AI 模型和平台
Sierra 在Agent Studio工作流程中心设置了幽灵作家
在 Google 上的首选来源中添加 Unite.AI
2026年11月2日，Sierra 宣布在Ghostwriter周围重新设计了Agent工作室，其agent建筑agent，将调查，旅行编辑和测试带入一次对话。 证据、拟议更改和测试结果仍放在一起，以便各小组在事前审查。
此前，用户必须在Agent Studio的不同部分之间移动，以编辑行程，运行模拟和审查性能。 新工作流程允许用户询问为什么解析率下降，检查对话，更新行程，测试固定点而不离开对话。
Sierra 推出幽灵作家于2026年3月25日作为agent用于建造其他agents. 其原始投入包括标准作业程序、支助电话记录、白板草图照片、流程文件、录音和普通英语指令。 Ghostwriter使用这些材料来识别agents在语音，聊天和电子邮件中运行的关键行为和边缘案例。 2026年9月28日，Sierra将其扩展为Slack和Teams的常备，主动的队友，可以审查客户互动，提出改进建议，并帮助将其转化为实验。
持久性背景和单独会议
随着用户在旅程、模拟和仪表板之间移动，现在仍然可以使用写鬼魂的人，它可以看到目前正在查看的项目。 一个用户可以询问一个明显的问题，或者让Ghostwriter在不重复周围上下文的情况下进行处理。 重新设计后的经验结合了先前分散在Explorer,Sierra助手和其他专业的agents的功能，消除了了解哪个工具处理特定任务的需要。
每个会话都作为聊天线索进行操作，既包含谈话内容，也包含"幽灵作家"所做的改变。 这些变化在会话中保留，直到有人审查和合并。 单独的会话可以让一个团队调查客户的增加，请求一个线程中的支持助理，同时在另一个线程中构建一个新的旅程，而这两个活动不会互相干扰。
可以共享会话，使同事能够一起解决一个问题，或者在另一个用户停止时要继续工作。 在合并之前，用户也可以要求Ghostwriter审查队友的变动。
多个 Agents 扩展项目
Agent Studio现在可以将关联的agents与他们共享的数据和频道放在一个单一的项目中，让团队一起测试并释放这些agents.
Sierra的放款人例子指定一个agent负责向申请人收集文件，另一个负责管理申请和后续，直到贷款结束。 当放款人新增对文件的要求时，Ghostwriter既更新了藏书agent要求的内容，又更新了应用程序agent在一个会话内如何跟进。
测试、核准和推出控制
逐行比较，可以确切地显示"鬼作者"改变了什么。 预览聊天和模拟让团队在合并前测试行为，而发布治理新增则需要批准和有控制的推出。
支持的发行管理控制包括Agent检查，一个自动插件，根据行程和它们所产生的对话来显示警告。 检查按严重程度优先进行，大多数包括从Ghostwriter中提出可以应用的修补方法。 各组织也可以要求模拟在转向生产之前通过。
对于人文审查，各组织在发布前可以要求同行审议，并利用审查员的作用来明确需要签到哪些人，包括客户经验牵头人、合规利益攸关方、工程管理人员或其他领域专家。 审查者可以逐行检查diffs，留下上下文注释和请求更改。 构建者可以与Ghostwriter合作处理这些反馈，并在一个人批准合并之前请求再次审查。
经批准后，拆分流量的发布可能会在扩大部分客户流量之前暴露出变化。 每次发行都是不可改变的快照，如果出现意想不到的问题，支持回滚。
由此产生的工作流程使调查、编辑、测试和审查在会话内连通，而合并和客户接触仍受小组的批准和推出控制。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
AI Models & Platforms
Sierra Puts Ghostwriter at the Center of Agent Studio Workflows
Add Unite.AI to your preferred sources on Google
Sierra announced on November 2, 2026 that it had redesigned Agent Studio around Ghostwriter, its agent-building agent, bringing investigation, journey editing and testing into one conversation. Evidence, proposed changes and test results remain together so teams can review them before anything goes live.
Previously, users had to move between different parts of Agent Studio to edit journeys, run simulations and review performance. The new workflow allows a user to ask why resolution rates dropped, examine conversations, update a journey and test the fix without leaving that conversation.
Sierra introduced Ghostwriter on March 25, 2026 as an agent for building other agents. Its original inputs included standard operating procedures, support-call transcripts, photographs of whiteboard sketches, process documentation, audio recordings and plain-English instructions. Ghostwriter uses those materials to identify key behaviors and edge cases for agents operating across voice, chat and email. On September 28, 2026, Sierra expanded it into an always-on, proactive teammate in Slack and Teams that can review customer interactions, suggest improvements and help turn them into experiments.
Persistent Context and Separate Sessions
Ghostwriter now remains available as users move among journeys, simulations and dashboards, and it can see the item currently being viewed. A user can ask about an apparent problem or have Ghostwriter address it without restating the surrounding context. The redesigned experience combines functions previously spread across Explorer, Sierra Assistant and other specialized agents, removing the need to know which tool handles a particular task.
Each session operates as a chat thread containing both the conversation and the changes Ghostwriter makes. Those changes stay in the session until someone reviews and merges them. Separate sessions allow a team to investigate an increase in customers requesting a support associate in one thread while building a new journey in another, without the two activities interfering with each other.
Sessions can be shared so colleagues can work through a problem together, or forked to continue where another user stopped. Before a merge, a user can also ask Ghostwriter to review a teammate’s changes.
Projects Spanning Multiple Agents
Agent Studio can now group related agents in a single project alongside the data and channels they share, allowing teams to test and release those agents together.
Sierra’s lender example assigns one agent to collect documents from applicants and another to manage the application and follow up until the loan closes. When the lender adds a document requirement, Ghostwriter updates both what the collection agent requests and how the application agent follows up within one session.
Testing, Approval and Rollout Controls
Line-by-line comparisons show exactly what Ghostwriter changed. Preview Chat and Simulations let teams test behavior before a merge, while release governance adds required approvals and controlled rollouts.
The supporting release-governance controls include Agent Checks, an automated linter that surfaces warnings based on journeys and the conversations they generate. Checks are prioritized by severity, and most include a suggested fix from Ghostwriter that can be applied in place. Organizations can also require Simulations to pass before a change moves toward production.
For human review, organizations can require peer review before release and use a Reviewer role to specify whose sign-off is needed, including a customer-experience lead, compliance stakeholder, engineering manager or another domain expert. Reviewers can inspect line-by-line diffs, leave contextual comments and request changes. Builders can work with Ghostwriter to address that feedback and request another review before a person approves the merge.
After approval, split-traffic releases can expose a change to part of customer traffic before expanding it more broadly. Each release is an immutable snapshot, supporting rollback if an unexpected problem appears.
The resulting workflow keeps investigation, editing, testing and review connected within a session, while merging and customer exposure remain subject to the team’s approvals and rollout controls.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.