I don’t want a screenshot of your Claude conversation

I'm now calling people who do this "LLeMmings" or "Meat-based LLM proxies". I mean how embarassing is this. Not to mention rude. Someone is trying to have a conversation with you, and that sometimes includes questions you have to think about for more than 0.1 seconds. The purpose is to find out what's in your head and see how you reason through things. If I wanted to know what an LLM would say, I'd go ask it myself.

I stumbled across a conversational technique yesterday which worked well.

The team was trying to understand a tricky programming problem. They had made a change to the logic in component A, which made the P99 execution time of the system take about 30% longer. Logs seemed to show that about half. This was due to slowdowns in component A and half with two to slow downs in component B, which hadn't even been changed. I'm sure there is some underlying issue, and perhaps by this time the team has figured it out.

But a director level manager was weighing in and trying to direct the work. She suggested they should try asking Claude. The team had done this, but the answer wasn't especially useful. The manager was pushing harder, saying they should really lean in on trying to follow the advice from Claude.

Then she happened to also mention that they should try searching on Google. That was the trick I leaned in on. I immediately said something like, "Yes, asking Claude and searching on Google are both reasonable starting points to get ideas of what to investigate." Then a couple more times in the conversation I mentioned "searching Google" and "asking Claude" together in the same breath.

You see, these tools ARE useful... but they are not a magic panacea. This manager already has a sense of the degree to which "search on Google" is useful (quick to try, often has useful information, lacks anything specific to this problem?) and by mentioning the two together I am emphasizing that "ask Claude" has very much the same set of limitations.