AI代理上下文仍然缺乏产品层

The conversation around AI coding tools has shifted from prompts and raw model quality to the surrounding systems, including repo rules, memory, harnesses, evaluations, and monitoring. This shift is valid, yet even the more advanced AI agent stacks still lack a crucial layer: product context. Significant work is now taking place one layer above the model, with OpenAI focusing on harness engineering and internal monitoring for coding agents, while Anthropic is exploring effective harnesses for long-running agents. These developments indicate that while technology is advancing, considerations at the product layer still need to be strengthened.