Codex 101 Speaker Script

Hi everyone, I am Brian. Welcome to Codex 101.

Today is not really a normal ChatGPT prompting class. The main thing I want you to understand is that Codex is not just a chatbot, and it is not only for writing code.

The simplest way to think about it is this:

ChatGPT is useful when you want an answer.
Codex is useful when you want to move work through tools.

That work might be code, but it can also be a Notion page, a brief, a portfolio website, a checklist, a research summary, a design prompt, or a review.

Codex is more like a workbench.

A workbench has materials, tools, rules, and a review step.

For Codex, the materials are things like files, notes, links, screenshots, folders, docs, or project context.

The tools might be browser, Notion, GitHub, terminal, image generation, Figma, Canva, Google Drive, or other plugins.

The rules are the boundaries we give it: what it can read, what it can edit, what it must not touch, and when it should ask for permission.

And the review step is where we check the output before accepting it.

The mistake people make is thinking Codex is magic automation.

I think the safer view is: Codex is a capable assistant, but the human still owns the goal, the judgment, and the final approval.

So the skill today is not, how do I write a perfect prompt?

The skill is: how do I set up a good workflow?

Codex has a few things that make it different from a normal chat interface.

First, threads. A thread is like a work room for one task. If I am planning a workshop, that can be one thread. If I am building a website, that can be another thread. Keeping work separate makes Codex much easier to steer.

Second, context. Codex gets better when you give it the right material. That could be a folder, a Notion page, a link, pasted notes, a screenshot, or an Appshot.

Third, tools and plugins. This is where Codex starts feeling different from ChatGPT. It can work with real systems, not just generate text.

Fourth, review. Codex can create things, but we should always ask what it did, what it changed, what it accessed, and what still needs human verification.

When Codex works, it usually follows a loop.

You give it a prompt. It reads context. It uses tools. It updates its understanding. It creates or changes something. Then it checks and reports back.

The important thing is that every action adds context. So if the initial context is messy, the output can be messy. If the context is clear, the result is usually much better.

This is the prompt structure I want you to remember:

• Goal: what do I want?
• Context: what should Codex know?
• Tools: what can it use?
• Constraints: what should it avoid?
• Done when: how do we know it worked?
• Review: what should it check before saying done?

If the task is fuzzy, do not ask Codex to immediately act. Ask it to plan first. Plan mode is good when you want Codex to think through the task before changing anything.

The biggest unlock is validation.

If Codex knows how to check success, it can improve its own work.

For code, validation might be tests. For a website, it might be opening the page in the browser. For a Notion guide, it might be checking readability and privacy. For a portfolio, it might be reviewing claims and making sure nothing is invented.

No review, no trust.

Now, the slides are just the concept layer. The hands-on part is in the Notion guide.

Think of the Notion guide as your workbook.

It has practice stations you can choose from. You do not need to finish every station. Pick one that matches your comfort level.

If you are newer, start with the profile-to-portfolio brief. If you are comfortable with tools, try the Notion, image, Canva/Figma, or website station. If you are technical, pay attention to how Codex scopes work, verifies output, and explains what changed.

For the hands-on portion, we will use a simple workflow:

profile material -> portfolio brief -> visual direction -> website draft -> review before sharing

You can use your own public profile, your own resume text, your own personal website, or the fake sample profile in the Notion guide. If you are unsure, use the fake Maya Tan profile. That is the safest option.

One important rule: do not paste private information.

Do not paste private phone numbers, private emails, private messages, salary information, references, or internal company material.

Also, do not let Codex invent achievements. If something is not in the source material, Codex should mark it as missing or uncertain.

Open the Notion guide and go to the practice stations.

Start with Station A: turn profile material into a portfolio brief. The goal is not to make the perfect portfolio yet. The goal is to see Codex take source material, structure it, identify what matters, and tell you what needs verification.

Before you accept anything Codex creates, use the review prompt in the guide.

Ask Codex:

• What tools did you use?
• What information did you access?
• What did you create or change?
• What should I verify?
• Are there privacy or permission risks?
• Is this ready to publish, or only ready to review?

That review step is the difference between using Codex casually and using Codex responsibly.

The main takeaway is this:

Codex is not about prompting harder.

Codex is about setting up a better workbench.

Clear source.
Right tool.
Concrete artifact.
Safe boundary.
Review step.

Use ChatGPT when you want a quick answer.

Use Codex when you want to move work through context, tools, artifacts, checks, and review.