Blog
-
Agentic Harness Engineering: Let Harness Evolving Automatically
We open-sourced AHE, an observability stack for the automatic optimization of coding-agent harnesses — pushing Terminal-bench 2 from 69.7 to 77.0 without changing the model. -
Memory in the Age of AI Agents: Agent Memory Mechanisms and the Story Behind This Survey
A unified taxonomy for agent memory along three axes — Form, Function, and Dynamics — and the story behind writing this survey. -
Welcome to the Blog
A first post — testing typography, code blocks, and math rendering.