The agent technologies I have researched myself are based quite purely on LLM and specifically on Transformer-based models. This utilizes task assignment with ordinary text and the Transformer’s continuously developing ability to infer.
These are sometimes also called LLM agents.
The agent’s own implementation receives actions from the LLM and executes them. In Sema4.ai, these actions can be added with Python.
DWF has been testing Sema4.ai’s agents in 2024, and according to Karli Kalpala, they are capable of performing tasks that are difficult to define through conditions with RPA.
DWF’s personnel largely consist of business process experts from the business areas DWF focuses on. The biggest investment is likely in Healthcare. For this reason, work efficiency is considered through genuine healthcare business processes, not technological hype.
RPA process optimizations have focused on tasks that genuinely enhance productivity. The task of AI Agents is likely to enhance processes in a similar way, but it only creates more opportunities.
Here’s a LinkedIn link to the message from the CFO of Länsi-Uusimaa (Western Uusimaa) concerning the pre-preparation work for client payment decisions agreed with DWF and the benefits this brings.
The post was liked by 100 people, a large part of whom are from other well-being services counties in Finland, so further work is likely to come if the planned savings are realized.
So, when managed correctly, AI Agents only bring more opportunities alongside RPA.
Edit Regarding the question of whether the term AI Agents is clearly defined, based on a Google search, it is. It’s worth searching for the term “LLM Agents” to find that consistent definition.
