in an area like legal, healthcare or finance."

Key Takeaway: Focus-based categories are pragmatic for businesses, directly linking agent deployment to desired outcomes and current real-world applications.

3.3. The KPMG TACO Framework: A Simplified Approach¶

KPMG offers a simplified "TACO framework" which organises agents into four categories based on complexity, human involvement, and system breadth:

Taskers: "Execute well-defined individual tasks and require a human in the loop."
Automators: "Manage more complex tasks that span multi-system workflows."
Collaborators: "Adaptive AI teammates that manage multi-dimensional goals."
Orchestrators: "Transformative agentic systems that coordinate multiple agents and tools to manage interdependent workflows."

Key Takeaway: The TACO framework provides a more intuitive understanding for non-technical audiences, simplifying the functional breakdown for broader accessibility.

4. The Emphasis on Orchestration and Multi-Agent Systems¶

A significant trend, especially evident from discussions with enterprises and private equity firms, is the strong emphasis on "orchestrators and multi-agent systems." Microsoft's Build conference further highlighted this by focusing on "software and agent infrastructure" and "multi-agent orchestration."

"It is quite clear if you're spending any time with enterprises or private equity firms that there is a huge amount of discussion of orchestrators and multi-agent systems."
"Microsoft really put the emphasis on software and agent infrastructure one of their big announcements was multi-agent orchestration in C-pilot Studio that was designed to allow people to deploy more comprehensive and complex agentic systems where the agents could actually interact with one another."

The rationale is that to achieve the "full value of agents," they "are going to have to work together in comprehensive systems." Thinking in "systems terms" is "more directionally aligned with where the world is heading."

Key Takeaway: While single "spot agents" are valuable, the future of AI agents lies in comprehensive, coordinated "digital worker organizations" powered by multi-agent systems and robust orchestration.

5. Strategic Deployment: Beyond Use Cases to Infrastructure¶

Successful agent deployment requires more than just identifying use cases; it necessitates building out the necessary infrastructure and tech stack. This includes:

Model training and fine-tuning - LLM and AI application development - Monitoring and observability - Inference optimization - Model hosting - Model evaluation - Data processing and feature engineering - Vector databases - Synthetic data and data augmentation - Coding assistance - DevOps and MLOps - Product and design

Key Takeaway: Organizations must consider the comprehensive "infrastructure that needs to be built as well" when planning for agent readiness, including common necessities like inference optimization, monitoring, and evaluation.

Conclusion¶

AI agents are a transformative force, rapidly moving from experimentation to widespread enterprise deployment. Understanding their diverse types, whether by functionality or focus, is crucial. The clear trend towards multi-agent systems and orchestration underscores the need for comprehensive infrastructure planning. By anchoring strategic thinking and systems design to this "agent systems future," organisations can unlock the full value of AI and navigate the complexities of this evolving landscape.