Hey friends,
Last episode, I showed you my AGENTS.md, the file that teaches Codex how I like to work.
This week, I wanted to show you what happened after I gave it those rules.
I opened my recent Codex tasks expecting to see websites and other coding jobs. Those were there.
So were 285 unread newsletters, my rebuilt Notion home and a motion infographic.
And sitting at the end was a Chief of Staff that starts work before I wake up.
That’s when I realised Codex had quietly become far more useful to me than I’d planned.
Codex is OpenAI’s coding agent. This episode is about Codex, but ChatGPT Work can do many of these jobs too.
The split in my head is simple:
ChatGPT Work helps me with admin, research and thinking things through.
Codex helps me build things.
Rather than give you another giant feature list, I picked ten real jobs from my own task history.
My guess is you’ll spot at least ONE part of your own work in here.
The first one started with an inbox I’d completely ignored..
📚 Clear The Work That Piles Up
✅ 1. Catch Up On 285 Newsletters
I’d let 285 AI newsletters pile up in one inbox. At that point, reading them one by one was pure fantasy.
I put the emails into Codex. It removed repeated stories, grouped similar ideas and saved the useful parts in a library I could search later. ChatGPT Work can also read emails directly through the Gmail plugin btw.
Pro tip: Start with a date range or a set number of emails.
✅ 2. Pull The Useful Parts From A Call
Once the inbox was under control, I tried the same idea with a Fathom meeting transcript.
I wasn’t interested in another neat summary. I asked ChatGPT Work to find the exact questions, problems and moments where the conversation got interesting. It kept the full quotes and timestamps, so I could actually use them later for analysis, content, or newer messaging.
Pro tip: “Summarise this” usually gives you the least interesting answer. Ask one clear question about the call.
✅ 3. Rebuild My Notion Home
That call reminded me how much useful work was already sitting inside tools I used every day.
Notion was next.
My databases worked, but the home page didn’t help me know where to start. (I wish I was better at organizing in Notion lol).
Using the Notion plugin, Codex rebuilt my home page, navigation and databases around my real workflow. This took a few tries, but Codex can use both the Notion plugin or browser use (you need to log in) or ChatGPT Chrome extension. Either way, Notion + Codex is underrated.
Pro tip: Tell Codex to use computer/browser use if it gets stuck using the Notion plugin.
✨ Make Things I Couldn’t Make Alone
✅ 4. Turn An Infographic Into A Video
I’d already finished an A4 infographic explaining five AI connectors.
The idea was useful, but it felt like it wanted to move.
Codex used HyperFrames plugin to animate each part. Codex planned the motion in the graphic, built it in, verified everything and exported the finished MP4. No new editing timeline. No learning another video app.
Pro tip: Finish the words and layout first. Animation won’t rescue an idea that still needs work.
✅ 5. Build My Personal Site
Seeing that infographic turn into a real video gave me the confidence to try something bigger.
My personal site had been sitting on my list for ages. I wanted something I could understand and change without waiting for a developer.
Codex built each page and then it opened the site in a browser and checked the links, layout and mobile version all on its own.
Pro tip: Try to break and test the site before you call it finished. Click everything and open it on a small screen.
🛠 Make AI-Built Work Better
✅ 6. Get A Second Opinion
When you’ve watched one AI agent build something, every choice starts to make sense.
That can be a problem.
I’ll open Codex and give it the finished work from Claude Code or Cowork without all the backstory. It reviews it with fresh eyes and catches confusing parts, weak assumptions and things the first agent missed.
Pro tip: Build with one agent. Review with another.
✅ 7. Save A Process As A Skill
The next problem was repetition.
I’d explain a task perfectly, get a great result, then explain the whole thing again a week later.
Now, when I notice that happening, I use the Skill Creator. I give it my instructions and a few real examples, then test the finished Skill on completely new work.
You know how much I love skills.. so more on this later. I’ve been working on something REALLY cool for you.
Pro tip: If the Skill only works on the example used to create it, it isn’t ready yet.
✅ 8. See What’s Working In My Content
Then I started using Codex to look at the work after it went out.
Content data can keep you busy for hours without helping you decide what to make next. I connected my Threadify MCP data and asked Codex a much simpler question:
What should I do more of?
It compared the posts that earned attention with the ones that didn’t, then helped me choose what to repeat and what to test next.
Pro tip: Check where the data came from and when it was refreshed before trusting the answer.
By this point, Codex had become part of my normal workday.
The next change was being able to leave my desk without stopping the task.
⚡️ Keep Work Moving Away From My Desk
✅ 9. Continue From My iPhone
Some Codex tasks are still working when I need to leave.
With Remote inside the ChatGPT app, I can open the task on my iPhone, see what its done and answer any questions from there.
I don’t need to stay glued to my computer just because the task needs one more decision.
Pro tip: Keep approval turned on for anything that sends, publishes, pays or deletes.
Remote changed WHERE the work could continue.
Scheduled Tasks changed when it could begin.
✅ 10. Wake Up To A Chief Of Staff
This is the one that made the whole setup feel different.
Every weekday morning, my Chief of Staff starts before I do. This agent is in a self improving loop, taking in feedback every day and over time has improved dramatically.
It reads my current work, checks my priorities and prepares what needs my attention. Nothing gets posted or sent without me. I just sit down and approve, provide feedback and kick off new work the agent suggests.
I still smile when I see it working.
I used to open Codex and wonder what to ask. Now, some of the useful work has already started.
Pro tip: Run the task yourself first. Schedule it after you know the result is worth waking up to.
🤖 Start With The Job You Keep Avoiding
While reading this, one of these use cases probably made you think:
“I could actually use that.”
Start with that one.
If you created your AGENTS.md after the last episode, Codex already knows how you like to work. Now you can give it something useful to do.
Try this:
Look at the work I currently have open
Choose one job you can help me finish this week
Tell me why you chose it, then start
Keep it simple
Ask before you publish, send, pay for or delete anything
You don’t need to choose the most impressive job.
Pick the one you’d be relieved to see finished by the end of the week!
📈 The Question I Ask Now
I used to open Codex, stare at the empty box and try to invent a clever task.
Now I start somewhere much less exciting.
My actual work.
I look at what’s unfinished and ask:
Can Codex move this forward?
That one question has changed how I use it.
I’m still not a developer. I just have more ways to finish the work that used to sit around waiting for me.
And once you start noticing those jobs, ten turns into a much longer list. Trust me.
🫡 Final Thoughts
Last episode helped Codex understand how you work.
This week you have 10 use cases for Codex you can try right now.
Choose one number from this list and give it a go. Then reply and tell me where you started!
And if your first job involves 285 unread newsletters, you’re in good company.
To getting one more useful job finished,
Nahid
