AI 模型和平台
Smartsheet 发布 vsActual 和企业AI的执行图
在 Google 上的首选来源中添加 Unite.AI
大赦国际可以帮助一个团队更快地完成任务，而不帮助一个组织实施其战略。 Smartsheet在EngAGE 2026客户大会上，正用VsActual这个新的独立，AI本地的规划和执行产品发布来瞄准这一缺口。
除了vsActual外，宣布公司还专门使用agents，经常性代理工作流程，新的集成，以及额外的安全控制。 这些更新旨在让大赦国际更清楚地了解项目、人员、预算和商业重点如何相互依存
vsual 将策略连接到已启动的工作
vsActual旨在跨已经使用的系统团队工作，而不是要求他们将每一项活动都转移到一个新的项目管理工具。 Smartsheet名称包括财务，HR，客户，以及服务平台，如Oracle,SAP,Workday,Salesforce,Jira，以及ServiceNow等，在它所设计的系统中，它可以全读。
起点是战略本身。 该产品从列报、预算文件或电子表格中获取计划，并将其转化为具有目标、时限和开支及人员配备能力等制约因素的活计划。 然后可以根据这些优先事项评估新的请求。
其情景规划能力是为了在领导人批准决定之前显示决定的预算和能力后果。 一旦选择了一种情景，它就会成为记录的计划。 Agents随后将正在进行的工作与该计划进行比较，查明新出现的拖延或资源冲突，并从基本数据中汇总执行摘要。
“每个公司都有策略。 胜者是那些真正能完成这项工作的人，他形容vsactual是缩小计划与实际结果之间距离的一种方式。
关键边界在于人：称 agents公司不能独立改变计划，建议包括领导人审查的理由。 公告指示客户注册进行更新，但没有规定 vsActual 的一般可用性或定价。
执行图供应业务背景
VsActual 背后是 Smartsheet 的执行图，它将各个任务通过项目和方案的关系映射到战略目标。 称公司借鉴了20年的工作流程智能，数百万项目，以及9个微字节的数据。
技术提议不仅仅是为聊天员收集文件。 一个有用的执行模式必须代表关系：哪个举措依赖于一个延迟的项目，哪个团队拥有稀缺的能力，谁拥有决定权，什么权限或批准会制约下一步。 该图包含这些业务关系和治理要求。
例如，一个滑落的里程碑可能孤立地可以管理，但当若干得到资助的倡议依赖同一个小组时，则变得至关重要。 将时间表、能力和战略优先事项连接起来，比单独的仪表板更快地暴露出这种冲突。 该建议是否有用，仍取决于相关记录的完整性和新鲜性。
Smartsheet现有的AI架构已经将智能层和知识图与通过Azure OpenAI和Amazon Bedrock交付的基础模型相结合。 其公开文献将这一点描述为在一般模型之上增加了具体工作的背景。 新的执行图将公司的重点扩大到规划和实施；该版没有确定新培训的基础模型，也没有公布比较性能基准测试。
新建 Agents 接受项目设置和重复检查
配套平台发布推出 Agentic Workflows，将自然语言指令转化为经常性工作，如编写每周项目摘要或检查风险和依赖性。 将结果交给相关人员，而不需要每次有人重复请求。
另有两个专业agents列于发布。 " 创建Agent " 从一个普通语言目标建立项目结构，包括工作表、仪表板、报告和自动化。 数据质量Agent按要求或按时间表按严重程度审查工作单和旗帜问题。
这些能力建立在Smart Assistance和Smartsheet的模型背景协议服务器上。 现有文档解释了嵌入式助理和兼容的外部AI工具如何在用户同意下查询现场项目数据并进行动作。 MCP提供了一个共同的连接接口；周围的应用程序仍然需要提供合适的上下文并强制访问。
Smartsheet也与宣布本土融合，与ServiceNow,Microsoft动态365,HubSpot，和Workday. 一个单独的集成Agent在早期进入，最初针对GitHub,Zendesk,AirTable，和星期一.com. 它旨在让客户请求外部数据连接对话。
治理成为执行层的一部分
随着agents获得行动能力，许可与其答案的质量同样重要。 Smartsheet 称人和agents使用相同的受管图，agents仅限于他们所代表的人允许的访问和更改。
新的控制包括安全事件流(Security Evention Streaming)，它将审计事件传送给Sprunk和Microsoft Sentinel等工具，几乎实时。 具有Smartsheet Security功能的数据分类让管理员定义敏感性标签并通过集中报告和审计线索跟踪其使用情况。
该公司还为AI管理提供宣布 ISO/IEC 42001认证和澳大利亚IRAP安全评估。 它们分别涉及组织人工智能管理和评估安全控制；它们不能保证人工智能产生的每一条建议的正确性。
Smartsheet更广泛的信托中心描述了物品级别权限，身份控制和审计能力，并声明客户数据不用于培训第三方基础模型。 这些承诺提供了相关背景，而新的发布则增加了专门针对agent活动的控制。
对企业来说，核心问题是AI能否保持对分散系统执行的有用和有规范的理解。 Smartsheet打赌它的工作管理历史给了它这样做的背景。 Vsual的意义在于如何有效地将背景转化为关于优先事项、资源和纠正行动的及时决定。
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
Smartsheet Unveils vsActual and an Execution Graph for Enterprise AI
Add Unite.AI to your preferred sources on Google
AI can help a team finish tasks faster without helping an organization deliver its strategy. Smartsheet is targeting that gap with vsActual, a new standalone, AI-native planning and execution product unveiled at its ENGAGE 2026 customer conference.
