If you've ever used Anthropic's Claude LLM, and in particular, Claude Code, to discuss and plan software architecture and systems you'll definitely start hearing some canned terms. LLMs have "tells". It can make reading some supposedly-human authored things like tech blogger posts somewhat amusing, but hey I still have unwavering respect for Sam and his accomplishments!
Is it dishonest? I'm honestly not sure. Would it have been more honest for Sam to strip out all those Claude crutches? He's been honest from day one about Roundhouse's journey and his use of Claude, so I think the blog posts are an amusing admission of just how much Claude is doing here. It also makes me feel a bit better because I was finding some of the posts hard-to-follow, mostly because I have other things going on and I know nothing about Ruby-on-Rails or that ecosystem.
Nation leaders don't credit speechwriters or their hair / make-up teams. CEOs don't list every software engineer, UX designer, product manager, etc that contributed to a product / service. We also don't list every developer that contributed to a language compiler or an art tool that was used to make a video game.
The floor is always going to feel vaguely shoulder-like.
I'm a fan of the Golden Age of Animation, so my mind drifted to that as a comparison... Was it dishonest for Chuck Jones and his chief animators to take all the credit for a cartoon when there were binders full of women in-betweening? Probably a little dishonest - but hey someone was in charge, so it makes sense to give credit to whoever was accountable and/or marketable (Chuck Jones was a "brand"). The industry at the time knew how cartoons were made, so it was semi-obvious that Chuck Jones didn't sit and literally draw and paint every single cel. Maybe not to that kid watching the cartoon on Saturday morning though.
Today we are re-learning how the software engineering industry works. From my own experience, I've watched LLM code assistance drive my velocity from something like 1.5x a year ago (mid-2025) to what feels like something much larger today (mid-2026). Maybe as much as 10x. It depends on the task/project and these are just subjective feels and we know that it's hard to properly gauge velocity. But I'm guessing that at some point in the future, maybe I won't even be able to quantize it, because writing code by hand will become such a specialized skill (in the same way that writing hand-tuned assembly is today). We'll just measure by the things built and shipped.
I liked it when Sam credited his success with Roundhouse to the skills he learned and used throughout his IBM career: Ask smart people to answer the hard questions. It's interesting, because I also interpret my LLM coding agent experience through the lens of my own career: as a Tech Lead who writes and reviews design docs, provides architectural guard-rails to the team, reviews code changes, highlights code smells, and helps prioritize projects/roadmaps. As a result of this lens, I haven't pushed too far into true vibe-coding - I still scan the LLM-proposed code changes and I've caught many instances of bugs and impending tech debt. This makes me still feel useful as an Engineer rather than a Product Manager or a UX designer. I don't know for how long. Knowing the right thing to build and the next task to pick up seems like a pretty human thing still.
What I know is that no one knows where this is going to finally lead. It is uncharted territory for human civilizations to have easy access to intelligent, non-human slave labor. What makes reaching consensus with groups of humans hard goes away when it's a single person project with a bunch of "yes-men", but it doesn't necessarily mean that everyone should become a solo artist. It makes sense when you're trying to get a project off-the-ground or scratch an itch no one else yet cares about.
It's a bit funny that the one time I met Sam face-to-face, we ended up lamenting about how dysfunctional the WHATWG/W3C was. Humans are hard!
In trying to predict the future here I had one observation: today's frontier artificial intelligence are rapidly becoming a commodity. I'm not sure what happens to those players when AI appliances that can run decent local models become affordable and common-place for consumers. I do get the sense that there is a "good enough" bar that is some N quarters away. As the bar lowers, we will surely see a lot of fall-out: very suss marketing, lock-in/bundling plans to keep folks token-spending, regulation around LLMs that will make running locally harder or impractical, and some interesting law suits emerge to attempt to shut down open models.