Live data from Hacker News

The Cost of Human-Centric Tools in LLM Workflows

joshbeckman.org

1–7 of 7 posts

Re: The Cost of Human-Centric Tools in LLM Workflows

#2
Having used multiple code review tools (Gerrit, Phabricator), GitHub's code review is frankly one of the worst UX's for humans to use for this purpose (go try to review a rebased patchset, or have to wade through the inefficient use of space in the UI), so I'm not surprised AI has trouble with it as well.

Re: The Cost of Human-Centric Tools in LLM Workflows

#3
post #2

Having used multiple code review tools (Gerrit, Phabricator), GitHub's code review is frankly one of the worst UX's for humans to use for this purpose (go try to review a rebased patchset, or have to wade through the inefficient use of space in the UI), so I'm not surprised AI has trouble with it as well.

Agreed. I think this is why we’re seeing a lot of new wrappers being built like Graphite.

Re: The Cost of Human-Centric Tools in LLM Workflows

#4
post #3
post #2

Having used multiple code review tools (Gerrit, Phabricator), GitHub's code review is frankly one of the worst UX's for humans to use for this purpose (go try to review a rebased patchset, or have to wade through the inefficient use of space in the UI), so I'm not surprised AI has trouble with it as well.

Agreed. I think this is why we’re seeing a lot of new wrappers being built like Graphite.

a lot? what are the other ones?

Re: The Cost of Human-Centric Tools in LLM Workflows

#5
post #3

Earlier quoted context omitted.

Agreed. I think this is why we’re seeing a lot of new wrappers being built like Graphite.

a lot? what are the other ones?

In exploring `gh` CLI extensions recently I found at least 2 devoted to being a better PR review interface. I've also seen [Trag](https://usetrag.com/) and others pop up.

Re: The Cost of Human-Centric Tools in LLM Workflows

#6
post #3

Earlier quoted context omitted.

Agreed. I think this is why we’re seeing a lot of new wrappers being built like Graphite.

a lot? what are the other ones?

Some others which are good:

  * Reviewable
  * Codestream (formerly GitContext)
  * Codelantis
  * CodeApprove (biased, obviously)

Re: The Cost of Human-Centric Tools in LLM Workflows

#7
post #2

Having used multiple code review tools (Gerrit, Phabricator), GitHub's code review is frankly one of the worst UX's for humans to use for this purpose (go try to review a rebased patchset, or have to wade through the inefficient use of space in the UI), so I'm not surprised AI has trouble with it as well.

You can try out Kodus (https://kodus.io)