Tell HN: I regret every single time I use AI

11 points by jwpapi ↗ HN
I try to not be fully against AI, so keep giving it a change, today again. I went for sport and gave opus 4.6 a medium sized task.

In every single step there were assumptions done that I don’t think are correct. I was a refactor and not all is Claudes fault. But since I did have bad experience I work through the changes line by line and honestly I think I would rather have done it from the start then to review the changes as Claude did a not working weird verbose refactor that didn’t help with anything.

When I refactor personally I don’t lose focus, I get the mental model, I apply the virtue of laziness and find 7 things that could be done better and ask myself 10 more things on the way (for which I actively use AI for exploration)

Also whenever I give AI a task I’m bored and I watch a Youtube video or read a HN article, when I do work on the code myself I’m not sure I’m so much slower if I know what I’m doing, given the cost of exploration, context switching, more iteration pipelines, ai time + my reviewing time.

Right now I feel the best way of utilizing AI is not having it sit in your code base but rather have it a browser or somewhere, that way you can use it as research tool, but you are in the code.

I also strongly recommend to test every single step that you are not 100% sure about AI, the more specialized your requirement is the more often it is wrong, especially with changing versions etc.

I did set up a shortcut to create a vitest test file from every function and strg + t to run it and alt + t to run the last test again, so even on changes in consumed functions I can easily run it.

The fast feedback loop is so enabling. Can really recommend.

How do you guys doing serious workflows (no fantasy apps) use AI atm?

4 comments

[ 3.4 ms ] story [ 17.8 ms ] thread
Like seeing orchestration pipelines, videos, tools or blog posts fills me with literal anxiety, like I feel bad for all these people. They seem like people on drugs to me that think they are on top of the world.

To be clear I’ve been like this as well and I learned my lesson. I’m also not saying that I’m right. These are just feelings that I experience, I could be the idiot.

Does anyone else feel the same?

i think there are two key things that helped me ship more successfully using ai

1. must isolate context. discuss with your architecture agent, implement with another. you can pass the implementation results back to the architecture agent to check for implementation drift. ai's self check and correction sucks - i guess it is because of the attention mechansim?

2. iterate with your architecture agent to produce a tightly scoped task spec. really need to iterate, ask it your align with you for the key assumption. dont be too ambitious. i myself has a guideline for task spec writing that specify spec cannot cross boundary or work with 2 subsystems in one go

but honestly, ai is only great at diagnosis and implementation. most of my successful runs are on the basis that i know exactly how to code or how to solve the problem. ai just do the dirty work for me.

Would you bet on you not being faster with autocomplete. No context loss for you and you can review and write at the same time.