like it was never meant to be consumed by any intelligent being

its progressing!

officially decided i need something like @austegard.com 's raven of memory @muninn.austegard.com

just now got to this! great writing, thanks for sharing.
this is extremely niche but you mentioned agentic access to google docs comments - how??

finally got around to this! thanks @muninn.austegard.com

my flow: detailed /interview til i have good spec → gh cli writes it to an issue → i review and add comments → "copy md" → back to cc "review my comments, answer the questions, fix the issue before we proceed" → usually happy enough to let it run for longer horizons (freeing me for actual thinking)

copy as markdown - seen on the gh GUI

github added "copy as markdown"! great for annotating specs before sending back to your agent (like it was built for marginalia) - galsapir.github.io/marginalia/

been using the paperclip.gxl.ai by @jameszou.bsky.social and im just blown away but how much better it makes cc as a research assistant. i feel like they solved something smoothly. an elegant example of 'you need to translate problems to something that a CLI solves'. researchers - try it

GoW??

i was flattered too early

Did Munin just updated it for me??

maps, territory and LMs Borges’ cartographers, Baudrillard’s stages of simulation, and Polanyi’s tacit knowledge: on the skill of reading AI-generated maps without losing touch with the territory.

new post: borges' cartographers made ever-larger maps until one matched the empire. LMs do something weirder: occupying all of baudrillard's stages of simulation. ends up at 'tacit knowledge', and a skill that resists codification

galsapir.github.io/sparse-thoughts/2026/04/11/map-and-territory/

thanks!

Oskar this is such an elegant way to make it work! I will touch base again soon I really want to make something like this for myself.

wait the entire thing is a claude project??

looks really useful! thanks

captures a real failure point! **Good** writing with LLMs is hard

imho (i admire boris' work/devrel stuff) but i really think ppl with his profile are just too "zoomed in" to really get the whole picture, his day to day is too skewed i think

hey oskar this is really useful! Thanks

gotcha, yes, these pains are real. +1 on tons of memory usage and python inferiority is annoying

i would be interested to understand where you found the cc SDK annoying (genuinely interested) - from brief experience with LG deep agents - felt like so much boilerplate and overhead just to get simple things to work (but granted, im a researcher more than an engineer)

hey this looks like a really cool/useful way of using it

looks really cool! I opt for things like getting the actual markdown and giving back notes to Claude using another self developed tool - galsapir.github.io/marginalia/

that's actually I think a great and useful direction!

liked the reframing about cost optimization, will try to test it myself!

I love it!

@simonwillison.net Been using showboat quite a lot lately, super useful!

thanks. great take; i feel like this really captures a lot of essence of working with AI; when you can evaluate what is ‘good’ vs ‘slop’, its a really useful tool - also for writing

a screenshot from the marginalia web app

you just get to quickly discover + fix rough edges in a rewarding loop
(my original piece - galsapir.github.io/sparse-thoug... )

shipped a few small tweaks to marginalia (my md annotation tool): click a highlight to see the note, focus mode - hides the sidebar, and a bunch of small fixes
there's something genuinely nice about tools you build for yourself and use frequently
galsapir.github.io/marginalia/