Building a Side Project with OpenAI Codex, FOMO, and Autumn Rain
While it rains outside almost every day, I’ve gone back to working on the project (the same one I first planned to launch in August, then in September).
Here is what is already done:
1. A solid MVP that covers the key functionality
2. A Telegram Mini App (about 80% done, it handles the core functions but not fine-tuning and extra features)
3. Payments via Stripe 😏
4. A future full architecture and a big backlog that I’ll tackle after release
Now to the interesting part – the challenges I’ve run into:
First, I wrote part of the code with OpenAI Codex, and that turned into a big problem. At first it looks great: you do not waste time on some basic stuff, you write a prompt and get working code. The problems start when you realize this ‘basic’ thing should not be basic at all and needs to be more complex. Codex is bad at making the logic more complex (or I’m not a prompt master), so I have to dive into code that is effectively someone else’s and spend time untangling it.
Second, as usual I’m facing FOMO. I do not just want an MVP, I want a bunch of cool features and almost a full mobile app right away. What helps me fight this is one of my favorite books, ‘ReWork: Change the Way You Work Forever’.
Third, a month before Web Summit I lost motivation and worked at half speed. But after Web Summit I realized the reason was the huge amount of background noise that does not just distract you, it pulls you in so much that you stop understanding what is going on around you.
Now to the good part – OpenAI Codex. I’m fairly skeptical about so-called vibe coding, simply because I believe that if you do not know the code and the architecture, nothing good will come out of it. So I use Codex as an assistant, not a replacement. And for that it works brilliantly. Say there is a class with several methods that I wrote myself following a predefined code style. I need to add a few new methods – Codex handles that perfectly and saves a lot of time.
On top of that, I fully switched to Windsurf this summer. It has its own agent, but since I pay for ChatGPT Business it is cheaper for me to use Codex, and I basically use Windsurf only for AI autocomplete (it is so good!).
I won’t commit to any deadlines for the project right now, but I will try to post here a bit more often and gradually share details not only about the progress, but also about what this project is actually going to be 🚀
More to explore
Startup Taxes Between Estonia and Portugal: A Quick Reality Check
As a tax resident of an EU country who files my own returns, today is my quarterly 'Tax Day'. On this day I set aside a few hours to file social security report…
Saylify Update: Fighting Perfectionism, Refactoring, and Finding the Right Focus
I have not written anything about Saylify for a long time, even though I planned to launch in January. Unfortunately, life likes to throw in challenges you can …
Human-Like Memory for LLMs
TL;DR I wrote a manifesto-style essay about a memory model for LLMs that is as close as possible to human memory and lets the system build a relationship histor…
When Companies Finally Say the Ugly Part Out Loud
Now we are finally fucking talking. Not all that crap like "internal policies", "no explanation needed", "just because".1Office are the first who wrote it plain…