unknown

Ruby on Rails creator David Heinemeier Hansson argues that current AI coding tools are inconsistent and unreliable, comparing them to a "flickering light bulb" that occasionally produces impressive results but often generates poorly structured, hard-to-maintain code. At his company 37signals, humans wrote 95% of the code for their new product Fizzy. He emphasizes that while AI can generate working code, it lacks the understanding of system architecture, edge cases, and long-term maintainability that even junior developers possess. AWS CEO Matt Garman echoes similar concerns, noting that eliminating junior developers creates future skills gaps. The core issue isn't AI's ability to write code snippets, but its inability to consistently understand context, design systems, and produce production-ready code that teams can safely maintain over time.

AI Not Ready to Replace Junior Devs Says Ruby on Rails Creator

kt123123123

<blockquote>
In real software teams, code is about whether someone else can understand it, fix it, and safely build on it later.
</blockquote>
No doubt. Although humans do a pretty bad job at this too.

I think an AI’s initial output is proportional to the level of detail in the context. Therefore, inconsistency comes from omitting instructions to write that kind of code. It’s like telling a brand-new project member to write code without looking at anything first.
Certainly, the reason AI may feel unhelpful as described in the article might be that the idea of “Vanilla Rails” as practiced by Basecamp, especially something like Fizzy differs somewhat from what is generally considered a typical Rails implementation.
If DHH and Basecamp had defined a truly reference-like, official Rails application back when Rails was at the height of its momentum, things might have turned out differently.
That aside, I can only be grateful that they made Fizzy public. It’s Rails in a reference-like form, and I hope this kind of direction continues in the future.

I like to think that this will stay true for a bit, lol. Don’t get me wrong, I am all about innovation and efficiency… Unfortunately, I am late to the party and made a lot of bad decisions that placed me where I am at in my current point in my life. I just recently (as of 8mths ago) started to take a deep dive in Networking basics and started branching out into more CS concepts like programming, and a little bit of security as well as architecture- small amounts, lol.
I must say that I feel like I cheated myself out of a lot of great opportunities but, is the reason why I work so hard these days - working FT for UPS in operations as mgmt, and taking online courses FT to obtain BSCS/SE in about 2.5 more years. The more I learn, the more I realize that school alone is nowhere near enough knowledge for me to grow into a backend dev, or something else, but idk. I am just hoping that I can home in on something and really learn all the frameworks around it. It is all so fascinating.

Makes sense. AI can help, but I still see a lot of weird bugs and messy code from it. Not ready to trust it alone yet.

I think it is really simple. AI is a tool for developers. How good it is depends on how good the developer who uses it is.