La Trasformazione Digitale
Agentic AI: Will this be the new catchphrase of 2025? Most likely this form of artificial intelligence that is already on the lips of analysts and commentators will be one of the technology topics of the year. Stealing the definition from Ibm, agentic AI is a “system or program capable of performing tasks autonomously, acting as a user or other system, designing its own workflows and using available tools.” Thus, not just a “co-pilot” (an expression initially used by Microsoft but now entering the common technology lexicon), finding information and making suggestions, but something additional.
While generative AI systems create content (text, images, video, audio, software code), the output of an AI agent is an action or decision, for example, making a purchase or responding to customer inquiries or processing bureaucratic procedures, and yet filtering and processing data or distributing marketing content or creating onboarding procedures for new employees or customers. An AI agent draws on the knowledge base of a Large Language Model but then combines the already learned data with new ones again and again, being able to interact with external applications and databases (via APIs, i.e., application programming interfaces). Of course, countermeasures can be built into the software to limit the risk of errors, and in addition, the agent continuously improves over time, learning from experience and adapting to the external environment.
If Large Language Models and chatbots built on them have been the big technology revolution of the last two years (a user experience revolution, if not yet from an adoption perspective), could Agentic AI be the next big thing? Are we ready yet, considering that previous innovations have not yet been fully internalized by companies? Many are betting on it. A few examples, among vendors that have recently launched AI agent innovations: Cisco in the area of customer support (with AI agents integrated into Webex), Salesforce for all in Crm workflows (with the ability to create custom agents), ServiceNow for IT troubleshooting automation. And consulting giant Accenture has created, in partnership with Nvidia, an entire business division dedicated to this technology.
According to Gartner, which listed it among the top ten technology trends of 2025, Agentic AI will take companies to new thresholds of productivity and efficiency, freeing staff from the most repetitive tasks. Analysts predict that in companies by 2028, AI agents will take over at least 15 percent of daily work decisions.
And it pays to prepare now: like it or not, artificial intelligence is better to know it, lest you be overwhelmed by it. The outlook painted by analysts is in some ways tantalizing, in others a bit disturbing. “It is clear that no matter where we go, we cannot avoid the impact of AI,” commented Daryl Plumer, distinguished vice president analyst, research director at Gartner. “Artificial intelligence is evolving along with its uses by people. We have to accept that AI can improve us before we get to the point where humans can no longer keep up.”
Among the supporters of AI agentics is Ibm, which has listed four reasons why it says this technology is destined for success. Number one: it is both flexible and precise, as it employs a “hybrid” approach to writing software (partly based on Large Language Model and partly on the more structured rules of traditional programming). Second, compared to classical Large Language Models an AI agent has broader knowledge and better contextual intelligence because it can draw on Web sources, logs, and queryable databases. The third success factor of AI agents is their autonomy, the ability to act even without human supervision, managing even complex procedures and monitoring their progress over time. Finally, fourth advantage, AI agents are intuitive to use because you can interact with them through natural language. All of this, according to Ibm, suggests that in the future Agentic AI tools will replace or transform many of the Software-as-a-Service applications currently in use.
