Comment by bastawhiz
21 hours ago
I have not had the same experience as the author. The code I have my tools write is not long. I write a little bit at a time, and I know what I expect it to generate before it generates it. If what it generates isn't what I expect, that's a good hint to me that I haven't been descriptive enough with my comments or naming or method signatures.
I use Cursor not because I want it to think for me, but because I can only type so fast. I get out of it exactly the amount of value that I expect to get out of it. I can tell it to go through a file and perform a purely mechanical reformatting (like converting camel case to snake case) and it's faster to review the results than it is for me to try some clever regexp and screw it up five or six times.
And quite honestly, for me that's the dream. Reducing the friction of human-machine interaction is exactly the goal of designing good tools. If there was no meaningful value to be had from being able to get my ideas into the machine faster, nobody would buy fancy keyboards or (non-accessibility) dictation software.
I'm like 80% sure people complaining about AI doing a shit job are just plain holding it wrong.
The LLM doesn't magically know stuff you don't tell it. They CAN kinda-sorta fetch new information by reading the code or via MCP, but you still need to have a set of rules and documentation in place so that you don't spend half your credits on the LLM figuring out how to do something in your project.
You do understand that 'you're holding it wrong' is a cop out about some brittle system?
It’s also a reference to the fact that the intuitive way of using AI isn’t always the correct way.
You need to hold it a bit weird.
I was wanting to build a wire routing for a string of lights on a panel. Looked up TSP, and learned of Christofides herustic. Asked Claude to implement Christofides. Went on to do stuff I enjoy more than mapping Wikipedia pseudo code to runnable code.
Sure, it would be really bad if everyone just assumes that the current state of the art is the best it will ever be, so we stop using our brains. The thing is, I'm very unlikely to come up with a better approximation to TSP, so I might as well use my limited brain power to focus on domains where I do have a chance to make a difference.
This is exactly the way I succeed. I ask it to do little bits at a time. I think that people have issues when they point the tools at a large code base and say "make it better". That's not the current sweet spot of the tools. Getting rid of boiler plate has been a game changer for me.
I think my currently average code writing speed is 1 keyword per hour or something as nearly all my time coding is spent either reading the doc (to check my assumptions) or copy-pasting another block of code I have. The very short bust of writing code like I would write prose is done so rarely I don't even bother remembering them.
I've never written boilerplate. I copy them from old projects (the first time was not boilerplate, it was learning the technology) or other files, and do some fast editing (vim is great for this).