Agents and automation
What sets an AI agent apart from a simple chat, how it acts, and why supervision still matters.
完成后你将能够
- Explain what sets an agent apart from a simple chat
- Describe an agent's action-observation loop
- Explain why human oversight still matters
课程安排 — 8 个步骤
- 1An AI agent doesn't just answer: it decides, acts, and loops back.
- 2Observe the situation or the previous result.
- 3True or false?
- 4Sort each task: needs an agent, or a single question is enough?
- 5What's a reasonable guardrail before letting an agent act on its own?
- 6Tap the passage that describes a bad practice for an autonomous agent
- 7Go watch a real AI agent in action and spot its observe → reason → act loop.
- 8Agent or simple question?
建议预修 : Creating with generative AI
你的进度只保存在本设备上 — 无需账号,无需邮箱。
本页目录
1. Agents and automation
- A small blue robot next to a gear
An AI agent doesn't just answer: it decides, acts, and loops back.
- A path splitting into two branches
Observe the situation or the previous result.
Reason about the next action to take.
Act: use a tool, write text, run code.
An AI agent is always limited to answering a single question, like a classic chatbot.
The more autonomous an agent becomes, the less human supervision is needed.
When an agent uses an external tool, it's the language model itself that decides which tool to use.
An agent with full access to a bank account that can spend money with no human confirmation at all poses no particular risk.
An agent's observe → reason → act cycle was formalized in 2022 under the name ReAct.
True or false?
Sort each task: needs an agent, or a single question is enough?
Find a cheap flight, compare 3 airlines, and book it
What is the capital of Spain?
Search for a bug in the code, fix it, and propose a commit
Translate this sentence into French
Browse several sites, compare prices, and buy the best product
What's 12 times 7?
What's a reasonable guardrail before letting an agent act on its own?
Tap the passage that describes a bad practice for an autonomous agent
A coding agent that proposes a fix and submits it to a human before applying it follows good practice. An agent with full access to a bank account that can spend money with no human confirmation at all poses no particular risk. An assistant that compares flight prices and proposes a booking for the user to confirm is a reasonable use.
Go watch a real AI agent in action and spot its observe → reason → act loop.
- A path splitting into two branches
Agent or simple question?
Browse several sites, compare prices, and buy the best product
What's 12 times 7?
Plan a multi-step trip and book each part of it
Summarize this sentence in 5 words
Train a mini-classifier
Train the classifier by sorting examples, then test it on new cases.