Amid all the talk about what AI can do, a friend of mine—who's a stock trader and expert in business taxonomy—asked me about the buzz around agentic AI.
He wondered if an AI agent could really work as a business agent. I didn’t have a clear answer at the time. He confidently imagined agentic AI systems thinking like human business agents and acting through algorithms, which could completely change the trading market.
This got me thinking: can agentic AI really act like an agency?
Understanding the Potential of Agentic AI
In the early days of AI, systems followed strict rules and instructions. These systems couldn’t adapt to new situations or learn from experiences. But as researchers studied human intelligence more closely, they started developing AI systems that could be more flexible and independent.
This led to the rise of agentic AI—a type of AI designed to act on its own, set goals, and learn from its surroundings. AI evolved from simple rule-based systems into more advanced systems that could handle real-world challenges with more independence.
Agentic AI, often referred to as autonomous AI, takes this further by giving AI the ability to make decisions and adjust to changes on its own, offering more autonomy than traditional AI. Organizations using agentic AI can expect it to be goal-oriented, adaptable, and proactive.
Tech teams around the world are improving current AI systems, which are good at recognizing patterns and generating outputs. They are doing this by adding techniques like reinforcement learning, function calling, and Graph RAG to build real agentic capabilities.
Agentic AI, Reinforcement Learning and LLMs: Reasoning and Planning
Anyone who’s familiar with reinforcement learning knows how these systems focus on specific tasks and make decisions based on a reward system. However, agentic AI systems need to go beyond just rewards and punishments; they require broader decision-making abilities.
With large language models (LLMs), it’s easy to get caught up in prompts and responses, but agentic AI is about more than just understanding and generating language. While LLMs are great at language tasks, agentic AI systems aim to show more general intelligence and independence.
Agentic AI and Ontology for Decision Making
Ontology in agentic AI has led to a lot of research on how AI systems understand and reason about the world. Ontology is like a map for organizing knowledge, helping AI agents make sense of their surroundings.
It allows different goal-oriented AI systems to share and understand the same ideas and goals, making it easier for them to work together. As new information comes in or things change, the ontology can be updated to keep the AI adaptable and flexible. This helps AI agents understand how things are connected, what actions to take, and why.
Embodiment is the process of combining different skills like recognizing patterns, seeing through cameras, understanding language, and performing actions. For agentic AI or systems like LLMs to act more independently, they need to develop the ability to perceive the world, take action, and integrate these abilities.
Perception involves gathering data from things like cameras or sensors, while embodiment gives the AI the ability to act on what it perceives.
Right now, LLMs don’t fully have agency, but they are starting to show signs of acting more independently. They can handle more than just following set rules or calling functions and are starting to use some form of general intelligence. Predictions say that agentic AI will continue to improve and get closer to true AI capabilities faster than many had anticipated.
Agentic AI Accelerating Digital Transformation
The impact of agentic AI on accelerating digital transformation is evident. Technology leaders across organizations are leveraging agentic AI capabilities to address Industry 4.0/5.0 challenges and enhance digital processes.
As agentic AI becomes a long-term sustainable asset for organizations, it is also expected to drive digital transformation initiatives.
Some key ways AI teams can support digital leaders include:
- Automating repetitive tasks: Agentic AI can free up human resources by handling time-consuming tasks. Teams will be free to focus on strategic goals.
- Enhancing decision-making: By identifying trends, patterns, and anomalies, agentic AI helps leaders make better technology investments, process improvement, and strategic decisions.
- Driving innovation: Agentic AI can stimulate new ideas and explore innovative possibilities within the organization.
- Optimizing resource allocation: Agentic AI helps organizations allocate resources more effectively. For instance, LLMs and narrative reporting are being explored for ESG reporting.
- Improving cybersecurity: Agentic AI can detect and prevent cyber threats by analyzing network traffic, identifying anomalies, and responding to incidents in real-time without human intervention.
Planning and Executing Agentic AI Systems
Implementing agentic AI systems successfully requires a structured approach.
Here's a breakdown of the key steps digital transformation teams should follow to ensure a smooth and effective implementation.
Step 1: Setting Clear Goals
The first step is to establish clear, measurable goals. Define exactly what the organization wants to achieve with agentic AI.
For example, you might track the number of teams integrating agentic AI frameworks into their workflows. A guiding metric, or "north star," will keep the entire project focused. Goals should be both ambitious and realistic to provide direction for the team and align with the organization’s long-term strategy.
