AI coding is gambling

Source: Hacker News

AI Coding is Gambling

I’ve been coding a lot with AI since November, when we all noticed it got really good. It’s impressive for instantly generating something that looks half‑decent, but the actual details—the individual parts that make a system—are still a challenge.

I’m not here to review a coding agent or nitpick its output, nor to brag about running Claude for eight days and ending up with an 8‑plus‑year portfolio of projects. I want to talk about feelings, a life well lived, and a nurtured soul.

The Gambling Proposition

Getting into a state where any change to your entire codebase feels trivial is intoxicating. Previously we were burdened by our own cognition and laziness: a ticket required us to estimate effort, look up documentation, read forgotten code, and reconnect with our thinking—often over months or years.

Now the AI can either handle the task or pretend to. It often pretends, but the result can be “vaguely plausible” and surprisingly wrong. This isn’t true coding; the thinking part is offloaded to the AI, and the actual code written can be minimal.

It maps perfectly onto the tech industry’s favorite mechanic: gambling. It’s like pulling a slot‑machine lever with a custom message. We’ve been refreshing for years, and now the “general intelligence” becomes a gambling machine. That explains why it’s so addictively appealing. I won’t decry the benefits or fear for my job, but you need to know what you’re doing to avoid holes in your code. I’ll explore a simpler problem instead.

The Simplest Problem

I divide my tasks into “good for the soul” and “bad for it.” Coding generally falls into the former, even when I do it poorly. Gathering inspiration—finding what others have built and figuring out how to integrate, refine, or iterate—is also soul‑nourishing. The infinite plagiarism machine makes that easier.

But it robs me of the most rewarding part: figuring out how something works for me, finding clever fixes, and getting it running. My job shifted from connecting ideas (the hard, rewarding part) to merely mopping up poorly connected code. It’s deeply unsatisfying, and while many factors contribute, the fix rests on me: I must avoid laziness and interact with my code more, using the methods I’ve honed for years to find inspiration and cleverness on the internet instead of defaulting to the infinite machine.

The Special Case

I’m not the average developer. I’ve never worked on large teams and have barely started a project from scratch. The internet is filled with freely available code and ideas to fork and change.

My experience includes small‑team work and solo development, which has taught me to reuse, minimize, and optimize code. Yet I’m more a designer than a traditional developer. Should I be happy that AI has made me a better developer?

I question that. AI has certainly boosted my confidence in trying new frameworks and stepping out of my comfort zone, and I’ve been spending more time coding. But is that because I’m becoming more efficient and smarter, or because I’m gambling on what I want to see? Am I just pulling the lever until I hit a jackpot?