GLM1 min read
Google Let AI Agents Rewrite Their Own Harness, and GLM-5 Gained 11.5 Points Without New Weights
Harness² edits the prompts, memory, tools and guardrails around a model instead of retraining it. Here is how the loop works, where the…Continue reading on Towards AI »
By Aya MahmoudTowards AI

Read the full story
Originally published by Towards AI on Oct 22, 2026.
Comments
Sign in to join the discussion.