Summary of "What is Agentic AI? Important For GEN AI In 2025"

Agentic AI — What it is and why it matters for GenAI in 2025

Core concept

Agentic AI = autonomous, goal-driven multi-agent systems (AI agents) that execute complex workflows without human intervention to achieve business outcomes.

Agentic AI differs from generative AI: generative models focus on producing content from prompts; agentic systems compose, orchestrate, and execute tasks by invoking tools, APIs, and multiple models to reach objectives and adapt over time.

Key technical ideas

Agents are independent modules that can call external tools (search, databases, finance APIs, payment gateways, etc.), other agents, and LLMs as needed. Multiple agents coordinate in a workflow to complete larger goals.
Typical workflow pattern:
1. Define a goal.
2. Dispatch agents/tools.
3. Gather data (web/news, yfinance, databases).
4. Analyze/compare results.
5. Return an action or recommendation.
RAG (retrieval-augmented generation) vs agentic workflows:
- RAG supplements LLMs with external context.
- Agentic systems go further by autonomously choosing and running tools and sequences of actions.
Agents can iteratively improve their own behavior during task execution (self-improvement / iterative refinement).
Open integration across many LLM providers: OpenAI, Google Gemini, Anthropic, Mistral, Grok, AWS/Bedrock, Claude, etc.

Tools, frameworks, and product features

Frameworks and tooling:
- LangChain (tools concept)
- LangGraph
- LangFlow (no-code / drag-and-drop visual workflow design; export Python/JS; deploy endpoints)
- Microsoft Autogen (open source)
- F data (Fdata) — open-source framework for creating many domain agents (financial, legal, marketing, construction); integrates any LLM and many external tools/APIs
Common tool integrations used in demos:
- DuckDuckGo search, Wikipedia search
- yfinance (stock data), news APIs
- Payment gateway integrations
Product features emphasized:
- Multi-LLM support
- Tool integrations and the ability to combine agents (e.g., web-search + financial agent)
- Code export and local endpoints (demo ran on localhost:777)
- Playground UI to test agents and quickly prototype

Demo / tutorial content (from the video)

Live demo (F data playground) built agents:
- Web search agent: DuckDuckGo + OpenAI chat to fetch recent news.
- Financial agent: yfinance + OpenAI chat to fetch and analyze stock data.
- Combined workflow: compared Tesla vs Nvidia and returned a buy/hold recommendation (example conclusion: Nvidia stronger contender; buy recommendation).
Creator plans:
- Convert an existing generative AI mock-interview platform into an agentic application.
- Produce more end-to-end framework tutorials and build agentic apps in 2025, with step-by-step guides starting from basics.

Implications / industry notes

Agentic AI is gaining traction and is expected to be a key skill area in 2025 for developers building GenAI applications.
It may enable new classes of software—potentially competitive with traditional SaaS—by automating business workflows end-to-end.
Many frameworks are open-source and rapidly evolving, presenting opportunities for experimentation and productization.

Main speakers / sources

Speaker: Krish (Krish Naak) — YouTuber who presented the video and demoed the F data playground.
Frameworks / products mentioned: LangChain, LangFlow, LangGraph, Microsoft Autogen, F data (Fdata).
Models/providers mentioned: OpenAI, Google Gemini, Anthropic, Claude, Grok, Mistral, AWS/Bedrock.
Tools/APIs mentioned: DuckDuckGo search, Wikipedia search, yfinance, various news APIs, payment gateway integrations.