Autonomous Agents

An autonomous agent is an AI system that pursues a multi-step goal without waiting for a click or a prompt at every turn. Ordinary software waits for a button. A chatbot like ChatGPT waits for the next sentence. An agent keeps going until the job is done or it hits a limit you set.

Ask a language model for a recipe and it prints a recipe. That is the whole job. Ask an agent to organize a dinner party for Friday and it breaks the goal into smaller tasks. It checks your calendar. It emails friends for times that work. It searches for a restaurant with strong reviews. It uses a wallet or an API to book a table. You can be asleep while that loop runs.

The loop is simple. You give an objective. The agent uses an LLM as a reasoning engine to pick the first step. It calls tools: a browser, a calculator, a database, or another model. It looks at the result, judges progress, and writes the next step. Reason, act, observe. Repeat until the goal is met or the budget, permissions, or time run out.

This is different from a script. A script follows a fixed path. An agent chooses the next action from the current state. That is useful when the world changes: the restaurant is full, the calendar has a conflict, the email bounces. It is also how agents fail. They loop. They book the wrong night. They spend money. They treat a hallucinated address as real.

The work changes when this is reliable enough to use. People spend less time driving the software and more time setting the objective, the tools, and the stop conditions. The useful design is a clear goal, limited tools, logs you can read, and a human who can halt the run. The standard AI textbook treats autonomy as acting without a human in every loop.

Today's coding and browsing agents are a practical version of that idea.