{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreic363khomgk7v4lq5i57fhdpgsg3i6woxpuqunxyzis7wuvfglyl4",
    "uri": "at://did:plc:pi6woz4d47bkuws673w2il2r/app.bsky.feed.post/3mkutavcmuwc2"
  },
  "path": "/t/anti-llm-sentiment-considered-harmful/14008?page=3#post_48",
  "publishedAt": "2026-05-02T13:59:31.000Z",
  "site": "https://discourse.haskell.org",
  "tags": [
    "recent article in The Atlantic"
  ],
  "textContent": "I somewhat share the sentiments of both sides in this discussion, if calling them “sides” even makes sense:\n\n  * Side 1: Haskell is a great match for \"agentic LLM\"s, because of the strong type system etc.;. Let’s embrace the new technogoly “agentic LLM”.\n\n\n\nVersus\n\n  * Side 2: Let’s be careful; LLMs are probabilistic models, they hallucinate, they are over-confident; let us question every single bit these tools spit out.\n\n\n\nI also have been critical towards LLM-assisted coding, or even commits generated entirely by a large language model. I still am, but my position has shifted quite dramatically.\n\nThere are estimates that by now (May 2026) about 50% or more of new code is written solely by AI (it is hard to tell, and I didn’t find any statistically well-founded sources, but the trend is quite clear).\n\nI consider the most likely scenario is that in a few years time, _most of the code_ will be written by LLMs or a newer technology outperforming current LLMs.\n\nNow, we are Haskell programmers. We love Haskell, and our shared goal is to increase the adoption of Haskell, and to ensure our ecosystem improves and continues to thrive.\n\nI believe this can be done best by taking _both_ sides: Embrace the technology “agentic LLM”, learn how to code with agents; _take the best out of this new technology_ ; while being extremely critical towards AI-generated code. If the review burden is too high, let an LLM help you review a PR that you are unsure about; do not blindly trust the results of the LLM-assisted analysis, but take them with a grain of salt. LLMs are pretty good at detecting sloppy code.\n\nA recent article in The Atlantic brings it to the point:\n\n> Six months ago, people arguing that AI was a bubble were pointing to real-world facts, whereas people arguing against the bubble hypothesis were making speculative promises about the future. Today, the roles have reversed. AI’s explosive growth may yet encounter some new unforeseen obstacle. But the burden of proof has shifted to the naysayers.\n\nWe have to be clear: In a few years we most likely will either be writing code with an agentic assistant (or some newer technology), or we will have moved on to some other work that can’t be done by an agent.",
  "title": "Anti-LLM Sentiment Considered Harmful"
}