Introduction
Artificial Intelligence is evolving at a very faster rate, and it can feel like a lot to keep up with. AI can be described as something that has gained a mass knowledge of everything miraculously. It’s not just about robots anymore, it’s about software that can think about problems that we give them and help us with our daily tasks, and even make complex choices for us.
In this short insights of mine, we’ll look at why this technology is jumping forward so quickly and why “intelligent agents” are the real picture right now. We will also explore how AI uses search and decision making to solve tricky puzzles.
In this blog I explore two recent research papers that discuss Agentic AI and LLM-based deep search agents.
Papers Analysed
1) The Rise of Agentic AI: A Review of Definitions, Frameworks, Architectures, Applications, Evaluation Metrics, and Challenges (2025)
This Research paper discuses agent based AI systems
2) A Survey of LLM-based Deep Search Agents: Paradigm, Optimization, Evaluation, and Challenges (2026)
This Paper discusses deep search agents powered by LLMs.
1st Paper: The Rise of Agentic AI
Problems addressed by the paper
This regular AI is passive and can only respond to a single prompt at a time, it can't make decisions on its own it mainly relies on the inputs given by the user. It is unable to achieve success in a big goal on its own without the help of Human
Agentic AI
Definition
Agentic AI?
Definition: Agentic AI refers to a smart, goal-focused systems that can operate independently with very little human supervision
How is it different from Traditional AI?
Traditional AI models are passive and can only respond to a single prompt at a time
while Agentic AI can manage complex, end-to-end workflows over a longer period of time. While regular AI is limited to following strict rules for individual steps, agentic AI can connect multiple steps together, track its own progress, and automatically fix its own errors.
Core Components of Agentic AI
The paper explains that agentic AI systems usually consist of several important components.
Planning
The system first understands the main goal and breaks it into smaller tasks.
Memory
The agent stores information from previous actions and uses it to make better decisions.
Reasoning
The AI evaluates different options before choosing the best action.
Tool Usage
Agentic systems can use external tools such as databases, APIs, or search engines to gather more information. These components allow the AI to behave more like an intelligent agent rather than a simple program.
2nd Paper: LLM-based Deep Search Agents
The second research paper focuses on search agents powered by large language models.Traditional search engines require users to manually browse through many results. Users must read different pages and combine information themselves. Search agents improve this process significantly. According to the research, these agents can understand user intentions and perform multi-step information retrieval automatically. Instead of a single search query, the system can plan multiple searches and combine the results together.
How Search Agents Work
Search agents follow a multi-step process.First, the agent understands the user’s question and determines the user’s intent.Next, it creates a plan to gather information from different sources. It may perform multiple searches and analyze the results.After collecting information, the system evaluates which pieces of data are most useful and combines them to produce a final answer. This process allows the agent to generate more detailed and accurate responses compared to traditional search engines.
Connection with AI Course Topics
These research ideas are strongly connected to topics we study in Artificial Intelligence.
Intelligent Agents
Agentic AI systems behave like intelligent agents that perceive their environment and take actions to achieve goals.
Search Algorithms
Deep search agents use multi-step search strategies to gather and analyze information.
Problem Solving
These systems break down complex tasks into smaller sub-tasks before solving them.This shows how the theoretical concepts learned in AI courses are applied in modern research and industry.
What I Personally Found Interesting
While reading these papers, one thing that really interested me was how AI systems are becoming more independent.Earlier AI systems required constant human instructions. However, agentic AI systems can plan tasks, make decisions, and complete objectives with minimal supervision.Another interesting aspect is the use of search combined with reasoning. Instead of simply retrieving information, these systems analyze problems and dynamically decide what to search next.This approach could make AI systems much more useful in areas such as research, healthcare, education, and business analytics.
Challenges and Limitations
Although agentic AI is promising, there are still many challenges.
Some of the main issues discussed in the papers include:
reliability of AI decisions
ethical concerns and safety issues
difficulty in coordinating multiple agents
limitations in long-term planning
Researchers are still working on improving these systems so that they can be used safely and effectively.
Conclusion
Artificial Intelligence is evolving rapidly, and the development of agentic AI and deep search agents represents a major step forward. These technologies move AI beyond simple prompt-based systems and towards intelligent systems that can plan, reason, and act independently.As research continues, agent-based AI systems may become powerful tools that assist humans in solving complex real-world problems.The future of AI may not just be about answering questions, but about autonomous systems that can work alongside humans to achieve meaningful goals.
Acknowledgment
I would like to sincerely thank @raqeeb_26 for encouraging students to explore modern AI research and share our understanding through blogging.
Top comments (0)