Ask HN: 有人在为跨工具协作的 Agent 构建写入保证吗？
1 分•作者: ajaystream•5 天前
我们在构建生产环境代理工作流时遇到了特定的故障模式。来自合约的字段导致订阅更新不准确——日期相差一天。产品在应该更新时却被随机创建。税额没有写入税额字段，而是被实例化为全新的产品。每一次失败都像是技术上成功但操作上错误的写入。
HITL（人机协同）有所帮助——一次处理一份合约，并在每一步都进行用户确认，从而保证了准确性。但用户最终表示：“我已经解释了 30 次了，赶紧把它做完。” 当我们减少确认步骤以使其运行起来时，它又开始失败了。
没有错误。没有警报。只有在几周后的对账中才显现出来的偏差。
提示和映射表在边缘上有所补偿，但从未奏效。代理程序没有关于字段在系统之间如何关联的已验证的真实依据——它每次都在推断。而且大多数时候推断得不一致。求助？
查看原文
We ran into a specific failure mode building production agent workflows. Fields from contracts creating inaccurate subscription updates — dates off by a day. Products created at random when they should have been updated. Tax amounts not written to the tax field but instantiated as entirely new products. Every failure a plausible-looking write that succeeded technically and was wrong operationally.
HITL helped — processing one contract at a time with user confirmation at every step kept it accurate. But users eventually said "I have explained this to you 30 times, just get it done." The moment we reduced confirmation steps to let it run, it started failing again.
No errors. No alerts. Just drift that showed up in reconciliation weeks later.
Prompting and mapping tables compensated at the margins but never held. The agent had no verified ground truth on how fields related across systems — it inferred every time. And most times inferred inconsistently. Help ?