Kris Brandow
Works on Go tooling and hosts Fallthrough, a podcast for Go developers that picked up after Go Time ended.
“In the long run, we will not be using LLMs to generate writing, like prose. I think they are so subpar at it, and I don't really think they're going to get better. If I start a YouTube video or a short, I can tell pretty quickly if the script was written with AI, even where it's been heavily rewritten, because the underlying voice and cadence is just this very average, monotonous type of prose.”
Fallthrough · Hands-On Engineering Podcasts · November 2026
Context
The hosts were discussing a drop in LLM-generated blog posts and company sites. Kris says AI-written prose is easy to spot, pointing to tics like the rule of three.
Transcript
…a catastrophic, maybe company ending problem. So you need to make sure that your software is solid because you're deploying it and to air-gapped machines that you have no ability to troubleshoot. So bugs in that are like exponentially more expensive than bugs in your little personal software app.
…people are starting to learn some of these long-term lessons on how they want to use AI. So, and at the same time, we're still learning more every day. It's like this weird mix of like settling and the comfort with acceleration and understanding that you do it to constantly re-evaluate what you do.
Yeah, I do think that in the long run, we will not be using LLMs to generate writing, like prose. I think that they are just way, they are so subpar at it. And I don't really think they're going to get better. Like as soon as I, if I start a YouTube video or a short, I can tell pretty quickly if the script was written with AI, even in circumstances where it's been heavily rewritten, because like the underlying voice of it and cadence of it is just this very like average, sort of monotonous like vert type prose. I don't know what to say there, but like the way that it's written is in a specific way that's easily identifiable as like either Claude or ChatGPT. You're like, oh, this was definitely, you can definitely hear when it was written by ChatGPT. There's certain ways that it structures its writing that is okay if you're interacting with a chat agent, still sort of annoying, but definitely annoying if you're watching a YouTube video because it just doesn't, it's just like, it's like basically hammering you all of the time. Like in Rain's episode, they brought up the rule, like the rule of three, where you say three things or use three things to contrast, and it just does it…
No, no, sorry. I didn't mean to interrupt. I mean, I just, I, I, I, I think the, there's a social contract, right? I produce something that I want you to watch or read or listen to. So why should you put in that time if I didn't put in the time to talk to another human? And almost like it actually…
On the topic of slop, I would say that we also need to dial in what the definition of slop is. Because there is this, there's this lobsters thread that Mitchell, you know, guest, multi-time guest on the podcast, Mitchell Hashimoto, tweeted out the other day or X'd out. So stupid. That he, he posted…
Yeah, absolutely.
More from Fallthrough
