What is Agentic AI?

Traditional LLM applications often follow a linear chain: Prompt -> LLM -> Output. Agentic AI breaks this linearity by introducing a control loop where the model decides what to do next.

The Control Loop

An agent typically operates in a loop:

1. Observe: Read input or environment state.
2. Reason: Decide on an action (using an LLM).
3. Act: Execute the action (call a tool).
4. Reflect: Observe the result of the action.
5. Repeat: Continue until the goal is met.

Agent vs. Chain

Feature	Chain (e.g., LangChain Runnable)	Agent
Control Flow	Hardcoded by developer	Determined dynamically by LLM
Flexibility	Rigid, predictable	Adaptive, handles ambiguity
Failure Recovery	Often brittle (fails if one step fails)	Can self-correct and retry
Complexity	Lower	Higher (requires guardrails)

Why Sruja for Agents?

Modeling agents is complex because relationships are often dynamic. Sruja helps by:

• Visualizing Dependencies: Showing which agents use which tools.
• Defining Boundaries: separating the cognitive engine (LLM) from the execution layer (Tools).
• Documenting Flows: Tracing the decision loop.

import { * } from 'sruja.ai/stdlib'


Agent = component "Research Agent"
LLM = component "Model Provider"
Tool = component "Search Tool"

Agent -> LLM "Reasons next step"
Agent -> Tool "Executes action"
Tool -> Agent "Returns observation"

view index {
include *
}