Hacker News Reader

Comment by ryandrake | Hacker News Reader

I've found that your step 6 takes the vast majority of the time I spend programming with LLMs. Like 10X+ the combined total of time steps 1-5 take. And that's if the code the LLM produced actually works. If it doesn't work (which happens quite often), then even more handholding and corrections are needed. It's really a grind. I'm still not sure whether I am net saving time using these tools.

I always wonder about the people who say LLMs save them so much time: Do you just accept the edits they make without reviewing each and every line?

View on HN

You can have the tool start by writing an implementation plan describing the overall approach and key details including references, snippets of code, task list, etc. That is much faster than a raw diff to review and refine to make sure it matches your intent. Once that's acceptable the changes are quick, and having the machine do a few rounds of refinement to make sure the diff vs HEAD matches the plan helps iron out some of the easy issues before human eyes show up. The final review is then easier because you are only checking for smaller issues and consistency with the plan that you already signed off on.

It's not magic though, this still takes some time to do.

I exclusively use the autocomplete in cursor. I hate reviewing huge chunks of llm code at one time. With the autocomplete, I’m in full control of the larger design and am able to quickly review each piece of llm code. Very often it generates what I was going to type myself.

Anything that involves math or complicated conditions I take extra time on.

I feel I’m getting code written 2 to 3 times faster this way while maintaining high quality and confidence

Maybe it subjectively feels like 2-3x faster but in studies that measure it we tend to see smaller improvements like in the range of 20-30% faster. It could be that you are an outlier, of course.

2-3x faster on getting the code written. Fully completing a coding task maybe only 20-30% faster, if we count chasing down requirements, reviews, waiting for CI to pass so I can merge etc.

This is my preferred way as well. And when you think about it, it makes sense. With advanced autocomplete you are:

1. Keeping the context very small 2. Keeping the scope of the output very small

With the added benefit of keeping you in the flow state (and in my experience making it more enjoyable).

To anyone that even hates LLMs give autocomplete a shot (with a keying to toggle it if it annoys you, sometimes it’s awful). It’s really no different than typing it manually wrt quality etc, so the speed up isn’t huge, but it feels a lot nicer.

If it's stuff I have have been doing for years and isn't terribly complex I've found its generally quick to skim review. I don't need to read every line I can glance at it, know it's a loop and why, a function call or whatever. If I see something unusual I take that as an opportunity to learn.

I've seen LLMs write some really bad code a few times lately it seems almost worse than what they were doing 6 or 8 months ago. Could be my imagination but it seems that way.

Reply on news.ycombinator.com