pull down to refresh

In my experience this has been true. LLMs have plenty of flaws but many can be overcome with skill/knowledge about how they work and how to minimize the effects. Time will tell how valuable these tools become and if they are profitable but it is promising for sure.
The counterargument I hear most often is that AI-generated code creates technical debt and quality problems. That’s absolutely true if you’re trying to ship Claude’s output directly to production. But for prototyping and validation, those concerns are premature optimization. You don’t need perfect code to validate whether people want your product.
Like many things there is a desire in many of us to say it sucks. Especially if we feel threatened by it. This is even more true when people are clearly hyping something and being lose with facts and logic while doing so. Also true of the AI hype cycle.
But, I have tried to push that stuff aside and see for myself if I can find use for these tools. I have access to many tools through work that I've been encouraged to evaluate. What I have seen so far does not threaten me in the slightest. One of the biggest things is that you can't trust these tools yet. I am not convinced you will ever be able to trust them. But if you have people that know what they are doing and have actual real world experience shipping software that works, on time and on budget these tools could be a massive boost to productivity.
They aren't for the lazy. They aren't a replacement for devs. They seem to be an accelerant. If an engineer has experience with breaking down a project into small chunks and delegating tasks they have some of the skills to use LLMs with code generation effectively. But like working with a new junior dev, there is a learning curve.
It is interesting to watch people that are clearly intimidated by the perceived threat. There's this idea that if they ignore them or refuse to use them they will go away. I really think my study of economics and economic history has helped me understand where to point my attention. I'm not gonna stop progress (if this is progress). As an economic actor my goal with work should be to make myself effective and indispensable. Its hard to do that with your head in the sand.
Depending on the framework, I’ve basically found the same as OP. On some tool chains it hallucinates and gives commands / scaffolding that is outdated or wrong. But for popular frameworks like next.js or similar and HTML/CSS it can build out the basics very quickly. I’ve also found it useful for explaining rust idioms in a way that I can understand. There are still issues when you have a long discussion and ask it to tweak something. Not an expert but that might be down to context window size. Sometimes it reverts to an older version of the framework and gives code that I can tell won’t work.
reply
They seem to be better with different programming approaches too. I have a colleague that believes they are better with FP than OOP. It kinda makes sense to me that this would be true.
reply
Fewer disparate concepts / objects and fields to keep track of, each function is its own self-contained programming context, kind of makes sense to me too.
reply
Indeed.
They're great for putting the whole skeleton in place, especially when it's a subject area I have less experience in. It shows me the common and expected patterns. I can always go in and change the parts manually that are causing problems.
reply
stackers have outlawed this. turn on wild west mode in your /settings to see outlawed content.