Joy Agwunobi 

A new report from McKinsey & Company has spotlighted agentic artificial intelligence (AI) as one of the most disruptive technology trends shaping the global business landscape in 2025, with growing interest among enterprises but only limited real-world deployment so far.

The study, which tracked the impact of 13 frontier technologies across industries, revealed that while many companies are experimenting with AI agents through small-scale prototypes, few have scaled them into core operations. Despite this cautious approach, momentum is building rapidly, with patents, research activity, and targeted investments in agentic AI accelerating off a small base.

According to McKinsey, agentic AI represents a step-change in how organisations can deploy automation. Unlike earlier chatbots or predictive systems, agentic AI is designed to act on the world rather than simply provide responses. These systems, built on large language models (LLMs), can autonomously plan and execute multistep tasks, interact with digital tools such as browsers and forms, and even communicate with one another to complete workflows.

“Agentic AI moves AI from a passive tool to an active collaborator with enterprise workflows,” said Delphine Nain Zurkiya, senior partner at McKinsey in Boston. “As these systems gain autonomy and decision-making capabilities, it is critical to invest in figuring out how to work with AI when it’s seen as a colleague versus a tool.”

This transformation enables businesses to manage AI agents as “virtual coworkers,” giving instructions in natural language and receiving readable work plans in return. Developers are also experimenting with multiagent structures where a “manager” agent delegates tasks to specialised subagents, mirroring how human teams organise projects.

Why businesses are paying attention

The McKinsey report pointed out that agentic AI holds distinct advantages over previous automation systems, particularly in the way it handles tasks and integrates into business operations. 

One of its most notable strengths is its ability to deal with unpredictable scenarios. While traditional rule-based software required painstaking programming and often left a long trail of exceptions that needed human intervention, LLM-powered agents can respond correctly even to inputs they have never encountered before. This allows them to manage a wide range of tasks that cannot easily be codified into preset rules.

Another defining capability lies in the way these systems interact with digital tools. Rather than relying on complex, custom-coded integrations, agentic AI agents can use the same platforms designed for human operators—such as web browsers—to read websites, process information, and complete forms. This makes them far more adaptable in real-world settings.

In addition, their capacity to understand and respond to natural language allows them to be managed much like virtual coworkers. Business leaders can give them instructions, provide coaching, or refine their performance using the same conversational language employed when interacting with human employees.

Finally, agentic AI brings a level of transparency to automation that was largely absent in earlier systems. These agents generate work plans that can be read, understood, and modified by humans, creating a feedback loop that improves efficiency while ensuring accountability. This ability to explain their actions and adapt to guidance strengthens their role as collaborative partners rather than black-box systems.

These capabilities have sparked strong interest in sectors with rich datasets, such as software development, mathematics, sales optimisation, and customer support. Specialised business-focused AI agents are emerging, reducing the complexity of prompt engineering while promising measurable improvements in performance.

“AI agents won’t just automate tasks, they’ll reshape how work gets done,” added Lareina Yee, senior partner and McKinsey Global Institute director in the Bay Area. “Organisations that build teams where people and agent coworkers collaborate will unlock new levels of speed, scale, and innovation.”

Advances and Applications

McKinsey’s research also pointed to a number of breakthroughs that are rapidly expanding the frontiers of agentic AI. One of the most notable is the rise of general-purpose platforms, as developers either embed agentic capabilities into existing AI systems or design entirely new frameworks capable of handling a wide variety of tasks. This approach is laying the groundwork for more versatile applications that can adapt across industries.

Equally significant are advances in multi-step reasoning. New methods now enable AI agents to break down complex problems into smaller, manageable stages, allowing for greater accuracy and a deeper awareness of context. Rather than relying solely on scaling up foundation models, developers are creating multi-agent workflows in which one system assumes a managerial role, assigning tasks to more specialised subagents.

The report also highlighted the emergence of so-called “knowledge agents.” These tools are designed to conduct in-depth research, comb through vast troves of data, and synthesise findings into comprehensive reports. Their ability to autonomously evaluate hundreds of sources reflects a broader shift toward AI systems that not only retrieve information but also reason and generate insights at scale.

Another area of rapid progress involves AI-to-AI communication. Increasingly, agents are being designed to interact directly with one another, even teaching and describing tasks in ways that reduce the cost of machine-to-human interaction. The report noted that while this development holds promise for robotics, problem-solving, and large-scale coordination, it also raises important questions about transparency and control, as the internal “languages” of AI systems are often opaque to human supervisors.

While interest in agentic AI was relatively muted in 2024, McKinsey noted that its growth trajectory is now steeper than any other frontier technology trend.

Talent and Workforce Shifts

The rise of agentic AI is also reshaping demand in labour markets. Job postings related to the field—especially for software engineers, data scientists, and machine learning specialists—have grown significantly since 2021, even though they still represent a small share of total AI hiring.

Agentic AI requires a mix of established and emerging skills, ranging from Python programming and machine learning to natural language processing and prompt engineering. Importantly, McKinsey noted, the technology is also changing the nature of work itself. Instead of deterministic coding, teams are now focusing on higher-order responsibilities such as task orchestration, decision-making, and contextual analysis.

Risks and Governance Challenges

Despite its potential, the report cautioned that agentic AI introduces new layers of risk. These include erroneous decisions, unintended actions, data quality issues, adversarial attacks, and model drift. As AI agents take on more autonomy—sometimes executing financial transactions or operating across multiple digital platforms—questions of trust, accountability, and liability are becoming pressing.

High-profile pilot deployments have underscored the need for strong governance frameworks, particularly around transparency, ethical guardrails, and oversight.

McKinsey further outlined several questions that companies must consider as they embrace agentic AI: How will the workforce evolve when digital labour works alongside human teams?; What trust and safety mechanisms are needed to manage autonomous systems?; Will agentic AI elevate expert talent by automating routine tasks or displace jobs reliant on structured, repetitive work?; How much autonomy should AI agents be granted, and what balance with human oversight is appropriate?; and How can companies translate experimentation into large-scale value capture?

For now, McKinsey observed that most organisations are still in the testing phase, piloting agentic AI without focusing on near-term returns. Yet the consultancy believes deployment and impact could accelerate quickly as advances in reasoning, collaboration, and specialised applications make the technology more business-ready.

The report added that agentic AI may be the fastest-rising star in the technology ecosystem, but its success will depend on whether enterprises can harness its promise while putting in place the governance structures needed to keep it accountable and trusted.

Delphine Nain Zurkiya, senior partner in Boston,said,“Agentic AI moves AI from a passive tool to an active collaborator with enterprise workflows. As these systems gain autonomy and decision-making capabilities, it is also critical to invest more in figuring out how to work with AI when it’s seen as a colleague versus a tool. At the same time, we will need strong governance, transparency, and ethical guardrails to ensure that these agents operate with accountability and build lasting trust.”