Alongside vsActual, the company announced specialized agents, recurring agentic workflows, new integrations, and additional security controls. Together, the updates aim to give AI a clearer picture of how projects, people, budgets, and business priorities depend on one another—and flag when actual work begins moving away from the approved plan.
vsActual Connects Strategy to the Work Already Underway
vsActual is designed to work across the systems teams already use, rather than require them to move every activity into a new project management tool. Smartsheet names financial, HR, customer, and service platforms such as Oracle, SAP, Workday, Salesforce, Jira, and ServiceNow among the systems it is designed to read across.
The starting point is the strategy itself. The product takes plans from presentations, budget documents, or spreadsheets and turns them into a living plan with goals, timelines, and constraints such as spending and staffing capacity. New requests can then be assessed against those priorities.
Its scenario-planning capabilities are intended to show the budget and capacity consequences of a decision before a leader approves it. Once a scenario is selected, it becomes the plan of record. Agents subsequently compare ongoing work with that plan, identify emerging delays or resource conflicts, and assemble executive summaries from the underlying data.
“Every company has a strategy. The ones that win are the ones whose work actually delivers it,” said Rajeev Singh, Smartsheet’s CEO. He described vsActual as a way to close the distance between plans and actual results.
A key boundary remains with people: the company says agents cannot change the plan independently, and recommendations include reasoning for leaders to review. The announcement directs customers to register for updates but does not specify general availability or pricing for vsActual.
The Execution Graph Supplies Business Context
Behind vsActual is Smartsheet’s Execution Graph, which maps relationships from individual tasks through projects and programs to strategic goals. The company says it draws on 20 years of workflow intelligence, millions of projects, and nine petabytes of data.
The technical proposition goes beyond collecting documents for a chatbot. A useful execution model must represent relationships: which initiative depends on a delayed project, which teams share scarce capacity, who owns a decision, and what permissions or approvals constrain the next step. Smartsheet says the graph incorporates those operational relationships and governance requirements.
For example, a slipping milestone may be manageable in isolation but become critical when several funded initiatives depend on the same team. Connecting schedules, capacity, and strategic priorities could expose that conflict sooner than separate dashboards. Whether the recommendation is useful still depends on the completeness and freshness of the connected records.
Smartsheet’s existing AI architecture already combines its intelligence layer and knowledge graph with foundation models delivered through Azure OpenAI and Amazon Bedrock. Its public documentation describes this as adding work-specific context on top of general models. The new Execution Graph extends the company’s emphasis on context into planning and execution; the release does not identify a newly trained foundation model or publish comparative performance benchmarks.
New Agents Take on Project Setup and Recurring Checks
The accompanying platform release introduces Agentic Workflows that turn natural-language instructions into recurring work, such as preparing a weekly project summary or checking risks and dependencies. The results are delivered to the relevant people without requiring someone to repeat the request each time.
Two specialized agents are also listed as available at launch. The Creation Agent builds project structures, including sheets, dashboards, reports, and automations, from a plain-language goal. The Data Quality Agent examines sheets and flags problems by severity, either on demand or on a schedule.
These capabilities build on Smart Assist and Smartsheet’s Model Context Protocol server. Existing documentation explains how the embedded assistant and compatible external AI tools can query live project data and perform actions with the user’s consent. MCP provides a common connection interface; the surrounding application still has to supply the appropriate context and enforce access.
Smartsheet also announced native integrations with ServiceNow, Microsoft Dynamics 365, HubSpot, and Workday. A separate Integration Agent is in early access, initially targeting GitHub, Zendesk, Airtable, and Monday.com. It is designed to let customers request an external data connection conversationally.
Governance Becomes Part of the Execution Layer
As agents gain the ability to act, permissions matter as much as the quality of their answers. Smartsheet says people and agents use the same governed graph, with agents restricted to the access and changes permitted for the person they represent.
The new controls include Security Event Streaming, which sends audit events to tools such as Splunk and Microsoft Sentinel in near real time. Data Classification, available with Smartsheet Safeguard, lets administrators define sensitivity labels and track their use through centralized reporting and an audit trail.
The company also announced ISO/IEC 42001 certification for AI management and an Australian IRAP security assessment. These address organizational AI management and assessed security controls respectively; they do not guarantee the correctness of every AI-generated recommendation.
Smartsheet’s broader Trust Center describes item-level permissions, identity controls, and audit capabilities, and states that customer data is not used to train third-party foundation models. Those commitments provide relevant background, while the new releases add controls specifically aimed at scaling agent activity.
For enterprises, the central question is whether AI can maintain a useful, governed understanding of execution across fragmented systems. Smartsheet is betting that its work-management history gives it the context to do that. vsActual’s significance will rest on how effectively it turns that context into timely decisions about priorities, resources, and corrective action.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.