Agentic AI is gradually going from a peripheral concept to one of the most talked-about shifts in enterprise technology. It is gaining traction as companies explore new ways to automate workflows and delegate tasks to “virtual coworkers,” rather than just interact with chatbots. Agentic AI can be proactive, often utilizing digital tools, rather than simply providing outputs. These systems are built on top of AI foundation models and can autonomously plan and execute multilayered tasks.

Just imagine a scenario where a customer service AI agent answers questions about products, processes orders, and manages returns by connecting to an organization’s logistics systems. Several organizations have released deep-research agentic AI that design their own workflows to research topics on the web and even produce reports. Even more companies are using software programming agents, which apply their multilayered reasoning to write, deploy, and test code, by only being given a description written in English (or any other known natural language).

Benefits of AI Agents

The benefits that agentic AI proffers over other previous systems include the following capabilities:

• Serving a long tail of unpredictable tasks: Previously, to create an application that could act autonomously, engineers had to painstakingly program step-by-step, rule-based systems. Many such software had a long tail of exceptions to their rules, which a human had to address. By contrast, large language models (LLMs) can respond correctly to inputs they have never encountered before, enabling an LLM-based agent to handle a long tail of tasks not easily codified into preset rules.
• Using digital tools designed for a person: Formerly, sending or receiving data required custom code that would connect each new digital system. However, AI agents can use the same tools that a person would, such as a web browser, to “read” websites using their LLMs and also to fill out forms.
• Receiving instructions in natural language: AI agents can be managed like virtual coworkers, including giving them instructions and coaching them on how to do their jobs better, using the same kind of language you would use to interact with a human coworker, because LLMs can process natural language.
• Generating work plans that can be understood and modified: AI agents based on LLMs can create work plans and, depending on their design, can communicate among themselves. Because these agents use language that humans can read, they can describe what they’re doing and be guided via feedback on their work plan.

The potential of agentic AI has persuaded many industries to explore enlisting agentic virtual coworkers for various functions and roles.

Latest developments

The development of agentic AI with the capabilities of autonomous decision-making and interagent communication presents interesting possibilities. However, the rapid progress of AI agents underscores a crucial need for solid governance frameworks to address trust, liability, and ethical concerns. The following includes the latest developments in agentic AI:

• Developers are building general-purpose AI agents. Some companies are adding agentic capabilities to their existing AI offerings, while others are building on these capabilities to create targeted, task-specific applications. These additional capabilities enable the development of agents that can interact with users through natural language and perform many different tasks. Progress is most rapid in fields that have more robust data sets for training and evaluation, such as software coding and mathematics.
• Increasingly long chains of effective multistep reasoning reflect meaningful progress in agentic AI. Over the past year, new techniques have improved AI’s ability to tackle complex, novel tasks by breaking them into smaller steps. Rather than relying solely on scaling foundation models, developers are now deploying multi-agent workflows in which a “manager” agent builds a work plan and delegates tasks to specialized subagents. While there is more to do to assure trust and security, this shift allows for more accurate, context-aware outputs, a major step forward in how AI systems reason and operate.
• There is a new focus on agentic AI for specific business solutions. AI agents are increasingly being developed to address specific, high-value business problems. More specialized and tuned to their specific tasks, these agents reduce the need for users to craft complex prompts. Early attention has focused on the use of agentic AI for software development, where capabilities have been advancing rapidly. In addition, there is significant interest in AI applications that can deliver measurable improvements in core business metrics, particularly sales optimization and customer support automation. As this trend evolves, enterprises will need to balance the use of specialized agents in workflows with the potential for more general agents to execute a variety of tasks.
• Momentum behind deep-research knowledge agents is growing. Multiple providers are advancing tools that can autonomously conduct multistep explorations for relevant content, execute searches, evaluate hundreds of sources, and synthesize information into comprehensive reports. These agents reflect a broader shift toward using AI not just for retrieval but also for reasoning, enabled by faster knowledge generation that can be scaled.
• AI agents can “speak” to one another. Recent advances in AI include models that can communicate with one another and create their own languages. Neural networks can now learn tasks and describe them to other AI systems. Processing this AI-to-AI communication costs less than processing an AI-to-human interaction. These developments in AI-to-AI communication have implications for robotics, complex problem-solving, and other fields, though they also raise concerns about transparency and control.
• Rising concerns about trust, governance, and liability are influencing the development and deployment of agentic AI. As AI agents take on more autonomous roles that include executing financial transactions and interacting across digital platforms, businesses are increasingly grappling with accountability and legal frameworks. Recent high-profile pilot deployments have brought these risks into sharper focus, especially as AI systems act independently across jurisdictions. Designing robust guardrails and providing the right operational context for agents will be essential to ensure reliability and accountability.

The Labor Markets and Talents

The volume of job postings related to agentic AI remains relatively small; however, it has grown significantly since 2021, especially for roles like software engineers, data scientists, and data engineers. This growth suggests an increasing interest and investment in developing AI systems capable of autonomous decision-making and action.

The development of agentic AI relies on a mix of technical skills, including Python programming, machine learning, and software engineering, and other skills in emerging areas like prompt engineering and natural-language processing. While the demand for talent is high, the picture is mixed. Some skills, such as using TensorFlow, are more readily available, relative to demand, while others, such as expertise in Python, are in shorter supply vis-à-vis demand.

AI agents are also reshaping the nature of work in its entirety, shifting responsibilities from deterministic coding tasks to higher-order activities like task planning, tool orchestration, and contextual decision-making. This evolution is changing how roles are defined, what skills are valued, and how organizations structure technical teams.

AI Agents Adoption

Organizations keep testing the functionality and viability of agentic AI with small-scale prototypes, typically without a focus on an immediate ROI. A few other companies are scaling or have fully scaled their technology.

Despite significant interest in the technology, it remains largely untested in real-world business contexts. Many organizations are actively testing the functionalities of AI agents through small-scale prototypes, but full-scale adoption remains limited. However, with the rapid advances in technology, agentic AI is worth watching closely, as deployment and its impact could accelerate swiftly.

Sample Organizations Using AI Agents In the Real World

• Anthropic released the Model Context Protocol (MCP) as an open-standard, open-source framework to standardize the way AI models such as LLMs integrate and share data with external tools, systems, and data sources. Google, Microsoft, OpenAI, and many others have also announced they would adopt MCP.
• Google introduced the Agent2Agent (A2A) protocol in April 2025, an open standard to facilitate secure collaboration between AI agents across vendors. Supported by more than 50 partners, Agent2Agent (A2A) enables use cases such as candidate sourcing and supply chain coordination, complementing efforts like MCP to unlock scalable, multi-agent ecosystems.

Considerations About the Future

Organizations and leaders may want to consider a few questions when moving forward with AI agents:

• What are the workforce implications of agentic AI at scale, which will include a combination of human and digital labor?
• What trust and safety techniques will be provided to mitigate risks as organizations adopt agentic AI?
• Is an AI agent more likely to elevate expert talent by automating routine tasks or to displace large segments of the workforce whose roles are built on structure and repetition?
• To what extent should an AI agent be allowed to operate independently? How do we have a good balance between the autonomy of AI and human oversight?
• How can organizations get ahead of their competitors and capture the value at scale, either in revenues or cost benefits, associated with agentic AI?