What are AI agents and where do they fit within AI today?

Natalie Smithson
AI enthusiast | Tea addict | Focused on using AI assistants to win the working week
Illustration of an AI assistant on a phone with an audio icon sits next to an AI assistant and a speech bubble

If you’re ready to catch the next wave of AI automation and sail ahead of your competition, read on.

Understand how and why AI agents are being built to automate business processes, learn how they work and how they differ from AI assistants, and how you get started with introducing them successfully into your organisation for the most impactful result.

Let’s take a dive deep into AI agents and the future of work.

TL;DR

  • AI agents disrupt today’s workforce in ways most people aren’t yet ready for or able to fully comprehend
  • AI agents carry out a limitless range of business processes while an AI assistant orchestrates what they do as the glue holding everything together
  • Creating a capable AI agent demands a complex arrangement of technology and the support of experienced technologists, language experts, psychologists, writers and designers who know how to apply human reasoning
  • AI agents can be trained on highly detailed and specific company information using large language models (LLM) and retrieval augmented generation (RAG)
  • We’ll all need to adapt to the displacement of some traditional roles that can now be automated with AI agents, take on new roles working alongside AI agents and assistants, and evolve with technological change

What are AI agents?

AI agents are the next step towards AI being ingrained in everything we do in business. Able to take on entire business processes accurately and at speed, AI agents relieve the need for teams to do any repetitive tasks and, in particular, perform functions that couldn’t easily be automated before.

AI agents disrupt today’s workforce in ways most people aren’t yet ready for or able to fully comprehend, and there’s an urgent need to address how we make this shift successfully to get people and AI working closely together.

Once more functions are handled completely by AI, business leaders and teams operating within this new structure can focus more on what pushes progress and makes money for increased development and growth.

Company AI

In the long term, AI won’t be defined by whether it’s a voice bot or online assistant. It won’t be categorised by the channel it appears on, be it your website or app, social media, SMS or phone lines. It’ll simply be your company’s AI, available 24/7 to take on tasks and functions you no longer need to do manually.

Within that AI system:

  • AI agents do the work
  • AI assistants assign, order, and communicate the work that’s being done
  • Your teams will both instruct and work alongside every AI agent and assistant

AI agents vs. AI assistants: What’s the difference?

The AI evolution has seen basic bots following a limited list of rules turn into smarter AI chatbots that learnt to recognise human language through natural language processing (NLP). More recent use of API integration, large language models (LLMs), and retrieval augmented generation (RAG) has enabled those AI chatbots to morph into complex AI assistants that can now take actions on our behalf.

Today, those smart AI assistants are learning to make decisions so they can operate autonomously, and the result of that is the formation of separate AI agents that break away from more general duties to perform isolated tasks.

While the AI agents we’re building today perform specific tasks, the AI assistants we’ve been creating for 10 years on AI Studio oversee the AI environment.

Example: AI-managed expenses

1. An employee speaks to the AI assistant to submit their expenses claim

The AI assistant’s been trained by the company to recognise requests from their employees and be able to speak with them in natural language, like a human would. It’s a true assistant, learning over time what people need to do and checking everything’s in place for them to do it, ready for an AI agent to complete the tasks needed for the employee to submit the expenses claim.

2. The AI assistant calls the appropriate AI agent to handle the expenses claim

As soon as the AI assistant knows what work needs to be done, it passes the task over to an AI agent that’s specifically trained to handle expense claims (there may be other AI agents employed to handle smaller, specific tasks within this too). The AI agents can pull information from other systems, or add to them, and make decisions on how best to process the claim.

It’s expert in its one task only, able to:

  • Read a scanned receipt to check the expenses claim is legitimate
  • Verify the employee’s identity
  • Confirm the expense is for business purposes
  • Check the date for the claim is within the acceptable timeframe
  • Identify any errors in calculations for the claim
  • Pass the claim back to the employee, if needed, or process it through integration with any system a business might use to log and approve expenses

3. Once the AI agent’s completed its role the AI assistant takes over again

The AI assistant lets the employee know if the claim is approved or rejected, based on the actions and decisions the AI agent has taken.

Rather than the employee having to make a phone call, send an email, log into a computer system or chase a colleague, they simply chat with the AI assistant on any channel it appears ― using text or voice. The conversation is instant, can happen at any time of the night or day at the employee’s convenience, and in any language from any part of the world.

Illustration showing a chat window an AI assistant is in, alongside images of customers making requests in that chat and AI agents of different colours and designs with arrows to show the back and forth liason

AI agents can carry out a limitless range of everyday business processes, from simple tasks like processing product returns, to more complex tasks like optimising your inventory levels and identifying potential supply chain disruptions.

