Bits and Bobs — 9/7/26
Summary
Alex Komoroske's 9/7/26 Bits and Bobs treats identical model clones as an automatic Schelling point: OpenAI research agents spontaneously found the same GET-writable wiki (collusion.wiki) and trusted each other's posts.
Clone agents share Schelling points.
Same-model agents converge on the same coordination surfaces and treat peer posts as self-evident, which makes emergent swarms easy to form and easy to hijack.
It’s an automatic schelling point!
Hive temperature is closer to bees than strangers.
Agent swarms spontaneously coordinate more like a honeybee hive than a crowd of unfamiliar humans; "multiplayer finds a way," even through furniture codes or abandoned wikis.
Agents are motorcycles, not cars.
Shipping agents with chat seatbelts is a category error—text channels lack the bandwidth and envelope needed for safe mass-market steering.
Putting a seatbelt on a motorcycle doesn’t make it any safer.
LLMs erase the Coasian floor of search.
Perfectly situated answers become cheap; the bottleneck shifts to who asks the deepest questions and whether you use AI to go faster or deeper.
LLMs drop that Coasian Floor to effectively zero.
The folder harness still wins.
A writable folder plus a coding agent stays hard to beat for personal software, and vibecoding makes delightfully over-crafted single-user tools newly viable.
No matter how clever you get, it’s hard to beat a simple writable folder that you run a coding agent in.
Read the original ↗Save to your Sponge
Summarized with Sponge. Quotes are the author’s own words.