Step 2: Assessing Capabilities
Before diving into development, assess the current internal capabilities. Does your team have the necessary expertise? If not, consider bringing in external partners or consultants to help.
A strong internal team is ideal, but external experts can offer quick wins by jumpstarting key phases of the project. Don’t shy away from outsourcing certain aspects if it accelerates the process and strengthens your overall approach.
Step 3: Choosing the Right Technology
Selecting the right technology for agentic AI is crucial. This choice will be guided by factors like technological compatibility with existing systems, cost-effectiveness, and the long-term vision for the company.
Ask: How will this technology fit with our current infrastructure? What are the costs and benefits over the next few years? Make decisions with an eye on future scalability and innovation.
Step 4: Developing a Broad Data Strategy
Agentic AI thrives on data. You need to develop a robust data strategy in order to fuel its learning and adaptability. This requires collaboration across different boards, such as data science, AI, and ethics committees.
Relying solely on data engineering teams is not enough for large-scale projects. Include input from diverse stakeholders to ensure your data approach is comprehensive and ethical.
Step 5: Deploying Agentic AI Modules
Deployment is where the rubber meets the road. Some models may need re-training to integrate effectively with your system, while new models may need to be built from scratch.
During this step, you need to prioritize team upskilling to make sure the staff are well-equipped to handle and maintain these new systems.
Additionally, focus on developing the necessary infrastructure for smooth implementation, which could include cloud-based platforms, data pipelines, or AI integration tools.
Step 6: Monitoring and Evaluating Performance
Regularly assess how your agentic AI models are performing in relation to your defined objectives. Use this feedback to optimize and improve the system.
Additionally, ensure visibility and accountability—track who, when, and how the system is being used. This is critical for both security and compliance reasons, as some regulators may require transparency in AI-driven actions, especially in sensitive fields like finance or healthcare.
Step 7: Addressing Legal and Ethical Considerations
Legal and ethical frameworks around agentic AI are evolving. Some industries, such as finance and law, may have strict regulations.
Your AI models need to be transparent and explainable, so users can understand how decisions are made. Always follow industry standards, like those from ISO or IEEE to make sure your system is compliant and ethical.
Step 8: Promote a Culture of Innovation
Innovation doesn’t happen in isolation. Encourage a company culture that supports rapid experimentation, learning from failures, and cross-department collaboration.
Break down silos between teams to support a more open exchange of ideas, which can drive faster AI development and integration.
Step 9: Communicating Technology Leadership
Lastly, ensure ongoing communication from technology leadership. Leaders should frequently share the innovation vision through events, hackathons, newsletters, and internal presentations.
Clear, consistent communication will help rally the organization around agentic AI initiatives and maintain momentum throughout the implementation process.
An Example of Agentic AI
Consider a predictive maintenance use case in a heavy engineering plant. Cameras capture images at regular intervals to gather information about the machinery parts. Inspections are carried out by experts to validate issues and take necessary remedial actions.
In the current setup, pattern recognition technology flags malfunctioning or deteriorated parts, and the information is passed to an expert to validate and determine the next steps.
Such plants often face a shortage of experts compared to the rate of failures or maintenance incidents. AI agents can play a key role in identifying and resolving issues by assessing the severity of incidents based on patterns derived from the images.
- Understanding the issue: The AI agent passes the image to a computer vision module, which recognizes any faults or wear-and-tear on the parts. The agent then classifies the issue based on prior information.
- Accessing the information: The AI agent retrieves relevant information from a pool of engineering documents stored in databases and matches the condition with pre-existing cases.
- Performing action: If the confidence in the classification is low, the AI agent can involve a human expert to review the case. The human expert may advise further investigation or guide the AI agent in interacting with robotic components to replace the part.
- Learning from the process: The AI agent will enhance its knowledge base by learning from each new instance, improving the system's overall capability.
Human experts will need to focus on upskilling to better interact with AI agents and address gaps in agentic knowledge that require human intervention. In this case, the AI agent plays a central role in resolving issues and demonstrates a higher degree of agency.
Final Word
Agentic AI is a powerful toolset that can help digital transformation leaders achieve their desired system landscape by automating tasks, enhancing decision-making, improving customer experience, driving innovation, and optimizing resource allocation.
While technology leaders are confident in the use cases that fit agentic system criteria, the technology landscape needs to mature quickly, with the necessary guardrails in place to ensure the ethical and fair use of agentic capabilities.
