AI Agents & The Need For An Agentic Spectrum

We need to stop being fixated with AI Agents and start considering how to introduce levels of Agency.

4 min readJan 19, 2025

--

Instead of relying solely on fully autonomous AI Agents from the get-go, introducing varying levels of agency/autonomy into everyday applications can provide a more balanced approach to automation.

This allows users to benefit from AI assistance while retaining supervision over critical decisions and actions.

By embedding incremental agency, applications can enhance user experience through adaptive support and intelligent suggestions without overwhelming autonomy.

This approach enables a collaborative environment where human oversight and AI capabilities complement each other seamlessly.

Introduction

I would like to make the case for seeing AI Agents for what they are at this present moment in time, and how different levels of Agency can be introduced.

I think a distinction needs to be made between AI Agents, Agentic X (where the X can be replaced by Workflow, Orchestration, Discovery and more).

Why Considering An Agentic Spectrum Is Important

The hype around AI Agents is undeniable, but their real-world performance falls short. The Claude AI Agent Computer Interface (ACI) achieves only 14% of human-level performance.

In recent benchmarking research, the top-performing AI Agent resolved just 24.0% of tasks, despite being the most expensive model at an average cost of $6.34 per task and requiring 29.17 steps, indicating high computational effort.

The second-best model resolved 11.4% of tasks, scoring 19.0%. It’s the least expensive, costing $0.79 per task, but requires 39.85 steps — the highest among all models.

The third-place model resolved 8.6% of tasks, scoring 16.7%, with moderate costs of $1.29 per task and the fewest steps at 14.55.

AI Agents

An AI Agent is a software program designed to autonomously perform tasks or make decisions using available tools. These agents, as illustrated below, leverage one or more Large Language Models or Foundation Models to decompose complex tasks into manageable sub-tasks.

The sub-tasks are then organised into a sequence of executable actions. The agent is equipped with a set of predefined tools, each accompanied by descriptions that guide when and how to use them in sequence. This structure enables the agent to address challenges effectively and reach a final conclusion.

And even-though, as I have mentioned, AI Agents do not fare well in long horizon tasks.

Human In The Loop

The autonomous nature of Agentic Applications has often been viewed as a threat and a barrier to the implementation of agents.

However, recent applied research demonstrated how human oversight can be effectively integrated as checkpoints before tasks are executed, addressing these concerns.

The human is viewed as a tool in the tool collection of the AI Agent, and when an appropriate next action or tool cannot be found, a human is pinged with a request.

Agentic Workflows / Orchestration / Discovery etc.

Agentic Workflows involve orchestrating tasks where an Agentic (Agency) layer in an application, autonomously handle complex processes through a series of sub-tasks.

Orchestration ensures these sub-tasks are executed in a coordinated manner, optimising efficiency and effectiveness.

Discovery plays a critical role, as the Agentic layer dynamically identify and adapt to new information or tools to enhance performance.

These human supervised workflows leverage AI’s capability to manage intricate sequences without constant human oversight.

By integrating discovery and orchestration, agentic workflows enable continuous improvement and adaptability in automated systems.

The phrases AI Agents, Autonomous Agents, Agentic Application, or what I refer to as Agentic X are all terms which are used interchangeably. Read more here about the differences here.

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.

--

--

Cobus Greyling
Cobus Greyling

Written by Cobus Greyling

I’m passionate about exploring the intersection of AI & language. www.cobusgreyling.com

No responses yet