51–60 of 81 posts
AI creates bloated, sometimes nonsensical output that often straight-up doesn't work. It's perfect for replacing web developers. It can make mobile apps, too.
Just wait, soon there will be a JavaScript framework: by AI, for AI.
It frees up code monkeys to become designer/testers.
But they are not the people companies want to get rid of, they never had any integrity to begin with.
Maybe idea of using AI to develop anything is going to die and instead it is just setting up AI that everyone then has to use...
Disclaimer, I did not use AI for this post.
It's missing one critical point: if AI makes a good developer 10% more productive, the industry needs 10% less developers to achieve the same output. You're not getting replaced by AI. You're getting replaced by a coworker using AI. It doesn't matter how poorly AI performs at any remotely complicated task, what matters is how many developer-hours it saves by not having to manually write as much boilerplate.
Alternatively: if AI makes a good developer 2x more productive (which I do think is possible based on my own experience), developers are now 2x more valuable to employers. The cost of building software drops in half, which means employers that previously wouldn't have developed their own software (too expensive) are now in the market for custom solutions. Demand for developers goes up. If you believe that making deve…
Because I suspect that comes with learning a lot less in the process and lower quality output.
It's missing one critical point: if AI makes a good developer 10% more productive, the industry needs 10% less developers to achieve the same output. You're not getting replaced by AI. You're getting replaced by a coworker using AI. It doesn't matter how poorly AI performs at any remotely complicated task, what matters is how many developer-hours it saves by not having to manually write as much boilerplate.
> the industry needs 10% less developers to achieve the same output Maybe. Or maybe the industry will simply churn even more software due to increased possibilities and everything stays the same.
It's missing one critical point: if AI makes a good developer 10% more productive, the industry needs 10% less developers to achieve the same output. You're not getting replaced by AI. You're getting replaced by a coworker using AI. It doesn't matter how poorly AI performs at any remotely complicated task, what matters is how many developer-hours it saves by not having to manually write as much boilerplate.
My company has had a backlog thousands of developer-hours long for years. Maybe eventually somebody will get fired because we're not needed, but the LLM would have to actually 10x our productivity for us to get through the backlog in a year.
You would likely lose absolutely nothing by throwing away most of it, and gain a lot.
It's missing one critical point: if AI makes a good developer 10% more productive, the industry needs 10% less developers to achieve the same output. You're not getting replaced by AI. You're getting replaced by a coworker using AI. It doesn't matter how poorly AI performs at any remotely complicated task, what matters is how many developer-hours it saves by not having to manually write as much boilerplate.
Dunno man, the times I’ve used AI to help me code, it got the solution wrong. I spent more time backtracking and fixing things than if I’d done it by hand in the first place. So it actually made me 20% less productive.
Very refreshing!
Earlier quoted context omitted.
Alternatively: if AI makes a good developer 2x more productive (which I do think is possible based on my own experience), developers are now 2x more valuable to employers. The cost of building software drops in half, which means employers that previously wouldn't have developed their own software (too expensive) are now in the market for custom solutions. Demand for developers goes up. If you believe that making deve…
Yeah, define productive. Because I suspect that comes with learning a lot less in the process and lower quality output.
You have to use them diligently, but it's definitely possible to avoid the obvious downsides of working in this way.
Earlier quoted context omitted.
Have you tried o3-mini? I'm finding it to be significantly better than GPT-4o for code - I'm even finding myself using it in place of Claude 3.5 Sonnet a lot of the time.
A lot of the time? What kind of code are you writing, if I may ask; and how long have you been doing it? I sometimes ask whatever AI open ended questions on subjects I know very little about, and sometimes I find some kind of starting point for less random research using better tools. Beyond that, and whatever fancy autocomplete my IDE forces down my throat, nope. If if was gone tomorrow I would barely notice.
I've published a ton of notes on this here: https://simonwillison.net/tags/ai-assisted-programming/ and https://simonwillison.net/series/using-llms/