What are AI Agents?

The terms agents, agentic AI, agentic development, etc. get used a lot in discussions today. I first paid attention to it after listening to DHH: Future of Programming, AI, Agentic Engineering, Vibe Coding & Linux. Then I started understanding what it was after listening to Cal Newport’s discussion of GPT 6.

At their core, AI agents are prompt loops. At each iteration, the agent assembles context from available inputs, invokes an LLM to reason and select an action, executes that action, observes the outcome, and feeds the observation back into the next iteration. This process repeats until the task is complete or a defined stopping condition is reached. (Copied verbatim from What is the AI Agent Loop?). This is the perceive → reason → plan → act → observe loop.

While this loop marks the essence of an agent, various engineering details (such as context management, stop conditions, verification, memory, state management, multi-agent coordination, cost control, etc.) are critical for making the agent performant.