h.olysh.it

Heads up - there is a bug where subagents can switch models when the coordinator sends messages to them. I’ve been telling the coordinator to work around this: github.com/anthropics/c... I’m… not impressed with the code quality of claude code tbh.

Subagent model override silently dropped after a continuation boundary (SendMessage follow-up / post-compaction resume) · Issue #68147 · anthropics/claude-code github.com

Just released v0.5.0 of modelith.sh, my domain modeling toolkit. Check it out. github.com/stacklok/mod... Brings in model importing, vendoring and subTypes. Let me know what you think!

Modelith - Domain Model Tooling | Modelith modelith.sh

I moved Claude Code onto a spare NUC so closing my laptop doesn't kill a running session. 2 interesting things: - I was able to set up a seamless config that includes things like pasting images - This stuff is so *easy* when you are working with an AI agent. joe.dev/posts/alwa…

An always-on home for Claude Code joe.dev

Rubber duck debugging is when you just talk about your problems and the process of verbalizing them helps you solve them. One of the best ways to use AI, IMO, is as an "animatronic rubber duck". It can really help you sharpen your thinking. Blog post: joe.dev/posts/animat...

The animatronic rubber duck joe.dev

AI can make stupid people feel smart. None of are immune. That includes you. (Stupid people = stupid in a domain. People are multi-dimensional and IQ is bullshit.)

Working with an AI agent is both fun and enormously frustrating. You know that co-worker that is super literal and pedantic and can do amazing things but also frustrates the fuck out of you? That is the AI agent. And also, if you are that co-worker: I love you. Don't change. Keep being you.

The more I work with LLMs and agents, the more I'm gobsmacked by the n^2 nature of how conversations are handled. For every turn of the conversation, you send the entire previous conversation. This means the costs compound the longer the conversation goes on.

Sleeper Agents but for AI. (This is a real concern for foreign open weight models.)

@michaelhobbes.bsky.social @notalawyer.bsky.social please can you guys do an episode on the AI doomer sex cults? There is no book associated but don’t let that stop you! More people need to hear about how unhinged and bizarre this all is.