You will be able to
- Explain the difference between chat, building, and agentic AI workflows
- Create a CLAUDE.md file that sets context, preferences, rules, and project structure for Claude Code
- Write a plain-English workflow file that describes a multi-step research process for an AI agent
- Run an agentic workflow in plan mode, answer clarifying questions, and approve the plan before execution
- Iterate on generated output by giving targeted feedback instead of starting over
- Avoid common agent-workflow mistakes such as vague prompts, skipping planning, and building too much at once
Good to know first
- Claude Code installed and used in the terminal or VS Code (covered in the creator's previous video)
- Basic comfort with navigating folders and files on a computer
- Basic familiarity with Markdown headings and bullet lists
- A general understanding of what an AI chatbot does
Syllabus
9 lessons and a final project
- 1
What makes a workflow agentic Free
You will learn the three levels of working with AI: chat, building, and agentic. You will also learn the three things that make a workflow agentic: it follows a process, it makes decisions, and it asks before it assumes.
2 exercises3 questions - 2
Setting up VS Code and the Claude Code extension Beyond Pass
You will install VS Code and the Claude Code extension, then create and open a project folder that the agent can work in. You will also learn the two VS Code features that matter most today: the live file explorer and planning mode (Shift + Tab).
5 build steps2 exercises3 questions - 3
Writing a CLAUDE.md project onboarding file Beyond Pass
You will learn what a CLAUDE.md file is and why Claude Code reads it automatically at startup. You will build one with a project context, an about-me section, rules, and a project structure, then see how it changes the agent's output.
1 code7 build steps2 exercises4 questions - 4
The three layers: workflows, agent, and tools Beyond Pass
This lesson explains the mental model behind a Claude Code agent: a workflow (a plain-English markdown recipe), the agent (Claude Code itself), and the tools it uses to get work done. You will learn why a good workflow matters more than fancy plugins.
2 exercises3 questions - 5
Designing the research workflow in plan mode Beyond Pass
You will learn how to switch Claude Code into plan mode, describe a goal instead of steps, and shape the plan before anything is built. You will also see how to add a requirement to an existing plan so the agent folds it in rather than starting over.
5 build steps2 exercises4 questions - 6
Reading the generated workflow file Beyond Pass
You learn how to open and read the workflow file that Claude Code generated, section by section. You see why a workflow written in plain English is easy to read, check and change.
5 build steps2 exercises3 questions - 7
Running the Agent and Iterating on the Report Beyond Pass
You will see how an agent runs a full workflow: it asks clarifying questions, searches the web, and writes a report file. You will also learn how to iterate, asking the agent to edit its own report instead of starting over.
6 build steps2 exercises3 questions - 8
Five common mistakes and pro tips with Claude Code Beyond Pass
You will learn the five mistakes that most often waste time when building AI agents with Claude Code, and the pro tips that help you avoid them. You will also learn how to organise workflow files, roll back a bad run and choose an effort level.
2 exercises3 questions - 9
Recap and your first workflow challenge Beyond Pass
This recap explains the difference between an agent that does exactly what you tell it and one that figures out steps on its own. It ends with your challenge: build one workflow file this week for a task you do regularly.
6 build steps2 exercises3 questions
Final project
Build, Run and Improve Your Own Weekly Research Agent
Turn one task you repeat every week into a working Claude Code agent. It has a CLAUDE.md onboarding file, a plain-English workflow file, a plan-mode design session, a real run and at least one round of targeted feedback. The point is to practise the full loop in the course: goal, plan, run, iterate.
Its 8 tasks open with Beyond Pass.
Free preview · Lesson 1 in full
What makes a workflow agentic
You will learn the three levels of working with AI: chat, building, and agentic. You will also learn the three things that make a workflow agentic: it follows a process, it makes decisions, and it asks before it assumes.
Draw and write next to the video.
The idea
Most people use AI in one of three ways. Level one is chat: you ask a question and get an answer, one and done. Level two is building: you tell the AI what to create, step by step, and it writes the code while you make every decision. In the last video, you were the director and the AI was the builder. Level three is agentic: you describe a goal, and the AI works out the steps on its own.
