AI Trends – Agentic AI
Agentic AI
Traditionally, AI applications serve as assistants or advisors performing tasks like recommending a product, suggesting an optimal route, advising on what can happen to an equipment or what will be the price of a particular stock tomorrow, what could be the diseases of a patient given the symptoms and lab results, and so on. In all these examples, AI acts like an intelligent assistant to humans who act upon the advice/suggestions given by AI.
The AI models today are able to think deeper, reason about the intent and context, plan a solution, and finally, act upon the recommendation. Such systems are called AI Agents who perform intelligent actions for the master (human user). They act more autonomously and often with specific instruction on how to solve a problem. Each agent can specialize on one particular task type such as collecting data, writing an email, making a reservation, etc., with support from Large Language Models (LLMs) which has high level of common skills and knowledge.
When we put together a set of such specialty agents who can coordinate among themselves, it is called an Agentic AI.
In short, agentic AI exhibits autonomy, goal-driven behavior and adaptability. The term “agentic” refers to these models’ agency, or, their capacity to act independently and purposefully.
Agentic AI systems have several advantages cover the traditional approaches such as:
Autonomous operation – ability to operate without human intervention
Planning and reasoning – ability to break down the given problem, identify the steps of solving, and discovering the resources needed.
Validating and self-correcting – ability to validate the results generated and modify it through iterative evaluation and improvement.
Basic Design Tenets of Agentic AI are
Planning: Analyzing the given problem and decomposing into smaller well-defined steps. This is typical to human approach to problem solving.
Tool Usage: AI agents are aware of the availability of tools and knowledge about which tool to use for a specific job. These tools can include web search, code execution, productivity tools, sending emails, managing calendars, etc.
Collaboration: Agentic workflows can involve multiple AI agents working together, each playing a different role, to arrive at better solutions than a single agent could typically in a single pass. The collaboration can be managed in different style – sequential, parallel, hierarchical.
Self-reflection: AI agents can examine its own work and identify ways to improve it, rather than simply generating output in a single pass. Agents can use feedback to course correct, try alternatives, and possibly learn.
References:
· What is Agentic AI - https://www.ibm.com/think/topics/agentic-ai
· What's next for AI agentic workflows ft. Andrew Ng of AI Fund -
· https://www.youtube.com/watch?v=sal78ACtGTc&t=43s
· AI Agents (IBM Tech Video) - https://www.youtube.com/watch?v=F8NKVhkZZWI