1. Quick intro:
Have you tried ChatGPT? I’m sure the answer is yes.
Have you tried Codex? I’m sure the answer is no.
Even if you did, the numbers speak for themselves: ChatGPT has 900 million weekly users. Codex has 5 million. Which is 5M/900M = 0.55%...
… for every 180 people chatting with AI, 1 person is actually putting it to work.
Source: openai.com/index/codex-for-knowledge-work
And that’s the issue. So I’ll say this as clearly as I can: you should be using Codex.
So, I decided to write this article and explain why Codex is the next level of using same ChatGPT… because as title of article says:
‘ChatGPT Can Explain Things.
Codex Can Change Things.’
For those who prefer watching video, here is my Youtube video based on this article:
2. So, what is Codex:
Let's start with the obvious: Codex — sucks.
Not the tool.
The name.
Say "Codex" and people hear "code," and instantly half the audience checks out: not for me, I'm not a developer.
Me too! And trust me — Codex is for non-developers first and foremost.
(But honestly, OpenAI has always been bad at this. Remember GPT-4.2, 3.5, mini, turbo, o1, o3? Nobody could keep them straight. So — Sam Altman, if you're ever shopping for a naming tool, I hear Claude's pretty good at that 😂)
So. Codex is OpenAI's AI agent for building and changing digital things. Technically it's a coding agent — most powerful in an IDE or terminal. But for the rest of us, that just means it builds the things code powers: websites, forms, dashboards, internal tools, scripts, reports, automations, workflows. And let me tell you the secret, my friends developers confimrs that majority of their time they also do these type of tasks, not coding.
Source: scaleyourweb.com
ChatGPT = CHAT | CODEX = AGENT |
|---|---|
A normal AI day: a question comes up, you open a model, paste context, get an answer, copy it back into your real work — and then personally carry the result the last mile. Rename the file. Check the source. Send the email. The model helped, but the loop still lives in your head and hands. | A Codex day: you open the actual work — the folder, the repo, the spreadsheet, the transcript — and give the agent a bounded job inside it. What it produces — the updated file, the fixed page, the report — becomes the starting point for the next job. Nothing depends on you remembering what you meant |
Codex can do everything ChatGPT can do,
BUT
ChatGPT can’t do everything Codex can.
3.1. Codex Interface (Level-1)
Good news: if you find ChatGPT easy to use… it's super easy to learn Codex — same chat, same way of choosing models, same idea of speed vs. intelligence.
So, in my opinion, that is it — that's all you need to start testing. The rest you'll learn by doing. Codex tells you what it needs and why as you go.
But if you're curious about the more advanced features, here's a handy list of symbols you can use to work with it more efficiently, and I personally love using it:
Symbol / shortcut | Correct meaning |
/ | Slash commands. Use it for commands like /plan, /review, /status, /init, /mcp, /goal, etc. |
@ | Mention or attach context, such as files, apps/plugins, and in some places skills. |
$ | Explicitly call a skill, for example $skill-name. |
/side | Opens a side conversation without disrupting the main thread |
Press both Command keys | Takes an Appshot on macOS: Codex captures the frontmost app window screenshot plus available text and sends it to a Codex thread. |
Cmd + D / Cmd + E | Codex app dictation shortcut. Hold or press while the composer is visible, speak, then edit/send the transcription. |
3.3. Codex Interface (Level-2)
Let's set up Codex so it works the way you want.
And here's the proof it's not just for developers: there are actually two modes. One for technical work, and one for everyday work — same power, less technical detail. Developer? Use the first. Everyone else? Use the second.
Personalization & Memory (experimental) — this is Codex getting to know you. Turn it on and it starts remembering how you work: your preferences, your projects, the way you like things done — so you stop repeating yourself every session. It's still experimental, so it won't be perfect, but it's the difference between a tool you re-explain every time and one that already knows the drill.
There are a lot more settings to play with, but those are best learned through real use, so I will show them in our Use Casees section. Here I'll just walked you through the basic ones I personally recommend setting from the start.
3.3. Codex Interface (Level-3)
Plugins (MCP & connectors)
This is where Codex stops living in a box. Connect it to the tools you already use — your Drive, your database, Slack, Notion, GitHub, Figma, Gmail, etc… — and it stops guessing about your work and starts touching it. The agent is only as powerful as what you plug into it. Plug in a lot.
Skills
Think of skills as muscle memory you hand the agent once. Instead of re-explaining how you like your reports formatted, your emails written, or your code structured every single time, you teach Codex once — and it remembers the how, not just the what. The more you use it, the more it works like you.
Automation
Here's the payoff. Everything above was you asking and Codex doing. Automation removes the asking. Set it to run on a schedule or a trigger — every morning, every new file, every incoming email — and the work happens while you're not even looking. The Monday report writes itself. The inbox sorts itself. The thing you kept forgetting to do, gets done. This is the line between using a tool and owning a system.
USE CASE 1: Work with local files (any type), OCR, dashboard
Imagine I'm preparing my tax report. I've got receipts scattered everywhere — a few phone photos (.jpg, .png, .heic), some PDFs I downloaded from Gmail, and plenty of a .csv exports.
Instead of building a category table by hand and typing every line in myself, I just point Codex at the folder and ask:
In the folder /receipts there are mixed files — PDFs, images, and a CSV — all receipts from different dates and categories. I'd like you to:
1. Read every file and extract the date, amount, vendor, and category from each receipt.
2. Build a clean, good-looking HTML dashboard that shows spending by category and by month.
3. Add manual filters for month and category so I can choose what to display.
4. Include clear charts (a category breakdown and a monthly trend).
If anything is unreadable or ambiguous, list it separately so I can review it myself.The result I got in 4 minutes:
So, Codex easily easily works with your local files like .JPG, .PNG, .CSV, read them, categorize and create an unified table, so that you can work with it…
Codex builds you a clean, readable dashboard — like an assistant you asked to put together a presentation on exactly where your money goes.
So, the takeaway from use case #1:
Codex works with local files in any format. Here I tested PNG, JPG, PDF, and CSV — but trust me, I've thrown all kinds at it (HTML, Python, OBJ, you name it) and it reads them all.
Codex uses OCR to pull data straight out of images.
And it turns that raw pile into a dashboard that points your attention at what matters — cutting the noise so you're not scanning irrelevant rows yourself.
USE CASE 2: Works with your cloud files — and automates
I already said it: with ChatGPT you ask, you get an answer, and then the copy-paste game begins — which is pointless. Instead, you've got an agent that does it for you. No copy-pasting — MCP connections instead. No sticky-note reminders — automation instead.
So let's build one:
Create a native Google Sheet in my Drive called "AI Headlines Automation" with a Headlines tab and these columns: Timestamp, Title, Source, Link, and an importance score for the AI industry.
Then set up a Codex automation that runs every 10 minutes, collects 5 fresh AI headlines, and appends only new rows to that tab.
Verify the sheet after it's created.And guess what — the screenshot below shows the automation after it ran, and you can even watch it running in real time:
The result? Exactly what I asked for — completely on autopilot. Codex created the spreadsheet in my Google Drive, built the table, set up the automation, and updates it every 10 minutes. Exactly as asked.
So Codex doesn't just touch local files — it reaches into your cloud, runs a web search, and keeps automations running while you sleep.
