Agentic AI: The Second Wave of Generative AI

Imagine solving business problems by simply defining the “what”—the desired outcome—without needing to preprogram the “how.”

In essence, this is the promise of Agentic AI: a new paradigm of autonomous multi-agent architecture that collaboratively manages complex tasks with simplicity and flexibility.

Recognized as the next significant evolution in Generative AI, this innovation involves intelligent software solutions (agents) that extend the capabilities of generative AI models by giving them the ability to “reason” to solve business challenges.

These solutions are designed with a set of corporate functions and utilities, referred to as “tools” (such as speech-to-text, computer vision, web search, web scraping, code execution, API calls, etc.). When faced with a new problem, the agent focuses primarily on understanding “what” needs to be achieved. Leveraging generative AI, it employs the “divide and conquer” strategy, breaking the problem into subproblems. Each subproblem is refined to a granular level where it can be addressed using one of the available “tools.” In essence, the agent plans the “how” and executes it effectively.

The emerging promise of Agentic AI, often referred to as the great promise of this technology, lies in its potential to automate, scale, and enable business processes that require cognitive actions in a flexible and adaptable manner. These include resolving incidents, designing and executing strategic actions, closing hiring processes, addressing new demands, or replanning delivery schedules.

Agentic AI becomes the cornerstone of what we call "intelligent or dynamic value chains." These chains are not pre-defined but adapt to various scenarios, providing maximum flexibility and adaptability to meet the ever-changing demands of today’s market.

The benefits proposed by Agentic AI extend to virtually every sector and numerous processes across all industries, embracing business operations and internal support activities, including, of course, the software lifecycle.

Software development and operations are characterized by repetitive cognitive tasks. Each day, developers, analysts, and operators apply their knowledge, experience, and intelligence to resolve issues that could be accelerated with this new paradigm. For instance, agents could monitor a platform, and upon detecting an error, review logs, identify the problem, and suggest an adjustment in the code or even deploy it directly.

Another significant expectation associated with Agentic AI is the concept of the data flywheel. This means that agents will not only resolve situations they had no prior knowledge of but also consider the outcomes of their plans to improve decision-making for future plans. This creates a "wheel" in which agents learn and continuously improve autonomously.

We are at the beginning of the Agentic AI paradigm. The foundational technology is reaching a level of maturity that allows companies to start exploring potential use cases and experimenting with confidence. In the coming years, this technology will drive significant change, provide important competitive advantages, and potentially transform enterprise architectures. It's now up to the AI community to take the leap and start trusting these solutions.