Huahai replies to Rich Hickey’s argument that language models are automation rather than tools. The disagreement is about control: does asking a model to do work necessarily mean surrendering the judgment and learning involved in doing it?

Huahai argues that it does not:

  • Automation can be part of a tool, as it is in compilers and spreadsheets.
  • A programmer can question an explanation, inspect the original code, test a change, reject it, and try again. That is a human-directed feedback loop.
  • Using a summary instead of reading a paper can cost you understanding. Reading the paper and using a model to challenge your interpretation is a different activity.
  • Unpredictability is a limitation to evaluate, not proof that something cannot be a tool.

The contribution is a more actionable distinction: skilled use versus delegation without adequate understanding or oversight. It leaves room for useful assistance without pretending that accepting an answer is the same as checking it.

The 211-comment thread on Hacker News adds concrete failure modes and disagreement about which skills survive model updates.

What the thread adds

  • vanuatu — on the verification burden: “they cannot be held accountable - therefore, you need the skill and expertise to verify outputs, or set up a verifier”.
  • gerdesj — reports a troubleshooting answer that was largely useful but invented a configuration detail: “it didn’t have access to exim4.conf and made it up”.
  • dzink — argues for keeping generated projects understandable: “The job of professionals now is to contain the complexity to keep projects maintainable and manageable”.
  • extr — questions how durable model-specific technique is: “those skills are all obsolete now”; their proposed alternative is “being able to communicate clearly what you’re trying to do”.
  • modo_ — pushes back on that obsolescence argument: “there’s still a gap between what these tools can do and what most people know how to get them to do”.

Handles are pseudonymous. HN publishes no per-comment scores; ordering is HN’s ranking. This is a slice of the thread, not a consensus.