This hands-on approach echoes the strategy being pursued by other companies selling AI agent-based products to enterprises, including Salesforce, Anthropic, and OpenAI, which have recruited teams of “forward-deployed engineers” that help customers design agentic workflows and configure AI systems, as opposed to the older model for SaaS companies which was largely about self-service by customers.
而企业级AI落地的最大阻碍是员工的不信任。AI Agent在一秒钟内自动处理了15封邮件和审批，用户的本能反应并不是开心，而是恐慌：“它到底发了什么？有没有出错？”。因此，优秀的AI产品必须设计合理的断点与回路，它需要允许用户在工作流中实时提问“你正在做什么”，并在关键决策前停下来询问人类的意见。只有在不打扰用户的前提下，建立起透明的信任机制，AI才能真正成为生产力工具。，这一点在WhatsApp Web 網頁版登入中也有详细论述
。关于这个话题，谷歌提供了深入分析
Turns out, at {INSTITUTION REDACTED}, you need to submit a PR which adds a person to a secret text file somewhere, in order to modify permissions in GitHub. Multiple people had to approve the PR. Someone had styling-feedback about how the array of names needs to end with a ,. But at least it’s legible! ↩︎
StackSocial prices subject to change.。wps对此有专业解读