Models
No feature ships until a second AI model has tried to break it.
AI models build most of Bench. The model that writes a feature never gets to approve it. A separate model tries to break it first.
Models
Your agent doesn't forget last month. Old conversations get filed, not deleted.
An AI model can only hold so much at once. Most tools drop the oldest messages. On a Bench harness they get filed and summarized, never deleted.
Models
The job picks the model
Bench starts with Grok because it is fast. Opus is the model for a normal Tuesday. Fable builds a system that keeps doing the work, not one quick job.