⌘K
Change language Switch ThemeSign In
Narrow Mode
How Ramp Uses AI to Ship 500+ Features with 25 PMs ==================================================
How Ramp Uses AI to Ship 500+ Features with 25 PMs ==================================================  ### Peter Yang
@petergyang
Ramp 去年仅凭 25 名产品经理就交付了 500 多个功能。以下是帮助他们实现这一目标的 Claude 代码技能 👇
它有 3 个阶段：
阶段 1：界定问题
Claude 会向产品经理提出以下问题：
→ 核心任务是什么？
→ 为什么是现在？
→ 成功的标准是什么？
→ 如果我们构建它，能带来什么？
如果产品经理无法充分解释他们的答案，Claude 就会提出异议。
阶段 2：调研
Claude 会启动 6-10 个并行智能体，同时扫描竞品、Gong 通话记录、Zendesk 工单和代码库。
每个智能体都会编写一个结构化文件，然后由 Claude 进行整合。
阶段 3：完善规范
Claude 通过对话达成设计共识。最终的规范文档简短，只需 2 分钟即可读完，并包含：
→ 背景
→ 原则（不可协商的约束）
→ 需求
...等等
我从 @geoffintech (Ramp 首席产品官 @tryramp) 那里学到了关于如何让公司实现 AI 原生化的知识，比我采访过的任何人都多。
📌 订阅即可获取完整节目和产品规范制定技能，明天发布：youtube.com/@peteryangyt?s…Show More
Mar 14, 2026, 2:30 PM View on X
8 Replies
3 Retweets
66 Likes
5,235 Views  Peter Yang @petergyang
One Sentence Summary
Peter Yang details Ramp's three-phase AI-native product development workflow using Claude to automate problem framing, research, and spec shaping.
Summary
This tweet breaks down the 'AI-native' product management strategy used by Ramp to achieve high velocity. The workflow leverages Claude in three distinct stages: first, as a critical thinking partner to challenge problem framing; second, as a multi-agent research tool that synthesizes data from various sources like Gong and Zendesk; and third, as a collaborative tool to generate concise, high-quality product specs. The insights come from an interview with Ramp's CPO, Geoff Charles, highlighting a practical shift in how AI can scale product teams.
AI Score
85
Influence Score 36
Published At Today
Language
English
Tags
AI-native
Product Management
Claude
Ramp
AI Agents HomeArticlesPodcastsVideosTweets
How Ramp Uses AI to Ship 500+ Features with 25 PMs | Best... ===============