How This Post Was Made
Prompt 1 (the request for this post): “This would be a great blog post, I think. Check the code at trevorturk.github.io. I think this is one of those examples where if you look at what I wrote in Slack versus what I was able to produce using Claude, it’s obviously so much more useful. There’s an ongoing debate about whether or not you should send people Claude output that you haven’t read yourself or that you didn’t author yourself, and I stand firmly on what I think is the minority opinion that it’s perfectly fine and in fact sometimes very good to just send stuff that Claude has written over to a coworker. If you look at the difference between what I had time to write in Slack and what I dictated into this session versus the artifact that Claude produced in terms of this GitHub issue for [a colleague] to follow, it’s just unbelievably valuable. […] The blog post structure should be using Hello Weather to serve as an example for how to do this kind of thing, having Claude generate an artifact, and giving a short idea of what the issue looked like. And then also a section at the start of this article about how I think this is actually really useful and I don’t think it’s offensive for me to send this to [a colleague]. I think it’s incredibly helpful and it’s going to help him a huge amount. Perhaps much like people need to quote unquote get used to AI, you know, like AI psychosis and all this kind of stuff, there’s a big debate about reading slop. I think I’m on the side of people need to get over it, just like programmers need to get over the fact that Claude can write code now.
Also people need to get over the fact that sometimes they’re going to be reading things that were generated by an LLM, and they can do a great job and be very useful. Maybe one trick that people could take away from this is that oftentimes I create things with Claude that are meant to be consumed via Claude. So [a colleague] is maybe going to skim this issue and then open a Claude session, paste in the link, and sit and chat with Claude about it to decide what is useful and what’s reasonable and how he wants to approach things.
He might not even necessarily read the whole issue, but I think it’s full of valuable information that is giving an LLM a great amount of context.”
Prompt 2: “One thing that I would like to also emphasize here is I am cooking dinner for my three children while I’ve done all of this, including the blog post and sending all this information to [a colleague]. And that is just not something that I would have had time to do otherwise without using Claude.”
Prompt 3: “Also can you add a little shout-out at the end about the VoiceInk transcription tool for the Mac, which I’ve really been enjoying and also is so helpful for getting your full context and everything into Claude.”
Generated by Claude Opus 5.5 using the blog-post-generator skill, in the same session that researched and filed the issue. Judgment calls: the limits on when to send model output (personal messages, apologies, hard news) are the model’s addition to the author’s argument, and the “AI psychosis” comparison was left out because it’s a different topic from reading model-written documents.