Lilian Weng summarizes 35 papers on Harness Engineering for RSI

Lilian Weng, cofounder at Thinky, published a summary of 35 papers analyzing the role of harness engineering in recursive self-improvement (RSI). Her work highlights that while harness improvements may eventually be integrated into core models, the fundamental need to specify goals and context remains constant.

• The post identifies main proven design trends in harnesses relevant to RSI.
• It recaps optimization literature, including the ACE paper and recent Meta-Harnesses trends.
• Anthropic expanded its "background agent" UX with Claude Cowork on mobile and web.
• Anthropic extended access to Claude Fable 5 on paid plans through July 12.

Weng's analysis suggests that harness engineering is becoming central to agent design, offering insight into the strategic focus of her new venture, Thinky.