Stop Telling Me to Ask an LLM

The author recounts an experience: seeking advice on a problem without industry consensus, she specifically called a senior person, hoping for judgment honed by thirty years of experience beyond textbooks. The response was: "Honestly, ask Claude." This is not an isolated incident. She has been brushed off with the same phrase on multiple occasions, and each time she contacted a real person, she had already spent two hours back-and-forth with the LLM, burning through many tokens.

This paradox highlights a new pain point in seeking knowledge: when human experts present LLMs as the ultimate answer, they ignore that the questioner has already gone beyond the LLM's limits. The problem was brought to a human precisely because the model failed to provide a satisfactory answer.

The author compares this new phenomenon to the early "LMGTFY (Let Me Google That For You)" but sees a key difference. LMGTFY was aimed at lazy questioners, while "Ask Claude" targets those who have already done their homework. A more precise analogy: you ask a friend for late-night restaurant recommendations, and the friend sends you a list of the top ten on Dianping—when what you really wanted was a personal suggestion based on shared taste and past experiences.

In other words, "ask the model" may have become a polite shell for "I don't know," "I don't have time," or "I can't be bothered to think." The author acknowledges that being asked for help does have a cost, but rather than fobbing someone off with an LLM, it is better to say "I'm busy" or "I can't think of anything you haven't tried"—these are honest answers that at least preserve candor.

The author emphasizes that what she seeks are "scars": the real experience gained only by witnessing decisions go wrong in a boardroom. Such experience cannot be obtained through search or the statistical patterns of an LLM. It encompasses personal perspective, historical context, and intuitive judgment—things a model cannot generate even when trained on vast data.

When "Ask Claude" becomes the standard answer, we effectively cut off the transmission of this unique knowledge. The questioner loses deeper understanding, and the responder avoids real thinking. Over time, this can erode trust in human relationships, reducing knowledge sharing to algorithmic interaction.

The author is not against using LLMs. She acknowledges that many questions can indeed be solved by models or search engines. But she points out that when the question goes beyond these tools, "Ask Claude" does not save any steps—it merely withholds the thoughtful, experience-based answer that should have been given.

The cost of honesty is indeed high: it requires time, focus, and real thought. Not everyone has the bandwidth amidst a busy schedule to bear that cost. But using an LLM to avoid it does not reduce the questioner's work; it only increases their disappointment. A healthier approach is to admit one's limitations or offer a brief, personal opinion. After all, the truly scarce resource is not the answer a model can provide, but the living, judgment-laden experience of a human.