Have you ever wondered what it would be like if you had your own personal assistant who could automatically solve complex tasks without the need for detailed instructions? This is the promise of large language models (LLMs) like ChatGPT, which act as a central logic unit for different tasks. However, even more advanced possibilities are promising with so-called LLM agents.

Over the past year, Medius has also been exploring how LLM models can act as the central brain of applications that independently access different tools and data sources. An example of this is the rabbit r1 device, which acts as a kind of central unit with access to different applications. As developers of customised solutions, we are of course particularly interested in the possibility of developing solutions for specific challenges.

What are LLM agents and how do they work?

The LLM agent is actually a smart language model that has access to various tools. These tools allow it to act as a smart assistant that adapts according to the instructions we give it. Instead of programming each task individually, the agent autonomously chooses the best approach to achieve the goal.

Let's try to explain with a practical example. In the user interface or dialog box, we write the following question: "Tell me how many years apart the first and the current President of the USA were born and multiply that number by 17". The LLM agent will first look up the data using a web search engine and then use a calculator to do the final calculation, which means that the agent will independently perform all the steps to arrive at the correct answer.

What are the restrictions on use?

LLM agents have limitations when it comes to the amount of information they can process at one time. If the accumulated knowledge is built up in a longer text prompt, problems can arise. One possible solution (which does not yet exist, but is very likely to be feasible and worthy of attention) is to break down more complex tasks into smaller chunks that can be solved separately by the agent and then formulated into a final answer.

How can companies use LLM agents to their advantage?

Imagine your company has different applications for project management, financial calculations and communication. An LLM agent can connect all these applications and allow you to get the information you need with simple questions. For example, you can ask the agent to summarise current projects, calculate monthly costs and send emails to the team, all in one conversation.

Employees can use LLM agents to automate all kinds of tasks that used to require several different applications and, above all, a lot of time. With simple instructions, agents can get useful answers without the need for specific programming for each task, as LLM agents work quite autonomously.

What's more, we can also give LLM agents access to internal knowledge bases or applications, and then use conversation and guidance to get the information we want. An example would be an LLM agent communicating with a call centre employee and asking even the most complex customer questions. In this case, because the LLM agent has access to the entire knowledge base of the company and does so extremely quickly, it can find answers to virtually any question in just a few moments. As personal assistants to your employees, LLM agents can help you solve a variety of day-to-day business challenges.

LLM agents are already a great opportunity

LLM agents represent the next step in the practical application of AI, providing simpler solutions to complex tasks, with huge benefits for businesses and process optimisation. LLM agents are likely to become an indispensable tool in the arsenal of any company that wants to harness the benefits of AI to solve challenges and improve the customer experience.

With simple instructions via a conversational interface (e.g. ChatGPT), LLM agents can autonomously perform tasks that would otherwise require manual programming each time. This means that businesses can achieve greater efficiency with less effort and time. At Medius, we believe that customised LLM agents are the future, as they open up many opportunities for innovation and progress for users and businesses.