Testing OpenAI Codex on Mobile

Last Thursday night, while browsing TechWeb, I saw the news that OpenAI launched Codex for mobile, available for both iOS and Android users, allowing remote access and management of development workflows. I thought this was perfect for a programmer like me who often needs to fix bugs on the go. The next day, I downloaded it, and my experiences over the next three days nearly got me fired on Monday.

Day One: The First Mistake

On Friday, I was supposed to take a day off to catch up on sleep. However, at 9 AM, my product manager called, saying there was an urgent bug where the shopping cart data wouldn’t refresh after user login. I was still in bed, with my laptop charging in the living room, and didn’t want to get up. Suddenly, I remembered the Codex mobile app and thought it would be a great opportunity to showcase new technology.

I opened the ChatGPT mobile app, found the Codex entry, and connected to my company’s testing environment. The interface was quite simple: a file tree on the left and a code editor on the right. I located the login interface file and started coding on my phone. Honestly, coding on a phone is quite challenging—the virtual keyboard is too small, I often hit the wrong keys, adjusting indentation takes ages, and matching brackets relies solely on my eyesight. It took me 40 minutes to write what usually takes 10 minutes.

After deploying the fix, I sent a message in the group: “Bug fixed, coded on mobile, Codex mobile is awesome!” I even included a screenshot. Half an hour later, a tester tagged me in the group: “Zhang, the logic you wrote has a problem. When a user logs out and logs back in, the shopping cart data gets cleared.”

I was stunned. I quickly got up to check on my laptop—turns out I had written the if condition backward. It should have been if (!user.loggedIn), but I wrote if (user.loggedIn). Such a simple mistake caused chaos in the entire testing environment, and operations spent two hours restoring the database backup.

That afternoon, my technical director called me into the office and scolded me: “Are you bored? Can you really code on a phone? Is what you wrote even usable? If I catch you coding production code on your phone again, you can just leave!”

Day Two: A Bigger Mistake

Originally, I thought that was the end of it, but on Saturday, I made an even bigger mistake. I was out having dinner with friends when operations shouted in the group that an interface timed out and needed a cache added. After a few drinks, I impulsively pulled out my phone and opened Codex again. This time it was worse; after making the changes, I forgot to add a try-catch block. The cache server crashed, resulting in a 500 error for the entire interface. That night, I spent over an hour in the restaurant restroom rolling back code on my phone, and my legs went numb.

Day Three: Testing Usability

On Sunday, I stayed home, afraid to touch my phone for coding. But I was still curious about the tool’s practical value. I found a personal project, a web scraping script written in Python. This time, I avoided complex logic and just added a simple logging feature.

To be fair, the code suggestion feature of Codex mobile is quite powerful, much better than other mobile editors. It can auto-complete variable and function names based on context, and syntax highlighting is decent. However, the phone screen is too small; one line of code often wraps into two or three lines, making it hard to see the overall logic. Debugging is also challenging since you have to squint at log outputs that are as small as ants.

I also tried its remote management feature, which allows you to see running processes on the server and restart services. This feature is genuinely useful; for example, if your server crashes while you’re out, you can restart it without needing a computer. But you must be very certain about what you’re doing; otherwise, a wrong tap could delete the wrong process, leading to disastrous consequences.

Conclusion

After three days of testing, I concluded that the Codex mobile app can be used in emergencies, but it should never be used for writing production code.

Recommendations:

1. Use it only for viewing code and logs; avoid writing more than 10 lines of logic on mobile.
2. If you must modify code, always review it on a computer before submitting.
3. Use the remote management feature cautiously; confirm which process you’re restarting.
4. Don’t brag in the group about coding on your phone; if something goes wrong, you’ll be the first to take the blame.

I’ve now moved the Codex mobile app from my home screen to the deepest folder on my phone. I won’t open it unless absolutely necessary. On Monday, I even bought my technical director a coffee to apologize. He smiled and said, “It’s great for young people to try new technology, but you need to know the right context. If I catch you coding production code on your phone again, you’ll really be working from home—permanently.”

Have you ever used a tool that seemed impressive but turned out to be a pitfall? Share your experiences in the comments so I know I’m not alone in this!