Bill Wong – AI Research Fellow, Info-Tech Research Group.
As the AI research fellow for a research firm, I have the opportunity to interact and share our best practices with hundreds of different organizations pursuing their AI strategy. So, it’s no surprise to me that agentic AI has captured the world’s interest and that organizations are trying to determine how to proceed with this technology. Drawing from my experience, I’ll briefly review agentic AI’s capabilities, use cases, challenges and recommended next steps.
The introduction of large language models (LLMs) has ignited the AI tools ecosystem. An explosion of new tools is now available for developers and data scientists to build or tune AI models. The AI technology stack is composed of the following layers: applications, data and AI tools, foundation models, data platforms and infrastructure. The latest category of tools enables the development, deployment and orchestration of AI agents. AgentOps tools work with these other categories of tools to accelerate the development of applications leveraging foundation models:
• DataOps: These tools focus on delivering quality data by accelerating and integrating tasks that ingest, cleanse and transform raw data into validated data to support AI and analytical applications.
• MLOps: These tools focus on accelerating the development and deployment of machine learning algorithms (e.g., natural language processing) or applications (e.g., speech or video) into production.
• LLMOps: These tools focus on the development and deployment of LLM-based applications, which include prompt engineering, deploying retrieval augmented generation (RAG)-based applications and fine-tuning AI models.
• AgentOps: These tools focus on the development and orchestration of AI agents to deliver an autonomous and adaptive AI system that can leverage AI models and external tools and integrate with the external environment.
Agentic AI Capabilities
Generative AI focuses on generating new content (e.g., text, code, audio and video), while agentic AI goes beyond those capabilities, leveraging AI models to deliver a system that’s goal-directed, autonomous and adaptive.
Although agentic AI and generative AI systems both use LLMs to perform tasks, agentic AI differs from generative AI with respect to how responses are generated. Agentic AI applications are autonomous; they can take LLMs and integrate them with a workflow. Agentic AI can do this because it can interact with the digital and/or physical environment via the use of sensors or IoT devices.
Agentic AI also uses LLMs to set goals and decompose those goals into subtasks. AI agents can then call tools to perform these subtasks. Examples of subtasks include performing a Google search to find information or calling a spreadsheet to perform a financial analysis on interim results. After the results from each subtask are returned and synthesized into a response, the agentic AI system stores its response and uses this information to learn and improve its performance for future requests.
So, while generative AI can generate a quick response to a user’s request to research a particular topic, agentic AI can take that same prompt and generate multiple possible responses, compare results, conduct a financial analysis and determine which research is more relevant before returning the results to the user.
Agentic AI Use Cases
Consider an example of agentic AI being used in the financial services industry. Generative AI can provide an assessment for a loan request by generating a credit report. Agentic AI, however, could take the same request to assess a loan application and also manage the entire process of evaluating the loan, perform additional assessments, such as producing scenarios that, if they occur, would make the loan more viable, and make the decision whether to approve the loan.
A 2025 Info-Tech survey identified the two most popular agentic AI applications planned for 2026. First, our respondents intend to use agentic AI systems to provide operational excellence by automating complex, multistep workflows requiring little to no human oversight. Second, they plan to use agentic AI to improve the customer experience by moving from simple chatbots to virtual assistants that can deliver highly personalized, proactive and goal-oriented actions.
Although the vision and promise of agentic AI are compelling, this technology also introduces new challenges. Most organizations will find that governing the actions and performance of AI agents will likely consume most of their development efforts. For each stage of development and when the application is promoted into production, the outcomes of what the AI agent generates need to align with the organization’s adopted AI principles and safeguards.
Agentic AI: The Next Generation Of AI Applications
Agentic AI is poised to be the next major evolution of generative AI, driven by applications that can transform operations by driving revenue growth, delivering hyper-personalized experiences, improving operational excellence, reducing time to market and mitigating risk.
I’ve found that many organizations are currently involved with assessing and working on agentic AI proofs of concept this year. By next year, I expect that the adoption of agentic AI in production use will increase exponentially.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
