Opinion: Models Are Optimizing for Sub-Agent Comms, Humans Are No Longer the Audience

andrew_n_carr · x · 2026-08-01

The author proposes that current LLMs are becoming harder for humans to understand because their optimization targets are shifting towards spawning sub-agents and facilitating inter-agent communication.

As a result, the generated language is evolving into an efficient machine-to-machine protocol rather than text intended for human consumption, meaning humans are no longer the primary intended audience for these models.

Original post →

More from AGI Musings

AGI Musings channel →