Lai Xinlu

Lai Xinlu is the Share AI founder interviewed in 探秘 Claude Code，搞懂 Agent Harness｜对谈来新璐 about Agent Harness design. He argues that harness means the model-external system around an agent: execution tools, context and state, memory, compression, handoff, and governance across multiple agents.

His technical position is model-centered. He accepts that harness design strongly affects what agents can do, but he says the model remains the first source of agent intelligence, followed by context and then tools. He favors “more context, less control” and CLI/Unix-style agent interfaces over brittle prompt-node or flow-graph orchestration.

Connections

• Share AI — company Lai founded.
• Learn Claude Code — Share AI learning project used to analyze Claude Code as an agent harness sample.
• K Computer — lightweight Unix-style virtual computer in Share AI’s K-series toolchain.
• Agent Harness, Agent-Facing Interfaces, and Context Engineering — main technical frames in the conversation.
• Model Harness Co-Evolution — adjacent wiki concept that Lai sharpens by stressing harness alignment with model operating logic.