Hacker News Reader

Comment by Sean-Der | Hacker News Reader

Here is another thing I find frustrating, but I don't know the right way to express it. A lot of this AI tooling/coding seems self-indulgent? 'I built a better software factory!'. The tools are very powerful, but I don't see the results yet. The frustration outweighs results.

I am underwater dealing with it. I am getting outlandish PRs against Pion. So I spend my time triaging and feel burned out. The 'AI future' I was expecting would be people swooping in and clearing up the backlog in high quality ways.

Instead I get people trying to add new APIs when it can already be done with existing APIs and then arguing about it with me.

View on HN

Please don't forget you are under no obligation to triage all the awful things.

You are well within your rights to require submitters reach a certain standard before you will review their request.

Your time is finite and precious. Your project benefits from your enthusiasm and knowledge. It does not benefit from you acting as human reviewer of a tsunami of low quality content.

You are allowed to be ruthlessly protective of your time to the benefit of your project.

This begs an importent question: how? How do you filter without putting up barriers that high quality contributors will not want to climb?

I ask both as a maintainer who has been overwhelmed and as someone who declines to participate in communities where the bar is set too high for casual drive-by contributions.

Make it a rule that any PR that changes over ... 100 LoC? 150? <some number you're comfortable with> needs to be broken down into smaller, more atomic ones.

If that can't be done, e.g. because it's a very large refactor that can only be published as a unit, get them to walk you through each design and architectural choice.

I don't think a high quality contributor will take issue with that.

I only maintain my own codebases, so my deepest commiserations my fellow.

I’d imagine I’d just want to turn repo’s private or gate who has access, that’s my default response.

Because the factories of factories is mostly bullshit. OpenAI blogged about how it was complicated for them to build a project using AI alone and the specs in the repo, how easy it was for models to get lost, not do what was expected of them and the burden it was on the team.

I don't think we will never get there but we're definitely not there yet. Models still require a lot of handholding to get things done.

Reply on news.ycombinator.com