You get my point. It definitely doesn’t look like my elderly neighbour, nor like my daughter, etc. It is confusing but very simple at the same time.

I’m not sure why would you want to think about how an AI would help. If you understand what you need to do and you understand the capabilities of the tools available you run the coding / learning loop. If you think you should use AI then you are solving the how to use AI problem, which just adds to the complexity. Orthogonal to first principle thinking, just a workflow issue.

You know what to do. Just don’t watch shit :)

Harder feedback: the writeup feels like a quickly hacked together memo, which claims it is “self healingly” good. Because it has this throwaway quality I’m not sure what is the goal? Shall I engage with it seriously, when the lead dev didn’t seem to put in the effort? But if I don’t engage with it seriously it means that I have to follow this patchy fremwork?

Polish: make it nicer

We have a hard time communicating about things. She sometimes can’t express herself at all, she sees things she thinks I should be able to see. I have very good ability to remember schemata, maps. I move really quickly in pages and pages of text. I can summarize the pages and pages of llm output in less time than it is generated. She can easily put 8 grammatical mistakes in a 5 word sentence. We can communicate easily when i write the code quickly and she can checks the visual output in the browser, this is really fluid.

Want more examples ? :)

These llm agents don’t have “agency”. No intrinsic goal other than respond. A tutor has judgment and agency. You will get a really good pattern match on the input text you give but you won’t get tutoring.

I think the rewrites are the main story for llms in code (hot take). Writing greenfield code at the seams also something which might work well.

> Workflows like jira_close.yml use deprecated atlassian JIRA actions and have a dependency on the gh-actions repo. This is not ideal and unecessarily complex.

And then goes on:

> PR updates jira_close workflow to use direct API calls via curl.

Duplicating the logic into OUR codebase via a hand rolled curl, so we can get rid of “needless abstractions”. Auch. And of course the whole thing embedded into a yaml file.

This code is the typical kaleidoscope sometimes written by junior devs (and LLMs). On review you just kindly ask to be rewritten into a simple program or just close it as the effort doesn’t worth it.

I think this is the core point of the article. I really hope that we will stop prompting llms to refactor codebases and start using deterministic llm supported toolchains for such tasks. I think you either blindly trust an llm or you don’t. And no post-hoc review will be enough to contain llms, we will super quickly create “dark factory” without the benefits.

Comments URL: https://news.ycombinator.com/item?id=49230778

Points: 3

# Comments: 0