Sign in
Capabilities

Agentic Capabilities

LLMs running in loops with tools to complete multi-step tasks autonomously

What it is

Agentic AI refers to LLMs operating in an autonomous loop: receive a task, use tools to gather information, take actions, observe results, and continue until the task is complete or the model determines it's stuck. This enables tasks that can't be accomplished in a single prompt.

A coding agent, for example, might: search the codebase to understand structure, write a function, run tests, read the error output, fix the bug, and repeat until tests pass. Each step involves at least one LLM call and one or more tool invocations.

Key challenges in agentic systems: error recovery (what happens when a tool fails), context management (long tasks exceed context windows), and reliability (one bad decision early can derail the whole task).

Why it matters

Agents are the primary reason AI labs are investing so heavily in capabilities right now. They extend what a single model call can accomplish by orders of magnitude. You'll be building with agentic frameworks (LangChain, LlamaIndex, Claude's agent SDK). Understanding the loop model, tool calling, and failure modes is the foundation for building anything non-trivial with AI.

Related concepts

Resources

Anatomy of AI Agents: Inside LLMs, RAG Systems, & Generative AI
youtube.com· Breaks down the internal components of AI agents (LLMs, RAG systems, and generative AI) and how they work together. Covers the perceive-reason-act loop, memory, planning, and tool use.
10 min
Intro to Large Language Models
youtube.com· The visionary "LLM as operating system kernel" talk. Covers tool use, customization, self-improvement, and multimodality as emergent agent capabilities. Prescient 2023 talk that predicted the 2025 agentic wave.
60 min
What Are Agentic Design Patterns?
deeplearning.ai· Andrew Ng's framework for understanding agents through four design patterns: reflection, tool use, planning, and multi-agent collaboration. Clear and authoritative.
8 min
What Are Large Language Models?
ibm.com· Solid section on how LLMs are extended into agents with memory, APIs, decision logic, and external systems. Explains the distinction between plain text generation and agentic task execution.
15 min
AI Agents in LangGraph (Intro lesson)
learn.deeplearning.ai· Andrew Ng + Harrison Chase discuss what makes an AI agent: function calling, tool use, memory, multi-step reasoning. Good conceptual intro even without doing the full course.
10 min