Skip to content
The Internet Compass

AI

Agentic AI

Agentic AI describes systems built on language models that autonomously plan a sequence of actions, call external tools or APIs, observe the results, and adjust their next steps to pursue a defined goal — as opposed to a standard chat interaction that produces one response to one prompt.

The practical shift from a chatbot to an agent is the addition of a loop: plan a step, take an action (often via a tool call), observe what happened, and decide the next step — repeated until the goal is met or the agent determines it cannot proceed. This loop is what lets an agent handle tasks that require more context than fits in a single exchange.

Reliability, not raw capability, is the main open problem in agentic systems: an agent that's right 95% of the time per step compounds errors quickly across a ten-step task, which is why most production agent systems constrain the tools available, add explicit checkpoints, or keep a human in the loop for consequential actions.