3 ms·

>> I also know which code I could write with LLMs but I never check in LLM code. I could, but that would be cheating. I spend the hours writing code by hand. S

>> I also know which code I could write with LLMs but I never check in LLM code. I could, but that would be cheating. I spend the hours writing code by hand. Spoken like someone who charges by the hour.... you are incentivised to do things the slowest possible way... Now, of course, slow might be better, or fast might be better, but you frame fast as "cheating" not as "inferior"... which is ... interesting..

rozenmd 5mo ago
Charging by the hour was one of the worst things for my career (was just starting out). Never realised at the time that I was subconsciously doing a slower job and not trying to make it faster or really improve until I left.
codegeek 5mo ago
Exactly. I hired someone who was a good developer but he was charging hourly and extremely slow for what we needed. I am a software engineer (and a founder) myself so I get what it takes to write good code but I am no longer waiting for a dev to turn something around in 20 hours when I can use LLM to write it in 1 or less. Going forward, I am no longer hiring hourly rate devs. Either fixed rate project or full time as needed. No hourly.
zerr 5mo ago
> you are incentivised to do things the slowest possible way... Not when you have long-term (multi-years) clients. It's basically a very flexible permanent job, where you are free to bill from 0 up to full 40 or more hours per week.
austin-cheney 5mo ago
I would never pay by the hour for LLM output. The expectation there is two-fold: 1. If LLMs are doing the work then why pay a consultant at all in the first place. 2. Secondly, I would expect the LLM output to cost just as much, or more, because it would require deeper code inspection and more thorough test automation cases. In my manually written code the test automation cases are there to account for future regression where in the LLM code the test cases to there to identify bounds checks against the requirements.
noduerme 5mo ago
My goal isn't to max out hours but to keep clients coming back and get the work done so I can move on to the next thing. Having LLMs write production code for clients presents a lot of extra risk for me in maintaining those codebases in the future. Writing it myself, I know exactly how every stitch of it works, how the logic is delegated, and where everything lives. With LLM-written code for something large and complex, you practically need an LLM to debug it. With my own code, if a bug crops up, I can reliably guess which file and which function out of hundreds is causing the issue. One thing I don't charge for is fixing bugs once software is delivered. So by avoiding LLM-generated code, I feel I limit my exposure to poring over files I may have only read once, didn't write, and feel less confident in understanding the implications of any changes I make.

rozenmd 5mo ago

Charging by the hour was one of the worst things for my career (was just starting out). Never realised at the time that I was subconsciously doing a slower job and not trying to make it faster or really improve until I left.

codegeek 5mo ago

Exactly. I hired someone who was a good developer but he was charging hourly and extremely slow for what we needed. I am a software engineer (and a founder) myself so I get what it takes to write good code but I am no longer waiting for a dev to turn something around in 20 hours when I can use LLM to write it in 1 or less. Going forward, I am no longer hiring hourly rate devs. Either fixed rate project or full time as needed. No hourly.

zerr 5mo ago

> you are incentivised to do things the slowest possible way... Not when you have long-term (multi-years) clients. It's basically a very flexible permanent job, where you are free to bill from 0 up to full 40 or more hours per week.

austin-cheney 5mo ago

I would never pay by the hour for LLM output. The expectation there is two-fold: 1. If LLMs are doing the work then why pay a consultant at all in the first place. 2. Secondly, I would expect the LLM output to cost just as much, or more, because it would require deeper code inspection and more thorough test automation cases. In my manually written code the test automation cases are there to account for future regression where in the LLM code the test cases to there to identify bounds checks against the requirements.

noduerme 5mo ago

My goal isn't to max out hours but to keep clients coming back and get the work done so I can move on to the next thing. Having LLMs write production code for clients presents a lot of extra risk for me in maintaining those codebases in the future. Writing it myself, I know exactly how every stitch of it works, how the logic is delegated, and where everything lives. With LLM-written code for something large and complex, you practically need an LLM to debug it. With my own code, if a bug crops up, I can reliably guess which file and which function out of hundreds is causing the issue. One thing I don't charge for is fixing bugs once software is delivered. So by avoiding LLM-generated code, I feel I limit my exposure to poring over files I may have only read once, didn't write, and feel less confident in understanding the implications of any changes I make.