AI agent

When you will meet it

This is the site's central term. Understanding how it differs from a chatbot explains why the same model performs so differently in different tools.

Minimal example

你說：把這個資料夾裡所有的 .log 找出錯誤行，整理成一份報告

聊天機器人：告訴你可以用 grep 怎麼做
Agent：     自己去跑 grep → 讀結果 → 發現格式不一致 →
            換個方法再跑 → 寫出報告檔案給你看

The difference is not model intelligence; it is whether the surrounding code gives it tools and lets it retry.

What people get wrong

• Assuming more autonomy is better. More autonomy means less predictability and harder tracing. Production systems deliberately limit scope.
• Assuming the four parts are mostly AI. Only the LLM is a model; context, tools and the loop are ordinary code.