According to a prediction from research giant Gartner, by 2028, a third of interactions with generative artificial intelligence (GenAI) services are forecasted to utilise action models and autonomous agents to accomplish tasks. These autonomous agents are described as combined systems capable of achieving specific objectives without consistent human intervention. With the ability to learn from their environment for continuous improvement, autonomous agents can potentially manage intricate tasks in the workplace.

Arun Chandrasekaran, Distinguished VP Analyst at Gartner, notes that this could cause a shift in humans' interactions with GenAI. He says, “In the future, human interactions with GenAI may evolve from users prompting large language models (LLMs) to users interfacing directly with autonomous intent-driven agents, which could allow for a higher degree of autonomy and much better alignment with human goals.”

These autonomous agents hold the ability to perform a multitude of tasks that could translate into various capabilities like controlling model output, accessing the internet, using applications, and automated complex business processes based on human intent. Chandrasekaran explains, “Autonomous agents can reduce the need for human intervention when interacting with LLMs and reduce the burden on business users across many sectors, as they are able to spend less time on advanced prompt engineering.”

Looking at potential impacts across sectors, it is predicted that these will include Healthcare, where autonomous agents could assist medical professionals with disease diagnostics, treatment planning, and patient care. In Education, autonomous agents could potentially offer personalised learning experiences and even adapt teaching methods to the needs of individual students. The Gaming sector could see autonomous agents observing and interacting with human players to provide more immersive and realistic experiences, and lastly, Insurance could benefit from autonomous customer service apps capable of handling most policyholder interactions via voice and text, assisting with claims and fraud, medical service, policy and repair systems, Marc thereby drastically reducing resolution times.

However, for autonomous agents to deliver meaningful value whilst controlling behaviours, clear objective functions will be required. Chandrasekaran adds, “The tasks autonomous agents can perform, such as verifying the output of a model before inputting into another model, has the ability to control model output and automate complex business processes based on human intent. But this can only be achieved with a clear objective function.”

To transition effectively towards this future use of GenAI, Gartner suggests organisations should pinpoint use cases where action models and autonomous agents could add value by reducing human effort and skill required. Next, they should create an architecture that will enable these agents to thrive, acknowledged chiefly through tool integration and access to various knowledge repositories and long-term memory. In this manner, agents might be able to demonstrate expanded reasoning and expertise. Equally important though, Gartner calls for a balance between autonomy and control, achieved through extended pilots and rigorous agent monitoring. Lastly, there is the understanding that action models and autonomous agents are not essentially a substitute for prompt engineering, as their potential remains reliant on the quality of the prompts they receive.