Gartner defines agentic artificial intelligence (AI) as goal driven software entities that have been granted rights by the organisation to act on its behalf to make decisions autonomously. “They can be combined with AI techniques which have memory, planning, sensing, tooling and some guardrails to complete the task and achieve a particular goal,” says Anushree Verma, a director analyst at Gartner.
She says that a common question Gartner is asked is how agentic AI differs from robotic process automation (RPA) and intelligent robotic process automation. Her view is that RPA is scripted and has a predetermined output. However, she says: “When we talk about agentic AI, this works with different levels of autonomy. It can have proactive planning as well, but its distinction is that it has a certain level of AI agency so it works autonomously to an extent, and it also works towards a particular goal.”
She believes the biggest impact of agentic AI systems is that they change the future of decision-making. “It just doesn’t analyse complex data sets, but an agentic AI system can also identify patterns and act on them.” For Verma, this means an agentic AI system can help a business solve problems better and reduce the time to action.
Given that enterprise software companies are embedding agentic AI technology into their products, there is a risk that these will end up operating as standalone systems and effectively siloed. Verma says the initial use cases have been around CRM systems, where they are used to improve the user experience. Such AI are generally deployed at the end of an entire workflow to improve the user experience. However, she says: “The real power of agentic is when you start looking at automating or orchestrating the entire business process, taking decisions based on complex workflows and automating them.”
A major barrier preventing the use of agentic AI systems for complex workflow orchestration is that AI is prone to errors, leading to incorrect decisions being automated. This inaccuracy will improve overtime, but there is little evidence that enterprise applications are being developed with tolerance to potential errors introduced by agentic AI systems.
Verma says Gartner has seen a lot of generative AI projects being abandoned because they have not been well thought through, after the proof of concept project has been tested. “They’re not being implemented due to poor data quality or inadequate risk controls, escalating costs or unclear business value.”
Since the majority of agentic AI initiatives are proof of concept projects, Verma believes that error tolerance has not been evaluated sufficiently. She says that the agentic AI market is still at a very early stage of maturity. Potentially, as the market matures, questions of error tolerance are likely to be addressed.
Gartner is seeing a shift towards domain specific models and lighter models that can be fine tuned. Over time, it is likely these AI models will work like experts, to compliment more generat agentic AI systems, which may help to improve accuracy.
#Gartner #agentic #Computer #Weekly #Downtime #Upload #podcast