Introduction
Artificial Intelligence is moving beyond simple rule-based systems towards agents that can plan, reason, and act autonomously. Present-day AI systems research is focused on building artificial agents capable of performing a variety of real-world tasks.
In this blog, I will share my findings from two very recent Artificial Intelligence research papers and relate them to concepts that would be studied in an Artificial Intelligence course: intelligent agents, search, and heuristics.
The Artificial Intelligence research papers used are:
"The Rise of Agentic AI: A Review of Definitions, Frameworks, and Challenges (2025)"
"Research on the A* Algorithm Based on Adaptive Weights (2025)"
The first paper discusses agentic artificial intelligence, a newer approach to AI where agents are able to perform tasks autonomously after independent planning.
The second paper discusses improvements to the A* search algorithm, one of the most commonly utilized algorithms for pathfinding and optimization tasks in Artificial Intelligence.
Both research papers highlight the continuing advancement of artificial intelligence from foundational principles that we learn in artificial intelligence courses and build on them in ways that allow more efficient systems to perform complex real-world tasks.
What is Agentic AI?
The first research paper focuses on agentic AI. This is a system of artificial intelligence that is autonomous, able to perform a variety of different tasks, and has the ability to plan.
Whereas previous AI agents could receive input and respond or predict information, agentic AI has the ability to respond and then perform several different actions.
Intelligent agents within the field of AI are typically described as agents which can:
perceive their environment
reason with information
act to achieve certain goals
Agentic AI further defines this as agents which can perform multiple steps of planning. For example, an agentic system could:
search for information
evaluate various sources
use the information to create an appropriate response
This would show a significant advancement from simple AI, which may just receive information and generate a simple answer, to an autonomous problem-solving system.
Frameworks Used in Agentic AI
Several frameworks have been developed in Agentic AI for creating agentic systems. The most important types have been discussed in the paper and can be briefly summarized below:
- Planning-Based Agents
These agents are the most direct implementation of what we learn in artificial intelligence courses about goal-based agents. They break tasks into smaller goals which are solved sequentially. As an example, an agentic system trying to compile a report would have goals of information retrieval, analysis, synthesis, compilation, writing, and presentation.
- Tool-Using Agents
Agentic systems are often designed with external tools in mind, such as Application Programming Interfaces (APIs), search engines, and other specialized tools for complex problems. This can expand the capabilities of an agentic AI system far beyond what its internal parameters or learning would provide.
- Reasoning & Action Loop
This refers to an iterative process of decision-making for an agentic AI system. At each step, the system would analyze the state of the world it perceives, decide on a course of action, then execute that action. The results are then analyzed and fed back into the cycle in order to adapt to the evolving world and task objectives.
Challenges with Agentic AI
As has been indicated throughout, Agentic AI represents an enormous leap in the field of Artificial Intelligence. However, there are some obvious challenges presented in the research paper.
Reliability
If an agent makes the wrong assumptions about what actions need to be taken or what data is needed, the agentic system could produce unreliable outputs.
Safety
Given that the agentic AI can automatically execute actions, there needs to be some safety protocol or control mechanism to ensure that it does not do something damaging.
Long-Term Planning
While simple actions are already an advancement for AI agents, carrying out long-term sequences of actions and maintaining context are areas which need a lot more research.
The challenges are presented clearly, demonstrating the extent to which agentic AI is still a new field which requires research to develop to its full potential.
Research on Adaptive A*
The second paper I reviewed concerned advancements to one of the most well-known and extensively utilized search algorithms in artificial intelligence, the A* search algorithm. The research has focused on improving the performance and efficiency of A* with adaptive weightings.
A* is used to solve optimization problems where the path has to go from A to B and is most commonly seen in games, navigation systems, logistics, robot navigation, and much more.
The A* algorithm finds the path by using the function f(n)=g(n)+h(n), where g(n) is the actual cost from the start to node n and h(n) is the heuristic estimate from the node to the goal node.
The A* algorithm prioritizes searching nodes with the lowest f(n) value, resulting in optimal search and efficiency provided the heuristic is good enough.
In the paper I researched, they implemented an adaptation to the A* search algorithm using a varying weight. Traditionally, a heuristic has a weight of one and does not change. This proposed algorithm utilizes a variable heuristic where the weight assigned by the algorithm to the heuristic can be modified during the search in order to balance the search between finding optimal and feasible paths. They discovered that the algorithm would favor heuristic results at the beginning of the search (giving it more room to explore), and when approaching the goal, it would favor the actual path to the goal in order to try to find a solution quickly and with minimal searching.
Benefits of the Adaptive Approach to A*:
Faster path identification
A reduction in nodes searched
Improved efficiency
Conclusion
The papers have offered two differing insights into the field of AI and artificial agents, but both show the progression that we are making. Agentic AI shows a step towards machines which have greater independent decision-making and can carry out tasks with much less interaction from users, whereas the improved A* search algorithm simply increases the speed of AI agents which utilize pathfinding. In essence, both papers are proving that the AI we have today is ever getting more capable and adaptable.
YouTube Video Link
Special Mention
Top comments (0)