Me: listening to a podcast during workout about dopamine depletion and realizing:” Wow modern capitalism really trained us to chase productivity that only creates more emptiness”
Also me: rowing harder so I can absorb the podcast faster and metabolize my existential crisis
Basic as it sounds, AI today still doesn’t know when to chime in & execute in a group chat, and when to listen without responding, which is essential to make any human team work in real life! the question remains if this should be solved on the model level or on the product level
Single-player mode with agents has gotten incredibly advanced - virtually every white collar job function in the world can now access coding agents on device or in the cloud.
Multi-player is still super primitive - people sharing ChatGPT sessions with each other, copy and pasting between their own AI sessions and Notion, Google Docs, etc. To do this right we need to solve hard problems in context, memory, integrations, and permissions management across an entire team.
We're starting to see glimpses of the future as we build in multi-player AI. One interesting angle is how to balance between "communication" and "execution" between multiple humans and AI. Will share more learnings soon!
Claude Sonnet 4.6 just dropped and it's now available on Coral 🪸
Opus-class intelligence at Sonnet pricing. Your AI employees just got a serious upgrade.
As usual with new model releases, the benchmarks are impressive. Details below 🧵