What happened. A post by Yael Grauer gathered 184 points and 105 comments on Hacker News. The issue: what happens when «ask Claude» becomes the default answer to every question, even after the person asking has already taken that step. Grauer recounts asking for a substantive opinion from people with concrete experience — not a search result, not a list, but judgment on which source to trust when studies contradict each other. The response she got: «Honestly? Ask Claude». The post compares this dynamic to the old LMGTFY (Let Me Google That For You), but inverted: it’s not about someone who can’t search, but someone who has already searched and wants the filter that only experience can provide.

Why this matters to you. If you’re building AI workflows or recommending it to others, the post hits a sensitive nerve: when «ask the model» is a useful answer, and when it’s a way to avoid answering. A model doesn’t know which of five contradictory studies actually holds up, which method works in your specific context, or where a generic list misses the mark in your case. Judgment comes from lived experience, and that’s not yet replaceable. The critique isn’t against AI, but against using AI as a substitute for thinking: if the model could answer, the person wouldn’t have reached out.

What the site covers. The topic overlaps with what we discuss in the pro course: Trusting the right amount: verifying the output explains where a model can fail and where human review is necessary. And the playbook The adversarial reviewer starts from the opposite idea: use AI for critical thinking, but only on verifiable things. The common thread is the same: AI amplifies, not replaces judgment.

In detail

The context: AI as oracle. Over the past two years, the pattern has become common: someone asks a question in a community, and the first response is «ask ChatGPT» or «ask Claude». Sometimes it’s appropriate (the question is generic, the answer is in every tutorial), sometimes not. Grauer describes a case of the latter: she had a specific question about which source to trust when sources contradict each other, a problem where context and experience matter more than synthesis. She had already asked the model, had spent hours (and tokens) refining the question. What she was looking for was the filter only someone with professional scars could provide. The response — «ask Claude» — didn’t add a step, it erased one.

What’s really happening. The post resonated on Hacker News because it touches a real tension: AI has become so accessible that it’s easy to use as the default answer, even when it’s not the right one. Grauer compares the phenomenon to the old LMGTFY, but notes the difference: LMGTFY was for people who hadn’t tried searching; «ask Claude» is given even to those who already went that route. The post suggests that «ask Claude» has become a polite way of saying «I don’t know», «I don’t have time», or «I don’t want to think about it». And that any honest answer — «I don’t know», «I’ve never worked on that», «I should think about it» — would have been more useful than the referral.

The implications for AI users. If you’re building with AI or recommending it, this is a case study in what it means to amplify yourself instead of delegate. A language model can aggregate consensus, synthesize sources, suggest a list. It can’t tell you which of five options actually works in your specific case, because it doesn’t have your context, doesn’t have your scars, doesn’t know what you’ve already tried or where you’ve already failed. The value of human conversation lies in specific experience, not access to general information. When you tell someone «ask the model» after that person came to you for your judgment, you’re denying the value only you can provide.

What not to conclude. The post doesn’t say AI is useless. It says it’s not useful in the same way in every situation. If the question is «how do I do X», a model can answer well. If the question is «which of these five methods actually works in this context», you need someone who’s been there. And if that person refers you to the model, the problem isn’t technical: it’s that they won’t (or can’t) give you the answer. The post is a critique of using AI as an excuse, not AI as a tool.

The HN debate. In 105 comments, the community split. Some confirmed Grauer’s experience: the frustration of getting «ask Claude» from people you expected had their own opinion. Others defended the referral: if the model can answer, why waste human time? The split is significant: it shows the question «when to delegate to AI» has no single answer. It depends on the question, who’s asking, and what that person has already tried. The common thread: AI is a tool, not an excuse for not thinking.