The World (of AI) Ends With You

There’s been a lot of discourse surrounding AI usage for about a year at this point. Most people praise the fact that ‘vibecoding’ has empowered people (even if they are non-technical) to build apps, while there’s still some who are skeptical of heavily depending on AI for coding.

I think there’s a lot more nuance than having to pick one of the two sides. I’ve more or less accepted that AI will be a large part of my day-to-day. It’s been useful in generating tons of code which I’d otherwise have to handwrite and especially helpful when it comes to debugging issues, especially ones that I do not have an expertise in, but know enough to know if the model is BS-ing me or not. To deny its positive impact in my work would be lying.

But people are overselling its benefits. AI still cannot make sound architectural decisions. It often comes to the wrong conclusions, and changes its mind at a whim if you convince the model that it is wrong, so all the hype around ‘AI will replace engineers’ is nothing but fearmongering. At the end of the day, you still need humans to understand, audit and verify the code that is written to make sure that it makes sense given the requirements of the system that is being maintained or built.

When you see tweets about how great ‘vibecoding’ is, keep in mind that some of these people have a vested interest in you not wanting to code. The founder of Anthropic obviously wants you to give in and use their models so that you eventually gain a reliance on their product. So do the other people making money off these models or wrappers around these models.

Nihilists would　give in to this narrative and completely give up coding because they believe the age of AI is already here and that they would be replaced. This is something very sad to see (and I’ve witnessed it a few times), and is completely untrue.

Optimists, on the other hand, see AI as a tool. The ‘10x dev to a 100x dev thanks to AI’ is a bit of a meme, but also somewhat true. You can, under supervision, build large scale projects at a much faster speed today thanks to AI’s speed at churning out code. We get faster iteration and experimentation, all at 10x the previous speed.

Correctness is still very much dependent on the implementors’ capabilities, as said in the linked tweet. One big reason why I’m skeptical of this vibe-coding craze is security. We’re already seeing a ton of breaches, malware and other security vulnerabilities surface, most likely due to agent-assisted blackhat activity, and I suspect this will continue especially with carelessly ‘vibe coded’ applications.

I do not believe in the future where coding will be done primarily by agents. If agents are doing most software work, they would pick a more efficient format anyway. Software engineering, from inception, has been a way for humans to communicate with other humans. That’s why there are design specifications for software, meetups, conferences and even a whole damn field on PLT. This whole narrative that software engineers will just disappear just seems absurd to me. What is software without humans to work on them and make them useful for other humans?

If all that did not convince you to stop over-relying on AI and start treating it as a tool, remember that coding is fun. Humans were not made to sit in office chairs for 10 hours a day but we do it anyway for the love of the game. You’re possibly an aspiring game dev or you would like to make your own programming language. Pursue those things again and write code by hand and enjoy the joy of programming. I’m personally learning Rocq using the Software Foundations textbook and Idris2 at the moment and would like to do game dev soon.

The title of the post refers to a game, and I didn’t quite get the meaning until recently. The world can be as small or as big as you want it to be. If you think AI will take your job, that will be your world. But if you treat AI as a tool to assist you, the world is as big as you will it to be.