Think of it like the difference between micromanaging a brand new employee and trusting an experienced one. In building, you say "do this, then this, then this." In an agentic workflow, you say "here is the goal, here is what you have access to, figure it out." The AI makes a plan, picks its tools, adapts when things go wrong, and asks you questions when it is not sure.
The video says an agentic workflow has three things that make it agentic. First, it follows a process: it works through a series of steps to reach a result instead of answering one question. Second, it makes decisions: it chooses what to do based on what it finds, and if one approach is not working it pivots. Third, it asks before it assumes: it asks clarifying questions first instead of guessing what you want. The video says this one change alone makes the output much better.
Here are some examples of goals that suit an agentic workflow. You could research a topic across several sources and compile a report. You could analyse a folder of files and sort them by category. You could draft personalised emails for a whole list of contacts. You could compare competitors and produce a side-by-side breakdown. The pattern is always the same: you give the goal, and the AI does the work.
In the course you will set up VS Code and Claude Code, then give Claude Code a goal rather than a list of instructions, and watch it plan, ask, research and deliver.
Key points
- Chat is one question and one answer. Building is you directing each step while the AI writes the code. 1:49
- Agentic means you describe the goal and the AI figures out the steps. 2:21
- An agentic workflow follows a process: a series of steps to reach a result. 2:51
- An agentic workflow makes decisions and pivots when an approach is not working. 2:51
- An agentic workflow asks clarifying questions before it assumes. The video says this one change makes the output much better. 3:00
- The pattern for any agentic task is the same: you give the goal, and the AI does the work. 3:20
Words to know
- Chat
- Asking the AI a question and getting one answer.
- Building
- You direct each step and the AI writes the code.
- Agentic workflow
- You describe a goal and the AI plans, decides and works out the steps.
- Clarifying question
- A question the AI asks you before it guesses what you want.
Watch out
- Agentic does not mean the AI works with no supervision. You still set the goal and check the result. 2:43
- Describing a goal is not the same as giving a vague request. A good goal says what you need to accomplish and what the AI can use to do it. 2:07
- Asking questions first is a feature, not a sign that the AI is failing. 3:00
Practice
In your own words, explain the difference between building and an agentic workflow, using the manager-and-employee idea from the video.
Solution
- Building: you tell the AI each step, like a manager saying do this, then this.
- Agentic: you give a goal and what is available, and the AI works out the steps.
- So in building you make the decisions; in agentic the AI makes many of them.
Building means you direct every step; an agentic workflow means you describe the goal and the AI plans and decides the steps itself.
Pick a task from your own life, such as planning a school trip or comparing phone plans. Write it as a goal (what you need to accomplish), then list three clarifying questions an AI should ask you before starting.
Hint
A good goal says what you want to end up with, not each click to make.
Solution
- Write one sentence starting with 'Here is what I need to accomplish...'.
- Think of what the AI would need to know: budget, dates, preferences.
- Write three questions that cover those gaps.
A goal sentence plus three clarifying questions, such as budget, dates, and must-have features.
On your scratchpad
On a sheet of paper, draw three boxes labelled Chat, Building and Agentic. Under each, write one sentence describing who makes the decisions: you, or the AI.
Quiz
Which level is an agentic workflow?
- Chat: one question, one answer
- Building: you direct each step
- Agentic: you describe a goal and the AI figures out the steps
Show the answer
C. Agentic: you describe a goal and the AI figures out the steps · The video says at level three you describe the goal and the AI figures out how.
2:21 watch the momentWhich of these is NOT one of the three things that make a workflow agentic?
- It follows a process
- It makes decisions
- It always writes the code without asking
- It asks before it assumes
Show the answer
C. It always writes the code without asking · The three things are: follows a process, makes decisions, asks before it assumes.
2:51 watch the momentIn the video, what does an agentic AI do when an approach is not working?
- It stops and waits for you to fix it
- It pivots and tries another approach
- It repeats the same step again
Show the answer
B. It pivots and tries another approach · It makes decisions based on what it finds; if one approach is not working, it pivots.
3:20 watch the moment
Lessons 2–9 and the final project
They open with Beyond Pass: every lesson of every course, in full. One payment: $14.99 for 90 days or $29 for a year. No subscription.
Made by Beyond from the video “Claude Code: Build Your First AI Agent” by Teacher's Tech. The video plays in YouTube's own player. Watch it on YouTube · All lessons