AI Agents Enter the Scene as the Next Step in AI Development

Author: Lider

September 19, 2025

Image by: foto Shutterstock

Two and a half years after the launch of ChatGPT, generative AI has reshaped the way companies utilize artificial intelligence. Artificial intelligence had its place in companies even before that, as it was used for predictions, classifications, and optimizations.

McKinsey’s research ‘The Economic Potential of Generative Artificial Intelligence: The Next Frontier of Productivity’ showed that the estimated potential value was already enormous – between $11 and $18 trillion globally, but AI was mostly the domain of experts, so the pace of adoption among employees was slow.

According to our research, from 2018 to 2022, the application of AI was relatively stagnant, with approximately 50 percent of companies using the technology in only one business function. Generative AI, through the synthesis of information, content generation, and communication in human language, has expanded the reach of traditional artificial intelligence, and our estimate is that this technology can unlock up to $4.4 trillion in additional value on top of the existing potential of analytical AI.

The Paradox of Generative AI

Our latest global research on AI ‘The State of Artificial Intelligence: How Organizations are Transforming to Capture Value’ showed that more than 78 percent of companies today use generative AI, but just as many report that this usage has not made any tangible contribution to the company’s earnings.

The paradox is that despite all the investments and potential of this technology, a greater economic return has not yet been realized for most organizations using it. LLMs (large language models) have revolutionized the way organizations work with data, but they are fundamentally reactive and isolated from business systems. This is where the next step in AI development comes in, which is AI agents, marking the transition of generative AI to autonomous task execution.

They operationally take over routine tasks, but also, as our research ‘Harnessing the Advantages of Agent-Based Artificial Intelligence’ states, transform processes in five ways: they accelerate task execution, bring flexibility as they can change processes on the fly, enable personalization by tailoring interactions to customer or service user behaviors, provide elasticity as their execution capacity can expand or contract in real-time, and increase resilience by monitoring potential disruptions within the system.

For example, we can take an e-commerce system where agents embedded in online stores or applications analyze user behavior and cart content in real-time and offer relevant suggestions for additional purchases. We have also seen results that far exceed mere efficiency gains with the development of our own AI system, McKinsey QuantumBlack, and by helping organizations build AI agent workforces.

Unlocking the full potential of AI requires more than just introducing agents into the company’s existing system. If you simply add agents to the existing system, without redesign, you will only get faster assistants, but the process remains inflexible, rule-driven, and shaped by human limitations.

Organizations therefore need to start reshaping their IT architectures around an agent-first model, where interfaces, logic, and data access are natively designed for machine interaction, not human. This requires a fundamental architectural change from static LLM infrastructure to an agent AI network (mesh) – a dynamic and modular environment built specifically for agent intelligence.

The Role of Company Directors

Trends indicate that AI agents are a turning point that will redefine how companies operate, compete, and create value. To realize the full potential of agent-based AI, you need to reshape your approach to AI transformation: not as a series of piecemeal initiatives, but as focused and comprehensive projects.

This specifically means that you will need to identify several business areas with the greatest potential and embark on redesign at all levels – from workflow to task redistribution between humans and machines to reorganization based on new operational models.

At McKinsey, we witness that some leaders have seriously taken action, not only by introducing agents but also by completely reorganizing their companies to leverage the full disruptive potential. The conclusion is that generative AI is quickly ceasing to be a toy for testing as it has already become an inseparable element of the business of almost every company and will increasingly shape the business environment in the years ahead. Are you ready for it?