|
|
The most powerful versions of AI live beyond the ChatGPT.com or Claude.ai homepage. |
How most people use AI: they ask, it answers, then they go and act on the advice themselves. |
Like a search engine, that's a useful way to get information -- no knocking it. But unlike a search engine, you're only scraping the surface of what AI can do. |
Moving from the default chat window to using Codex (made by OpenAI) or Claude Code (made by Anthropic) is one of the biggest steps a beginner can take to unlock real leverage and graduate to intermediate power user. |
The best part is that if you pay for access to ChatGPT or Claude, you likely already have access to these tools. And if you’re still using a free plan, what I have written below just might give you the gentle nudge you need to try upgrading -- all I ask is that you read it with an open mind. |
So today, I want to break down: |
What are Codex and Claude Code -- and why you should try them this weekend 3 prompts you can steal to get immediate value on your subscription Why it’s easier than ever to learn to use these apps today
|
—@dharmesh |
|
|
A Beginner’s Guide to Codex and Claude Code |
You can secure a quick win right now by introducing yourself to Codex or Claude Code. Simply moving your day-to-day AI usage into these apps is a step in the right direction. |
Codex comes from OpenAI, and Claude Code from Anthropic. Beyond the new name and slightly different interface, everything (more or less) works exactly the way you’re used to. You still chat to it in plain English. Under the hood, it's still GPT or Claude, the same models that sit behind the chat window you’re well acquainted with. |
What changes is where the AI is allowed to work. |
In the chat window, it works at arm's length: you bring it things, it hands things back (a summary, some code, a draft), and then you go put them to use yourself. |
But when you use Codex or Claude Code, the work is done inside your actual files -- you point the app at a real folder and the AI sets up shop. It makes changes to files within that folder, runs things inside, and otherwise does everything you instruct it to do, without you copying results between windows. |
The difference between dragging-and-dropping a few crumbs of context and your agent working directly alongside all the context you have is massive. |
(Sidebar on safety best practices: For your first few runs, select a copy of your real work folder, or a folder you won't miss. At least until you've seen how your agent behaves. Better safe than sorry!) |
So how do you get started? |
You don’t have to jump right into using a command line interface. Start by using the official apps: Codex has its own dedicated app, separate from ChatGPT, whereas Claude Code is built right into the Claude app. These are very powerful apps -- while developers still typically favor using their go-to coding interfaces, it’s not uncommon for even technical folks to dip in and out of using the official Claude Code or Codex app. |
|
|
3 Practical Prompts to Steal Yourself |
With Claude Code or Codex installed, you’re now equipped to raise the ceiling on the potential of your AI workflows. Exciting! |
Here are a handful of leverage unlocks you may be interested in tinkering with: |
Create persistent context |
If you find yourself constantly re-explaining the same facts (who you are, what your company does, what you're trying to accomplish) in every conversation you have with AI, this one’s for you. |
Both Claude Code and Codex read markdown files at the start of every session. You can store key context inside these files to (1) make sure every conversation is aware of the information written inside and (2) to tailor the way your AI interacts with you. |
The key files function nearly identically but differ by name. There is CLAUDE.md for Claude Code and AGENTS.md for Codex. |
Try this prompt: “Update my (CLAUDE.md or AGENTS.md) file to include a new rule. The rule is that you should always end each message with a simple dad joke.” |
Create autonomous loops |
Last week, we built our first loop inside a chat window by giving the AI an objective, a metric to measure progress, and a boundary to know when to stop. |
Codex and Claude Code have a built-in /goal feature to help you create loops that are even more powerful and customizable. I’ve seen loops that run for over 24 hours when given an ambitious goal! |
Try this prompt: “Goal: Rename every file in this folder so the name starts with the date it was created, then a short topic. Done when: every file is renamed that way and nothing is left with an old name.“ |
Create real artifacts that exist outside of the chat window |
You may have used Claude's Artifacts or ChatGPT's Canvas before. A chart, a doc, or a small app appears in a panel beside the chat. Handy, but the artifact lives inside the conversation. It can't reach your real files, and once you scroll on, it stays behind. |
Codex and Claude Code produce actual files on your computer (a spreadsheet, a script, a working web page) that you own, open in your usual apps, and run again whenever you want. |
Try this prompt: “Build a beautiful standalone HTML page that helps me set my top 5 goals for this month, each with a checkbox I can tick off, give it a clean look. Save it as a file I can open in my browser.” |
|
|
Start Small |
You don't have to overhaul how you work to use Codex or Claude Code. Simply using these apps exactly how you’d use their simpler cousins puts you on the right track. |
Put yourself in the right place enough times and naturally you’ll get more comfortable, try a few more things, and your AI expertise will grow accordingly. |
The nice thing about the models we have access to today is that they can onboard us themselves. If you’re unsure about something, just ask Claude or ChatGPT for help. The learning curve for software this powerful has never been flatter. |
I leave you (until next week!) with a question: How do you interact with AI most? ChatGPT.com? Claude Code in the Terminal? |
Hit reply and let me know -- I read every response, and I’m eager to test my assumptions on this one! |
—Dharmesh (@dharmesh) |
|
|
|
|
What'd you think of today's email?Click below to let me know. |
|
|
