[Daishiman]

> The biggest problem with AI PRs is the sheer amount of churn, extra code and lack of intent with the PRs it generates.

But this isn't an LLM problem; this is a problem of undisciplined engineers who feel they need to cram extra stuff in a PR. If an engineer doesn't look at the output of the LLM and generate extra work then it's still on them, right?

> The only way I can describe the latter is that an AI-only PR feels to me like a painting where everything is high detail - and you have to comb over each part before you understand why it's there because so much is superfluous

This just indicates that the engineer doesn't know how to use the tool. Hell they can ask the LLM to split the work into focused PRs and Claude will be happy to do it and the results might no even be half bad.

> Also when I'm _using_ the agent; at least 50 percent of my time is spent telling it to stop with it's approach so it doesn't go down a useless rabbit hole and waste tokens.

If this is happening often then the tool is probably not fit for the job.

[yw3410]

I don't believe so - it's not as though the original prompts asked for extra code churn (note that as soon as you look and edit the LLM code output extensively it ceases to be vibe coding, which I was talking about in my OP).

I'm not talking about extra feature s; I'm talking about for the same single feature the code is either convoulted because the algorithm is overly complicated or the abstractions are just wrong for the domain.

The PRs typically are already focused in that they address a single feature; or at least a single "usable" feature in a complex system which necessarily has a lot of connected parts and behaviors.

> then the tool is probably not fit for the job.

Perhaps; but with an LLM I haven't found which jobs it _does_ work for and which it doesn't. I already use planning mode extensively; and capture the major points, but then it makes a stupid decision mid implementation and just starts churning.