•
Agentic AI: planning, reasoning, and acting with Autonomy
By ryu • 3 minute read •
Disclaimer
All views are educational in nature.
What is an Agentic AI
Agentic AI refers to systems that don’t just answer questions, but can plan, decide, and act on their own.
They use LLMs + Tools + Memory to handle Multi-Step Tasks without being told exactly what to do at each step.
For example:
- You say “Book me a trip to Delhi this weekend under Rs.5000.”
An Agentic AI might:
- Search for flights & buses.
- Check Hotels.
- Compare Costs.
- Make bookings via APIs.
- Email you the plan.
Now, this is way beyond just generating text - this is actual Decision + Action.
Characterstics of Agentic AI:
Goal-Oriented Planning
- It starts with a goal, then breaks it into smaller steps and executes.
Multi-Step Reasoning
- It doesn’t stop at one reply - it iteratively reasons until the goal is met.
Autonomous Decision-Making
- It chooses the best path or the tool on its own, based on the whatever context it has in that situation.
Uses Memory, Tools and Knowledge
- Remembers Past Events (long term memory or conversational memory), uses APIs/Scripts (tools), and refers to the docs or rules (Knowledge Base).
What makes Agentic AI different from Basic LLMs?
- Normal LLMs (Chat GPT, Gemini, Perplexity etc.) are Passive:
- They just respond to a prompt.
- Agentic Systems are Active:
- They observe, think, plan, act, and even revise their actions.
Basic Agentic Structure
Agentic AI VS AI Workflow
| Term | What it means |
|---|---|
| AI workflow | Pre-defined steps using AI (e.g. RAG pipeline) |
| Agent | Dynamic reasoning & acting system that adapts |
| ReAct Agent | Reason + Act: Think, Decide, take Action |
| Control logic | Like if/else to guide actions |
| Tools | External Functions/APIs |
| Memory | Past info/context retained |
| Planner | Creates a step-wise plan |
LLMs in this context
- LLMs (like GPT, Gemini, Claude) are the Core Brains of the system.
- But on their own, they can’t remember, act or plan.
- We wrap them in Agents to extend their ability.
Agentic AI Example
- AI Coding Assistant
- Can Understand bug report, edits code, runs tests, commits.
- Travel Agent
- As we have already seen the example above; There are a gazillion use-cases of Agentic AI in various domains.
- A few more examples would be an HR Onboarding Bot, An Agent that posts for you on LinkedIn, the list goes on~
Let’s recap
- LLM = Brain
- Agent = Brain + Body(Tools/Memory/Planner)
- Agentic AI = Autonomous Worker that thinks and achieves any goal given to it
- Multi-Turn Planning = Not just simple answers, but strategic answers driven by context
- ReAct Loop = Reason -> Act -> Observe -> Repeat.
Important
Before finishing up this blog; I would like for you to read this Amazing Blog by Anthropic on building Agentic AI.