{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreian5jbbekzaxp7osbxcem7o3fq3s37x2jmjtrviwa3cu42wxccndq",
    "uri": "at://did:plc:pi6woz4d47bkuws673w2il2r/app.bsky.feed.post/3mlfp4us56qg2"
  },
  "path": "/t/anti-llm-sentiment-considered-harmful/14008?page=5#post_93",
  "publishedAt": "2026-05-09T05:53:36.000Z",
  "site": "https://discourse.haskell.org",
  "tags": [
    "The GitHub Blog – 7 May 26",
    "Agent pull requests are everywhere. Here's how to review them.",
    "link"
  ],
  "textContent": "This blog post is very relevant for this discussion. It’s about both submitting and reviewing AI PRs:\n\nThe GitHub Blog – 7 May 26\n\n### Agent pull requests are everywhere. Here's how to review them.\n\nA practical guide to reviewing agent-generated pull requests: what to look for, where issues hide, and how to catch technical debt before it ships.\n\nEst. reading time: 8 minutes\n\nIt also clearly lays out some of the shortcomings of LLMs and how human judgement can hold up a high standard.\n\nBeyond the article, I can also recommend a strong “throw it back” attitude in reviewing PRs. By that I mean setting and enforcing strong boundaries to save review time. One is mentioned by the article: ask for smaller PRs.\n\nThere are more. I recently had two outrageous PRs on a Lua project, using Claude to rewrite the whole thing in C#. See link and link. The first one I closed after writing a supportive comment. The second one I just closed immediately, and the third step would have been to just block the user. In this extreme case I only spent seconds looking at the code.\n\nFor the less extremely obvious slop, you can set boundaries by asking to explain what problem this solves, and what important decisions were made and what the alternatives were. I think it’s fair to first demand that there is a convincing story here before you look at the code. LLMs can answer those questions, but not well. If you get bad answers back, you can simply say you’re not convinced of the premise of the PR and are thus closing. Well engineered PRs imo have this story right from the start, and you don’t even have to ask.\n\nThe big challenge here is to set strong boundaries while also not being extremely discouraging for genuine people to contribute. There is balance there I think, but it is difficult to find. I think it already helps to have a “please and thank you” style of setting boundaries.",
  "title": "Anti-LLM Sentiment Considered Harmful"
}