here's an interesting idea, within a company / team:

- a single person is responsible for all the production code, kind of the way an architect was

- everybody else can slop around as much as they want, the key is that at the end comes out a good document / a good feature (could be just a screenshot or literally 2 sentences: "yes this is good"), and the single integration person is responsible for prompting it reliably into the main codebase, instead of reviewing PRs....

that integration human role can be passed around, or some people will find that they really enjoy it (I know I do), and it removes that whole "review 200kLOC of slop". From the get go you kno they're destined for the trash can, so there is no animosity or so. The responsibility is to get the "core" of your work across.

#llms #vibecoding #llm #claudecode

@mnl What’s the feedback loop for the everybody else? (I.e. when it turns out their screenshot/doc missed crucial information and the agent-whisperer doesn’t have the right context to recreate it, how do they find out and learn from it?)

(This thought comes from my team trying a bit to do that to ourselves: we’re *all* practising using the slop to refine the doc then simply starting again fresh, a few rounds really helps to clarify the core idea.)

@tikitu good question, I think it’s obviously traditional team communication techniques, but I think new tools and workflows will emerge (most certainly very tailored to teams). I have no idea but I think it’s a very exciting new domain to explore.

One thing we are playing with with a friend is that every idea implementation comes with a fully functional dynamic essay (think Bret victor) to work people through, which also has the side effect of putting the implementation through its paces. Definitely more pleasant than a pile of red/green lines.

We’re kind of talking how teams are now more like little xerox parc research entities, and that you there are a lot of workflows from science that now apply.

@mnl yeah we’re exploring some similar ideas.

Writing the code : reviewing the diff :: prompting an agent : ???

We’re investing a lot more in (deterministic) tools to show interesting things about the new code. (Simon Willison’s showboat tool is in a similar space I guess.) Feels like Jupyter notebooks and similar tech should be applicable here. (I build mobile apps so, um, not so much for me.)

@tikitu look at https://artifacts.yolo.scapegoat.dev which is like 0.5% of my experiments in Claude. I then just download the barf and recreate it in my codebase. More complex vibe slops are the date numbered repos on my GitHub, after something like GitHub.com/go-go-golems/go-go-labs or GitHub.com/go-go-golems/vibes just wasn’t tenable anymore…
Claude Artifact Server