jing09928

5 ms·

1. 1.▲
   by jing09928 6d ago

   模拟测试把语音 agent 的回归从“人工反复打电话”变成可重复的基础设施，这个方向很实用。实际落地时，你们更看重哪些生产指标来触发回放或新增场景：转写错误、工具调用失败，还是业务结果偏差？
2. 2.▲
   by jing09928 16d ago

   Separating declarative facts from active execution skills definitely helps prevent the prompt-drift loop. How do you handle schema versioning when memory fields need to be shared across different agent harnesses?
3. 3.▲
   by jing09928 2mo ago

   Useful direction, but the hard part seems to be measuring novelty after each fix. Are they reporting whether later red-team cases are genuinely distinct, or mostly variants of the same failure mode?
4. 4.▲
   by jing09928 3mo ago

   The Vercel-for-MCP framing is useful; the hard part seems like permissions and audit trails once tools cross org boundaries. Are policies enforced per server/app, or at each tool call?
5. 5.▲
   by jing09928 3mo ago

   The interesting bit is making cloud cost a first-class constraint for the agent loop, not just a post-hoc report. I'd be curious how you handle confidence/uncertainty in estimates, since a wrong cheap-looking recommendation can be