The world of AI is evolving rapidly. We now cannot think of a world without AI as it has become a part of our life now. Frequently, we are seeing innovations in AI and now Agentic AI has taken the center stage. To understand better about Agentic AI, we first need to understand the evolution of AI that led us here.
From Monolithic Models to Compound AI Systems
Traditional AI models are powerful, but they are inherently limited by one thing – the data they are trained on. It is challenging to make these models adapt and often requires huge investments in data and resources for tuning. Compound AI systems were introduced to overcome this challenge. These systems follow system design principles which involve creating modular systems with multiple components. One can add tuned models, programmatic elements, output verifiers and various other tools as components in the compound AI systems. This helps in this faster adaptation and hence became more efficient in problem solving compared to traditional models.
Retrieval Augmented Generation (RAG)
RAG systems can retrieve information from external sources when required and combine it with model-generated responses to answer user queries. The control logic of these systems ensures the accuracy and relevance of the answers generated by the models.
The Agentic Approach
In the simplest manner, we can say that Agentic AI acts like a detective. If we ask a complex query, it does not just check on its library, instead it makes a plan, figures out what steps to take, and decides which tools to use to solve the problem. This approach made a significant shift in the traditional AI system design. By placing a large language model (LLM) in charge of the control logic, AI agents can reason, plan, and act autonomously.
To illustrate how Agentic AI can be applied in real-world scenarios, let us consider one example where a user asks a query: “I want to plan a road trip from Bengaluru to Mumbai, stopping at Pune and Goa. Can you help me figure out the best route, places to visit, and where to stay?” Now, let us see how the Agentic AI solves this:
- Reasoning:
The AI agent starts by understanding the goal: planning a road trip with specific stops. It breaks down the task into smaller parts: finding the best route, identifying places to visit, and locating accommodations. - Acting:
Route Planning: The agent uses a map tool to calculate the best driving route from Bengaluru to Mumbai, including stops in Pune and Goa.
Places to Visit: It searches travel websites and databases to find popular tourist attractions and activities in each city.
Accommodations: It checks hotel booking sites to find suitable places to stay in each city, considering user preferences like budget and amenities. - Memory:
The agent remembers the user’s past preferences for travel, such as preferred types of accommodations or favorite activities, to personalize the recommendations.
This example helps us understand the modularity and adaptability of AI agents, showcasing their ability to handle complex, multi-step problems.
Advantages of Agentic AI
- Autonomy: Agentic AI can operate independently, making real-time decisions.
- Enhanced problem-solving: Agentic AI can tackle complex problems better than traditional ones because of its ability to reason, act and memorize.
- Personalization: Agentic AI can remember user preferences and past interactions allowing it to provide personalized responses and solutions.
- Innovation and insights: Users gets more informed decisions and can drive innovations dues to the data-driven insights given by Agentic AI
The shadow side
Like every other AI implementation human oversight is still needed in the case of Agentic AI as well. Risk of errors, bias and the need for quality data are also factors that affect the output of Agentic AI.
Agentic AI is a big step we have taken in the technology realm. The innovations, assistance and services that it can bring are up to the imagination of us.
