By John Burton
Agentic AI and you
The term “Agent” has become one of the most popular in the field of technology over the last few years, becoming almost as ubiquitous as “GenAI” in certain circles. But what is an AI agent and why has it become one of the biggest focus points in the last few years for companies who are claiming to be at the forefront of innovation? This article aims to explore what an agent is, why they’re useful and what you need to make the most out of them.
What is an agent?
An AI agent is a workflow that uses an LLM to complete a task. The main difference between an AI agent and an LLM output is that an agent has the ability to access tools and take actions on the users behalf. When I ask ChatGPT to give me a recipe for a lovely chocolate cake, it searches through the data it’s been trained on and gives me back a step-by-step instruction to make the cake myself. What it doesn’t do is make me a cake, and that’s the fundamental difference between Generative AI and an agentic solution.
Where’s my cake?
The analogy, while delicious, has an obvious flaw – technology now isn’t able to autonomously create a cake out of a set of disparate ingredients. What it can do though is access software tools and carry out actions based on the prompt I give it. For example, if I want to create an agent to automatically approve my teams leave based on who’s already off, how much leave has been taken in the year and what urgent deliverables are due around that time, it’s a matter of prompting an LLM and giving an agent permission to access the leave request tool on my behalf. I am then able to review the leave requests that have come in and ensure that it’s not exhibiting any unexpected behaviour.
Is this the end of work as we know it?
Whenever a possibility of automation comes up, there’s a natural anxiety that emerges about whether people’s jobs will be no longer required. Whilst most companies want to find a way to maximise the money coming in and minimise any outgoings, the main focus as people have explored this new technology has been primarily how would agents compliment the human expertise that people have built up over the years. Any company that is looking at how best to utilise agents cannot underestimate the need for having a robust and well supported data landscape. Automation without human input and proper guardrails will result in disaster, every time. A worthwhile conversation that will change the way we work is how best to use data, human experience and technology to improve everyone’s lives, both in and out of work. Companies that rush into automation at the expense of human knowledge are always doomed to fail, and this is exactly what we’re seeing now.