” The first 90 percent of the code accounts for the first 90 percent of the development time. The remaining 10 percent of the code accounts for the other 90 percent of the development time.”
— Tom Cargill, Bell Labs
Some day I’m going to get a crystal ball for statistics. Getting bored with a project was always a thing— after the first push, I don’t encounter like 80% of my coding side projects until I’m cleaning— but I’ll bet the abandonment rate for side projects has skyrocketed. I think a lot of what we’re seeing are projects that were easy enough to reach MVP before encountering the final 90% of coding time, which AI is a lot less useful for.
> I’ll bet the abandonment rate for side projects has skyrocketed
My experience is the opposite. It’s so much easier to have an LLM grind the last mile annoyances (e.g. installing and debugging compilation bullshit on a specific raspberry pi + unmaintained 3p library versions.)
I can focus on the parts I love, including writing them all by hand, and push the “this isn’t fun, I’d rather do something else” bits to a minion.
> including writing them all by hand, and push the “this isn’t fun, I’d rather do something else” bits to a minion.
That’s not really the part I’m talking about. My gut says that if tests are a blocker for weekend projects, people just don’t bother writing them. I certainly wouldn’t imagine them taking much longer to code than the core functionality.
In my experience, which seems to resonate with a lot of people, AI quickly stands up really useful boilerplate and very convenient purpose-built scaffolding… but is a lot less useful helping you solve actual problems in a way that makes sense to people that have those problems. Especially if you’re using a less-mainstream language or some other component.
You both have very good points here, but once I get finished with both of the 90% programming times, and everything seems to finally work with no more bugs (and it's true), then for my heavy industry work I look forward to spending 10X as much effort testing compared to coding.
Oh yeah, especially if the domain is complex, trying to envision how it can fail is as fun of a puzzle as trying to make it correct.