What are AI agents?

An AI agent is an AI system that doesn't just answer a question. It works toward a goal: it decides what to do next, uses tools, looks at the result and adjusts, often over many steps.

Agents vs. workflows

AI company Anthropic draws a useful line between two kinds of AI systems:

• Workflows follow a path that developers set in advance. The code decides which steps run and in what order.
• Agents let the AI model direct its own process: it chooses which tools to use and when, based on what it sees along the way.

In practice an agent runs in a loop: act, get feedback from the environment, decide the next step, repeat.

What they're good at

Agents make sense for open-ended tasks where the steps can't be predicted up front, such as researching a topic across many sources or handling a phone call. Google's Gemini calling businesses on your behalf is an everyday example.

The tradeoffs

• Cost and speed: many steps mean more computing time than a single answer.
• Compounding errors: a small mistake early on can snowball over a long run.
• Safety: agents that act on their own need testing in sandboxed environments and firm safeguards. When those fail, agents can do things nobody intended, as in OpenAI's agent breaching an Australian government website.

Anthropic's advice to builders: start with the simplest setup that works, and add autonomy only when it clearly helps.