Agentforce評測:Salesforce七個AI代理各有名字與職稱 | Agentforce Review: Salesforce Gives AI Agents Names, Jobs
By Kit 小克 | AI Tool Observer | 2026-09-14
🇹🇼 Agentforce評測:Salesforce七個AI代理各有名字與職稱
Salesforce 在 9 月 11 日一口氣推出七個有名有姓的 Agentforce AI 代理——Casey、Paige、Carter、Hunter、Marshall、Piper、Fin,每個都對應一個具體職位:客服、IT/HR、購物助理、陌生開發業務、供應鏈、進線業務、顧客體驗。這是目前企業級 AI 代理最積極的「擬人化」嘗試,也讓 Agentforce 這個關鍵字這週衝上企業軟體圈熱搜。
七個代理在做什麼
- Casey:跨語音、簡訊、WhatsApp、網頁聊天處理客服問題,內建常見問題、退貨、帳戶管理與真人轉接。
- Paige:IT/HR 員工代理,在 Slack、內部入口網站處理員工申請。
- Carter:購物代理,幫消費者比價、問答、直接在對話中結帳。
- Hunter:陌生開發業務代理,能跨週跟業務員一起經營一條 pipeline。
- Marshall:供應鏈代理,自動化後勤流程並留下稽核紀錄。
- Piper:進線業務代理,把網站與信箱的詢問轉成有效名單。
- Fin:顧客體驗代理,處理跨通路的複雜客服流程。
真正的技術亮點,不是名字
取名字、給職稱是行銷包裝,真正值得注意的是兩件事:一是 long-horizon runtime,讓 Hunter 這類代理可以用「週」為單位持續追蹤目標,而不是單次對話就結束;二是 Agent Script,一個把 AI 推理和確定性規則混合寫在一起的開源語言,讓企業可以限制代理「哪些事一定要照規則做、哪些事可以讓模型判斷」。Salesforce 同時發布了 Trusted Enterprise AI Harness 治理框架,涵蓋情境、代理權限、執行動作、治理、安全、模型六個面向,主要是為了因應企業同時使用多家供應商代理時的管控問題。
誠實的提醒
七個代理裡目前六個正式上線,Hunter(最有技術含量的長期任務代理)還在試營運,要到 11 月才會全面開放,代表宣傳裡最吸睛的功能其實還沒成熟。而且這些代理全部建立在 Salesforce 自家的 Customer 360 資料平台上,導入等於把業務流程再往 Salesforce 生態系綁得更深——如果你本來就是 Salesforce 重度用戶,這是低門檻導入 AI 代理的好機會;如果不是,這則新聞跟你關係不大,別被「AI 員工」的說法沖昏頭。
這波新聞的意義,不在於 Salesforce 發明了什麼新技術,而是印證企業 AI 代理正在從「一個萬用聊天機器人」走向「一群分工明確、有稽核紀錄的專職代理」,這個方向會是接下來一年企業軟體的主流敘事。
好不好用,試了才知道。
🇺🇸 Agentforce Review: Salesforce Gives AI Agents Names, Jobs
On September 11, Salesforce launched seven named Agentforce AI agents — Casey, Paige, Carter, Hunter, Marshall, Piper, and Fin — each built for one specific job: customer service, IT/HR, shopping, outbound sales, supply chain, inbound pipeline, and customer experience. It is the most aggressive attempt yet to give enterprise AI agents a human face, and it is driving this week search spike around Agentforce.
What Each Agent Actually Does
- Casey handles support across voice, SMS, WhatsApp and web chat, with built-in FAQs, returns, account management, and human escalation.
- Paige is an IT/HR agent resolving employee requests inside Slack and internal portals.
- Carter helps shoppers compare products, get answers, and checkout in-chat.
- Hunter works an outbound sales pipeline alongside human reps — over weeks, not one chat session.
- Marshall automates back-office supply chain work with a full audit trail.
- Piper converts inbound web and email inquiries into qualified pipeline.
- Fin resolves complex customer-experience workflows across channels.
The Real Story Isn't the Names
Naming and job titles are marketing polish. What is actually new is the long-horizon runtime that lets Hunter pursue a goal across weeks instead of resetting each session, plus Agent Script, an open-source language that mixes AI reasoning with deterministic rules — so companies can decide which steps must follow fixed logic and which can be left to the model judgment. Salesforce paired this with the Trusted Enterprise AI Harness, a six-pillar governance framework covering context, agency, action, governance, security, and models, aimed at companies already juggling agents from multiple vendors.
The Honest Take
Six of the seven agents are generally available; Hunter, arguably the most technically interesting one, stays in pilot until November — the flashiest capability is not actually shipped yet. All seven run on Salesforce own Customer 360 data platform, which means adopting them ties your workflows more tightly to Salesforce ecosystem. If you are already a heavy Salesforce shop, this is a low-friction way to deploy agents into specific roles. If you are not, this launch does not change much for you — do not let the 'AI employees' framing oversell it.
The bigger signal here is not a new technology — it is enterprise AI agents shifting from one general-purpose chatbot to a roster of narrow, auditable, job-specific agents. Expect that framing to dominate enterprise software marketing for the next year.
好不好用,試了才知道。
Sources / 資料來源
- Salesforce Expands Agentforce With a New Portfolio of AI Agents Built for High-Value Work
- Salesforce Debuts Job-Ready Agentforce Agents and Long-Horizon Runtime
- Salesforce agents gain a runtime that pursues goals over weeks, not chats
延伸閱讀 / Related Articles
- Real-SWE評測:企業真實程式碼,AI最高只答對38.8% | Real-SWE Review: AI Tops Out at 38.8% on Real Code
- Asimov晶片評測:8.75億美元豪賭不用HBM晶片 | Asimov Chip Review: $875M Bet on Memory Over HBM
- Gemini 3.8 Flash Cyber評測:Google補漏洞模型,政府才能用 | Gemini 3.8 Flash Cyber Review: Bug-Patching AI, Gov Only
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