Vercel details how it built its inbound sales agent
What happened
On October 6, Vercel COO Jeanne DeWitt Grosser described how the company built an inbound sales agent. One engineer spent about 20% of their time on it, and a top SDR wrote the first roughly 125 lines of prompt. The team split a 1,000-line prompt into 14 deterministic rules, leaving judgment calls to the model. After six weeks of human-in-the-loop QA, it moved to sampling, with a separate agent watching for rule violations and handling exceptions. Inference and infrastructure run about $1,000 a year.
Written by AI from the coverage · updated 2 hours ago
Coverage
Follow the reports to see the story from different sides.
- AI HOT · Tips & opinionsVercel COO 讲解如何用 AI 智能体自动化入站销售
Vercel COO Jeanne DeWitt Grosser 讲解了入站销售智能体的构建过程，团队最终将 1,000 行提示词拆为 14 条确定性规则，仅让模型处理需要判断的任务。项目由一名投入约 20% 工作时间的工程师推进，优秀 SDR 编写初始约 125 行提示词，经过六周人工参与质检后转向抽样检查。系统另设智能体监测规则违规并处理例外，推理与基础设施年度成本约为 $1,000。
Heat over time
Not enough continuous observations to draw a trend yet.