3 ms·
Estimates for trivial programming tasks are trivial estimates. In a real software engineering process not one task is trivial. The implementation depends on th
by zare_st 3y ago
Estimates for trivial programming tasks are trivial estimates.
In a real software engineering process not one task is trivial. The implementation depends on the architecture of the project and specifics around it. These things are out of LLM scope, even if it knows the language.
Even non-trivial feature estimations should be easy if the process is good.
It's the bugs that you can't estimate because they're unforeseen consequence by definition.
- amelius 3y ago> It's the bugs that you can't estimate because they're unforeseen consequence by definition. Well, it doesn't need to be better than human estimates. Just less time-consuming.
- zare_st 3y agoMy point is that some things cannot be precisely estimated in any case. What can be estimated by humans and computers alike is how much time does it take an average programmer to implement something if it's generic enough. An expert system consisting of LLM+more would spit out the stack overflow copy/paste answer and the 'more' part would give you the average effort needed to copy/paste that code in the project and adapt the variables,etc. If you have "bugs" that you can resolve by simple stack exchange googling you need to up your project tooling and programming guidelines. What's left is real bugs where you operate in the dark and no computer will able to estimate the size of that room because the light switch is off. Once you run head into it and light turns on, you're able to clearly see the size of the room and do the calculation yourself.