Agentic AI

Definition

AI systems that can autonomously plan, use tools, and take actions to accomplish goals.

Why It Matters

Traditional chat AI responds to one prompt at a time. Agentic AI breaks a goal down into steps, decides which tools to call, executes a multi-step plan, and adapts when individual steps fail. That shift is what makes prompts like "fix this bug across 12 files" or "book me a flight" tractable without a human supervising every action.

Key Points

• Agents loop: Observe → Plan → Act → Observe, repeating until the goal is met or a timeout fires.
• Tool calls (web search, code interpreter, file read/write, API calls) are the primitives that make agents non-trivially useful.
• Most frameworks (LangGraph, AutoGen, CrewAI) implement the loop in Python; the underlying LLM is swappable.
• Reliability drops with chain length: each step at ~90 % accuracy means a 10-step plan succeeds only ~35 % of the time without error-recovery logic.
• Human-in-the-loop checkpoints (ask before destructive actions) are the standard safety mechanism for production agentic systems.

Example

An agentic coding assistant given "add login to my app" will read the existing routes, draft the auth handlers, run tests, fix the failures it finds, and only ask for confirmation before applying changes. A single LLM call couldn't do any of that.

Related Terms

• LLM (Large Language Model)A neural network trained on massive text datasets that can generate, understand and manipulate human language. Examples: GPT-4, Qwen, Claude.
• PromptThe input text you give to an AI model. Better prompts lead to better outputs.
• InferenceThe process of running an AI model to generate a response. When you send a message to ChatGPT, the model performs inference.

Quick Facts