Loading...
Loading...
Loading Curriculum...
Loading Subject...
Loading Topic...
Loading Lesson...
Loading Lab...
An AI agent uses a language model as a reasoning engine to plan, use tools, and take actions toward a goal — not just produce text.
01 — Definition
A traditional generative AI model takes a prompt and returns a response. An agent goes further: it pursues a goal over multiple steps, decides which tools to use, observes results, and adapts.
Think of the LLM as the brain, tools as the hands, and memory as the notebook. Agents close the loop between thinking and doing.
02 — Anatomy
Most agents are built from four building blocks working together.
An LLM that interprets goals, plans steps, and decides what to do next.
Inputs from users, documents, APIs, or sensors that ground the agent in context.
Functions, APIs, and code the agent can call to take real actions in the world.
Short and long-term storage for context, past actions, and learned preferences.
03 — Lifecycle
Every agent runs some variation of a perceive–plan–act–reflect loop.
01
Receive a goal and gather relevant context.
02
Break the goal into ordered, actionable steps.
03
Call tools, run code, or query data sources.
04
Evaluate results, adjust, and iterate until done.
04 — Taxonomy
Respond to a single prompt with a tool call. Simple, fast, predictable.
Plan multi-step workflows and adapt their plan based on intermediate results.
Specialized agents collaborate, delegate, and review each other's work.
05 — In practice
Browse, summarize, and synthesize sources.
Read repos, write patches, and run tests.
Resolve tickets by querying internal systems and applying fixes.
As reasoning improves and tools get richer, agents will quietly handle more of the workflows that used to require humans clicking through software.