The AI assistant orchestrates every part of all they do as the glue that holds everything together.

How AI agents work

You can use a platform like AI Studio to create an unlimited amount of AI agents, but don’t be fooled by their speed and agility. Creating a capable AI agent demands a complex arrangement of technology and the support of experienced technologists, language experts, psychologists, writers and designers who know how and when to successfully apply human reasoning to the technological tools.

Example: Handling of invoice submissions

To take on this task, the AI agent needs to know the process for how a supplier submits an invoice to your company for payment.

  • Every element of the process is stripped down to individual tasks
  • The AI agent needs to be taught how all these different tasks slot together
  • It also needs to know what happens if there’s a problem along the way too, able to make decisions based on the knowledge it has to make sure the process is still followed through to completion

To achieve this, a human AI administrator puts in place instructions for an AI agent to follow repeatedly without deviation (until the administrator makes a change). These instructions mean the AI agent will know what text response to send when, what links to share when appropriate, be able to offer multiple choice options, ask questions, accept files, or take an action through one of your favourite business systems, like:

All of the actions and responses used by the AI agent are curated by a human AI administrator who understand how to start and finish the process to get an invoice submission through. Traditionally, they’d expertly handle twists and turns that can happen along the way face to face, over the phone or by email to make sure the supplier is still paid on time. With automation, they simply train the AI agent to follow their same reasoning to complete each necessary task.

Training materials

AI agents are trained by the AI administrator on highly detailed and specific company information using large language models (LLM) and retrieval augmented generation (RAG). Using an AI platform with these features, the AI administrator will:

  • Add website URLs to instruct the AI agent to process the information that exists on a page, so it can use this knowledge to answer requests where needed.
  • Upload all kinds of documents, from Word and Excel to PDF and text, no matter how long or complex they are. These will give the AI agent deeper instruction from your customer service, marketing, sales and product teams ― or any other team, about your brand, processes, and customer experiences.

All of this means your AI agent is capable of delivering precise, up to date information and to make appropriate decisions based on everything it’s processed. To keep control of your AI-driven communications and keep your messaging consistent, though, you must introduce LLMs safely with strict guardrails in place.

Job displacement and new job creation

As their knowledge grows, AI agents can take on more and more tasks on behalf of your teams and, while it’s vital to keep a human in the loop working behind the scenes to keep everything tickety-boo, the opportunities this opens up for your teams are vast.

How many of the mundane, monotonous, and manual processes you do today could be automated? How many routine or repetitive tasks could be done for you, so teams can be working on more rewarding, demanding, or profitable tasks? AI agents make all this possible, curated by humans to better serve humans.

As AI agents begin to take on more and more for us, we’ll all need to adapt to the displacement of some traditional roles that can now be automated, take on new roles working alongside AI, and adapt to change that evolves with the growth of technology.

“The good news is that worker displacement from automation has historically been offset by creation of new jobs, and the emergence of new occupations following technological innovations accounts for the vast majority of long-run employment growth. The combination of significant labor cost savings, new job creation, and higher productivity for non-displaced workers raises the possibility of a productivity boom that raises economic growth substantially, although the timing of such a boom is hard to predict.”

Goldman Sachs

Global Economics Analyst, 2023

Getting ready to introduce AI agents

As you prepare for increased AI activity within your organisation, it’s important to recognise there’s no confirmed definition for what AI agents are yet across AI companies and providers, or for those who use them. AI technology is advancing as we type, making it confusing and overwhelming at times for business leaders to do due diligence, so to get the results you need from AI, our advice is to:

  • See proof of what an AI platform can do for you. Ask for a demo, get a proof of concept for your AI agent, question how the technology has been built, how it operates ― LLM alone won’t bring you ROI, and the experience of the team.
  • Check pricing carefully and know what you’ll pay for long term. Some AI providers will charge you per utterance (every back and forth in an online chat or phone call), but with us you’ll only pay for what you use.
  • Look for a free trial period so you can get comfortable with the technology to know it’s right for your organisation and able to give you results you want and need.
  • Make sure you’re tracking key metrics before you create an AI agent, so you can see immediately the positive impact it has, deflecting business processes to AI.

This post highlights our definition of AI agents after 23 years in data management and a decade in AI. We’re always happy to go into more detail on a call, or show you how AI Studio works ― just book in online to speak with us in person.

Let us know what problem you’re trying to solve, and we’ll show you how to do it using AI agents and assistants to establish your company AI.