A friend strongly recommended Codex to me recently, so I gave it a spin. After trying it myself, I’d call it an “idea extractor.”
Before this, my impression of AI coding was that it could build things, but the resulting code would be a mess to maintain. But after actually getting my hands dirty, I burned through 72M tokens in one go. My ideas kept getting pulled out and turned into reality. It was a massive productivity explosion.
What I Built
Let’s talk about what I actually built over the last two days. First, this blog. I did a major refactor, added a glassmorphism background, optimized the link layout, and finally revived a background wallpaper that had been gathering dust for years. Using the wallpaper from my very first blog instantly brought back old memories.
Also, since I’ve been messing around with AI covers lately, I used AI to rewrite a tool I originally planned to code from scratch. If I did it myself, it probably would have taken me weeks. I would’ve slapped together a half-assed GUI just to get it over with, and it wouldn’t look anywhere near this good. (I might have had to relearn C# and WinUI just to make it look decent). But with AI, it’s different. You feed it your ideas, wait a bit, and see the results. It even “conveniently” fixes logical bugs on the fly. Seeing your ideas come to life constantly sparks new ones, and you get “addicted” because building stuff is the ultimate dopamine hit.
Finally, I updated my personal homepage, which hadn’t been touched in years (the root domain of this site: https://yexca.net/ ). It looks awesome.
The Experience
The overall experience is just amazing. It feels like you are the product designer, and the AI is slowly turning your ideas into a real product. Major PM vibes. I was honestly wondering why humans even need to sleep—having your creative flow limited by physical fatigue is incredibly frustrating.
Depending on the scenario, the experience varies. Overall, it feels like being an architect. You shift from raw coding to high-level design. Think of the traditional software engineering lifecycle: requirements gathering, high-level design, detailed design (like API design for frontend/backend), and finally, development. AI is basically taking over that final step. (You can look up the Software Development Life Cycle; I’ve written about it before).
I saw this video right before publishing, which shares a very similar viewpoint: https://www.bilibili.com/video/BV1YP5W6ZEP9
Anxiety
The results are so mind-blowing that it actually made me worry about what kind of jobs will be left for us in the future.
But looking at it objectively, AI is still just a tool. A complete non-programmer and an experienced dev building the same thing will get very different results. Experienced devs describe things with a programmatic mindset, which helps the AI generate much more accurate and maintainable designs.
Also, what really matters is human creativity. AI only generates or visualizes things based on your prompts. This creative spark cannot be replaced by AI—at least not yet.
Still, theoretical reasoning doesn’t quite cure the anxiety. So, I tried to look at this through the lens of history.
Micro-level
If AI is just a relatively minor shift, let’s look at video games and online shopping as examples:
| Phase | Video Games | Online Shopping |
|---|---|---|
| Start | 1972 (Release of arcade “Pong”) or 1983 (Nintendo NES rise after the Atari shock) | 2003 (Launch of Taobao) |
| Mainstream Adoption | Mid-to-late 1990s (PS1, N64 era; seen as standard entertainment) | 2013-2015 (Mobile payment boom, Double 11 becomes a massive shopping festival) |
| Duration | ~15-20 years | ~10-12 years |
| Notes | Accompanied the growth of a generation. Went from being labeled “electronic heroin” to a major part of pop culture. | Key breakthrough was trust mechanisms like Alipay, and the widespread adoption of smartphones. |
Macro-level
But what if this is a massive paradigm shift? Let’s look at the Industrial Revolutions.
| Phase | First | Second | Third |
|---|---|---|---|
| Start | 1765 (Watt’s steam engine) | 1882 (Edison builds first power station) | 1991 (Birth of the World Wide Web) |
| Mainstream Adoption | 1840s (Railway mania, rise of factories) | 1920-1930s (Home electrification, Ford Model T) | 2007-2010 (Smartphone boom, mobile internet adoption) |
| Duration | ~75 years | ~40-50 years | ~16-19 years (half a generation) |
| Notes | Spanned several generations. Slow migration from rural to urban areas, with long struggles like the Luddite movement (smashing machines). | Spanned two generations. People were initially terrified of electricity until home appliances and assembly lines reshaped middle-class life. | Went from panic over “internet addiction” to everyone owning a smartphone in under 20 years. |
The Evolution Process
| Phase | Core Feature | Public Reaction | AI Parallel |
|---|---|---|---|
| 1. Breakthrough & Awe | New tech blows old efficiency out of the water. | Geek hype, public curiosity, early worries from some. | The internet-wide buzz when ChatGPT was first released. |
| 2. Shock & Panic | Threatens legacy industries and actual jobs. | Boycotts, lawsuits (copyright issues), doom-posting, ethical panics. | Current phase: artists boycotting AI art, copywriters/support agents worrying about layoffs, data privacy concerns. |
| 3. Regulation & Integration | Laws catch up, and tech starts solving its own pain points. | Trust builds, industries start adopting the tech as a utility. | Enterprises deploying custom AI tools, governments drafting AI safety acts. |
| 4. Invisible Infrastructure | Tech becomes part of the baseline infrastructure. | Taken for granted; people can’t live without it. | The future: just like how no one thinks googling something on a phone is “high-tech” today. |
The Future of AI
Clearly, history shows we panic, then accept, and finally adapt—and this cycle is getting faster. If we mark the release of ChatGPT (late 2022) as year zero for generative AI, we can project what’s next based on past patterns:
- The infrastructure is already there: The first three industrial revolutions required laying down rails, stringing power lines, and building cell towers. Building physical infrastructure takes forever. AI, however, piggybacks on the existing internet and devices—no new physical rollouts needed.
- Super-compressed adoption cycles: The first industrial revolution took 75 years to go mainstream; the third took 16. E-commerce took about 10 years.
- Predicting AI’s adoption timeline: Starting from 2022, the shift from “panic and novelty” to “invisible daily utility” will likely be compressed into 5 to 8 years (putting us around 2027–2030).
There’s a funny saying about how people react to new things: Ignore it -> Look down on it -> Don’t understand it -> Too late to catch up. A more formal way to put it: Panic/Resistance -> Friction/Adaptation -> Business as usual.
Wrapping Up
So, we should just run with it and embrace the AI ecosystem. The current kinks will get ironed out. The current anxiety about losing jobs is likely just like the panic during the first Industrial Revolution—over time, a whole new wave of jobs we can’t even imagine right now will pop up.