Curriculum

Topic 07 · Autonomous Systems

Agentic AI

Agentic AI systems autonomously pursue goals across multi-step tasks, calling tools, managing memory, and adapting their strategy based on environmental feedback.

The idea

An AI agent is any system that perceives its environment and takes actions to achieve a goal. Modern LLM-based agents implement a loop: Observe → Think → Act → Observe. The LLM acts as the "reasoning engine", while tools (web search, code execution, APIs) extend its capabilities beyond language.

Key paradigm: ReAct (Reason + Act). The LLM generates a "thought" (reasoning trace) followed by an "action" (tool call), then receives an "observation" (tool output), and iterates until the task is complete. This pattern is the basis of LangChain, CrewAI, AutoGen, and the OpenAI Assistants API.

aₜ = π(oₜ, hₜ) hₜ₊₁ = memory_update(hₜ, oₜ, aₜ)

Agent policy π maps current observation and memory to action. Memory updates track interaction history.

What to hold onto

  • Memory TypesEpisodic (conversation history), semantic (fact store), procedural (workflow templates), working (current context window).
  • Tool CallingLLMs structured to output JSON specifying a function name and arguments. The runtime executes the function and returns results. Foundation of GPT function calling and Claude tool use.
  • Agentic LoopsSingle-agent (simple ReAct). Multi-agent (orchestrator delegates sub-tasks to specialist agents). Hierarchical (nested orchestration trees).

Live: ReAct Autonomous Agent Task Loop

Interactive state machine cycling through Observe $\to$ Reason $\to$ Act $\to$ Reflect.

State: [THOUGHT]

Quick answers

What is Agentic AI?

Agentic AI refers to AI systems that autonomously pursue goals through multi-step reasoning and action. Powered by LLMs, they follow an Observe-Think-Act loop, calling external tools (search, code execution, APIs) and maintaining memory across steps to complete complex, open-ended tasks without human intervention at each step.

What is the ReAct framework?

ReAct (Reasoning + Acting) is a prompting framework for LLM agents where the model interleaves reasoning traces (thought) with action calls (tool invocations). The resulting observation is fed back into the next thought step. This interleaving improves factual accuracy and task completion over pure reasoning or pure acting alone.