(It also triggers a recursion limit bug in CC after an hour or two, but I've reported that 🙂)

Do you have trouble getting Claude not to stop? I eventually created a stop hook that tells it (kindly) to keep going, and to ask sub-agents if it needs advice/help. Every hour or so, it tells it to ask subagents to review code and look for architecture/refactor opportunities. Seems to work ok…

Oh, also, I am absolutely _not_ a frontend person.

Crazy times, my programmer friends, crazy times…

Now I want to give Claudius sub-agents whose context is stuffed with apenwarr Google housing market simulation articles and can run simulations and report back. MBA in a box! 😂

Does it specify the tokenizer, or is everyone just using OpenAI's Byte Pair Encoding?

I should mention that it was (as always) a thoroughly fascinating episode!

If you have a chance to follow up, I've be very curious to hear what Josh thinks about Kotlin coroutines vs Java Lightweight Threads

He seemed bullish on both, but from what I've been able to gather, they don't love each other! At the very least, Kotlin *does* have (a bit of) function coloring…

He did a pretty good job of appearing to ask a couple of tough questions while ensuring that the experience was completely friendly to Vance. Joe Rogan style!

No joke, nothing has made me think more about distributed, anonymous government-proof communication more than listening to the Martian Revolution, and thinking about what that level of control over information dissemination would mean 🙂

Us computer nerds are building some really cool stuff out there!