Sierra Ghostwriter 完整解析：Bret Taylor 打造用對話就能建 AI Agent 的工具，按鈕介面時代結束了？ | Sierra Ghostwriter Explained: Bret Taylor Build AI Agents by Talking — Is the Button Era Over?
By Kit 小克 | AI Tool Observer | 2026-04-13
🇹🇼 Sierra Ghostwriter 完整解析：Bret Taylor 打造用對話就能建 AI Agent 的工具，按鈕介面時代結束了？
Sierra Ghostwriter 是前 Salesforce 共同執行長 Bret Taylor 創辦的 AI 新創 Sierra 在 2026 年 4 月 9 日推出的新產品。它的核心概念很簡單：用自然語言對話，就能建立一個可以接電話、聊天、發 email、支援 30 種語言的客服 AI Agent。不用寫程式，不用拖拉介面，只要用講的。
Sierra Ghostwriter 是什麼？怎麼運作？
Sierra Ghostwriter 是一個「用來建造 Agent 的 Agent」。你透過自然語言描述你想要的客服行為、邊界條件和例外處理，Ghostwriter 會自動把這些對話轉化成一個可上線的 AI Agent，支援語音、文字聊天和 email 三種管道。
整個流程的關鍵在於：Ghostwriter 會從對話中辨識關鍵行為和邊界案例，自動加上防護欄（guardrails），然後產出可以直接部署的 Agent。傳統做法需要工程師寫規則、QA 測試、PM 定義流程，現在一個人跟 Ghostwriter 聊天就搞定了。
為什麼 Bret Taylor 說「按鈕介面時代結束了」？
Taylor 在 HumanX 大會上宣稱，未來使用者不需要再點按鈕來完成任務，自然語言將取代傳統軟體介面。這不只是 UI 層面的改變，而是整個軟體互動模式的重新定義——從「人適應軟體」變成「軟體理解人」。
這個願景並非空談：Sierra 成立不到 21 個月就達到 1 億美元 ARR，最近一輪融資估值 100 億美元，由 Greenoaks Capital 領投 3.5 億美元。客戶包括多家財星 500 大企業。
Ghostwriter 跟其他 Agent 平台差在哪？
市場上的 Agent 建造工具不少，但 Ghostwriter 的差異化在三個地方：
- 完全自然語言：不需要拖拉流程圖或寫 YAML，用對話定義一切
- 內建防護欄：自動從對話中抓出邊界條件，不用另外設定安全規則
- 全通路部署：一次建好，同時支援語音通話、網頁聊天、email，還有 30+ 語言
對比 Anthropic 的 Claude Managed Agents 走開發者路線，Sierra Ghostwriter 更像是給非技術人員用的——客服主管、營運經理這些角色可以自己建 Agent。
Kit 小克觀點
「用講的就能建 Agent」聽起來很美好，但魔鬼在細節。實際上線後遇到的邊界案例，光靠對話能涵蓋多少？當 Agent 出錯時，非技術人員要怎麼 debug？這些問題 Sierra 還沒完全回答。不過以他們的成長速度和客戶規模來看，這個方向顯然有市場需求。
好不好用，試了才知道。
🇺🇸 Sierra Ghostwriter Explained: Bret Taylor Build AI Agents by Talking — Is the Button Era Over?
Sierra Ghostwriter is a new product launched on April 9, 2026 by Sierra, the AI startup founded by former Salesforce co-CEO Bret Taylor. The concept is straightforward: build a production-ready AI agent that can handle phone calls, chat, email, and 30+ languages — entirely through natural language conversation. No code, no drag-and-drop interfaces, just talking.
What Is Sierra Ghostwriter and How Does It Work?
Sierra Ghostwriter is an "agent that builds agents." You describe the customer experience behaviors, edge cases, and guardrails you want through natural language, and Ghostwriter automatically converts that conversation into a deployable AI agent across voice, chat, and email channels.
The key innovation: Ghostwriter identifies critical behaviors and edge cases from your conversation, automatically applies guardrails, and produces a production-ready agent. What traditionally required engineers writing rules, QA testing, and PM-defined workflows can now be accomplished by one person talking to Ghostwriter.
Why Does Bret Taylor Say "The Button Era Is Over"?
At the HumanX conference, Taylor declared that users will no longer need to click buttons to complete tasks — natural language will replace traditional software interfaces. This is not just a UI change but a fundamental redefinition of how humans interact with software: from "humans adapting to software" to "software understanding humans."
This vision is backed by results: Sierra reached $100 million ARR in under 21 months, with a $10 billion valuation after raising $350 million led by Greenoaks Capital. Their clients include multiple Fortune 500 companies.
How Does Ghostwriter Compare to Other Agent Platforms?
The agent-building market is crowded, but Ghostwriter differentiates in three ways:
- Fully natural language: No flowcharts, no YAML — define everything through conversation
- Built-in guardrails: Automatically extracts boundary conditions from conversation, no separate safety configuration needed
- Omnichannel deployment: Build once, deploy across voice calls, web chat, email, and 30+ languages
While Anthropic's Claude Managed Agents targets developers, Sierra Ghostwriter is designed for non-technical users — customer service managers and operations leads who can build agents themselves.
Kit's Take
"Build agents by talking" sounds compelling, but the devil is in the details. How many real-world edge cases can conversation-based setup actually cover? When an agent makes mistakes, how do non-technical users debug it? Sierra has not fully answered these questions. But given their growth trajectory and enterprise client base, the market demand for this approach is clearly real.
You never know until you try.
Sources / 資料來源
- TechCrunch: Sierra Bret Taylor says the era of clicking buttons is over
- The AI Insider: Sierra Advances AI Agent Model with Ghostwriter
- Sierra Blog: Agents as a Service
常見問題 FAQ
Sierra Ghostwriter 是什麼？
Ghostwriter 是 Sierra 推出的 AI Agent 建造工具，使用者透過自然語言對話就能建立可上線的客服 AI Agent，支援語音、聊天、email 和 30 種以上語言。
Sierra Ghostwriter 需要寫程式嗎？
不需要。Ghostwriter 完全透過自然語言對話定義 Agent 行為，不用寫程式碼、拖拉流程圖或設定 YAML 檔案。
Sierra 公司估值多少？
Sierra 在 2025 年 9 月由 Greenoaks Capital 領投 3.5 億美元，估值達 100 億美元。成立不到 21 個月就達到 1 億美元年經常性收入。
Ghostwriter 跟 Claude Managed Agents 差在哪？
Claude Managed Agents 走開發者路線需要技術背景，Ghostwriter 設計給非技術人員使用，如客服主管和營運經理，用對話就能建 Agent。
延伸閱讀 / Related Articles
- AI 模型蒸餾攻擊全解析：OpenAI、Anthropic、Google 聯手反擊中國 AI 偷學，Frontier Model Forum 首次啟動威脅情報共享 | AI Model Distillation Attacks Explained: OpenAI, Anthropic, Google Unite to Stop Chinese AI Theft via Frontier Model Forum
- ICML 2026 用隱形浮水印抓 AI 代寫審稿：497 篇論文被退，學術誠信怎麼辦？ | ICML 2026 Catches AI Peer Review Cheating With Hidden Watermarks: 497 Papers Rejected
- Microsoft MAI 自研模型實測：Suleyman 領軍打造語音、圖像三模型，正式向 OpenAI 宣戰 | Microsoft MAI Models Explained: In-House AI Takes on OpenAI With Speech, Voice and Image
AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends
留言
張貼留言