The Rise of AI Agents: From simple tasks to autonomy
Our AI Lead, Pernille, explores the evolving role of agents, where personal productivity tools are becoming the interface for business process agents
What is an "agent"? It’s one of those terms in the realm of AI that’s on everyone’s lips but still feels a bit elusive. Imagine a digital assistant that's more than just a tool – it’s a proactive helper that can either wait for your command or jump into action autonomously based on the context that you are in.
Currently, we are looking at agents from two perspectives: the ‘personal productivity on job’ agent and the ‘business process’ agent. These are connected from a user interface perspective, with the idea that the personal productivity agent is becoming the user interface for your business process agents. This is good news as it reduces the number of applications and thereby limits the user interfaces in the workplace.
In a perfect world (warning: these are my speculations) this might limit the amount of adoption needed, since there are no buttons to click or user interfaces to figure out. There might just be ‘approve’ or ‘reject’, the rest is done by an agent.
Well, we are not there just yet, so we will leave the future for now, but I guess this just once again underpins that we need to start rethinking the way we use IT altogether.
Let’s take a closer look at the agents for Business processes, and here we currently observe three different types of maturity levels among our customers, which reflects the way Microsoft defines it pretty well: Retrieval, Task and Autonomous agents
Our customers are maturing by starting with creating a Retrieval Agent.
A Retrieval Agent is the well-known ‘chat to your data’ agent. An agent that surfaces knowledge from your organizations data storage or knowledge bases like the intranet, employee guides or customer service.
Most of us know these types of agents by now as chatbots since they often have a chat user interface.
The second type of agent is a Task Agent, which can perform a variety of tasks without requiring human intervention. A Task agent could for example:
- Organize and categorize emails in an inbox for example the customer service inbox, ensuring that messages are distributed to the right people and junk mail is filtered out.
- Prefill forms with data retrieved from relevant databases.
- Schedule appointments and meetings by checking the availability of participants and sending out calendar invites including agendas.
- Generate reports by collecting and analyzing data from various sources and then presenting the information in the right format.
- Listen in on meetings and filling out required documentation afterwards.
These task agents can draw upon the Retrieval Agent and over time, these Task Agents could evolve into Autonomous Agents, capable of making decisions and taking actions independently, without the need for human oversight. For now, our customers want to be able to approve or reject the tasks that the agents take, but I’ve already heard of companies that no longer require this. Instead, they evaluate the agents' work the same way they would assess a human's performance, accepting occasional errors as part of the process
Please remember that even though some agents are simple to create, they should be supported with proper governance to ensure they function effectively and prevent your organization from becoming cluttered with agents popping up like Mr. Smith in the Matrix.
At Fellowmind, we’re inspired by Microsoft’s vision of the diverse spectrum of agents and the immense potential they offer to our customers and industries worldwide —these agents represent not just an elusive buzzword, but a fundamental shift in how we interact with technology. As we explore how to leverage these capabilities, it’s clear that the AI-driven future, powered by these dynamic agents, is already here unlocking a world of new possibilities
