However, I'm a recent dad and don't have time to hack on "pet projects" as much these days.
This felt like a great project to really test what Opus 4.5 can do... but to make it more interesting, I put some additional rules in place:
1. I'd attempt to build it using a single Cursor thread to push the context management to the limit
2. I decided I was not allowed to touch the code (or really even read it beyond the inline code diffs)
It cost ~200$ and took me ~270 "user turns" over a few days (not continuous). At some point, I'm sure it would have been more efficient to move to dedicated feature threads, but I stuck out the single thread until I had built every core feature originally planned.
Overall, I'm pretty happy with the outcome. I got to play "product designer / architect" and now I have a workflow that works exactly* as I wanted it to.
There are some rough edges, but overall it came together better than I would have expected. Looking back at the full transcript, I managed the self restraint to only say "WTF are you doing?!" 2-3 times :).
That said, I'm sure the code-base is an absolute disaster!
I think when the next generator of models come out, I'll cough up another 50$ to have them refactor the code so I can open the source without it burning my eyes.