monday.com 如何将平台重塑为人类与 Agent 协作的 Agent-first 产品
英文标题：How monday.com transformed its platform into an agent-first product where humans and agents collaborate
monday.com 为超过 25 万家企业提供工作管理服务，如今已围绕人类与 AI Agent 协作模式，从底层重构整个平台，并以 Claude 为核心模型能力。该公司认为，仅在既有流程中叠加零散 AI 功能会形成“AI dust（AI 尘埃）”，难以带来持续使用，因此将 Agent 直接嵌入工作流、权限体系、看板和治理机制中。自 2026 年 5 月新平台上线以来，客户已在 monday 平台上与 Agent 互动超过 500 万次；开发者和企业可通过自定义 Agent、BYOA、预构建 Agent 及 Claude Coding 集成，将 AI 能力接入真实业务流程。monday 的经验表明，Agent 落地不仅取决于模型能力，也依赖数据基础设施、权限治理、透明度、可靠性以及用户信任。
10.2 KB
403 B
10.7 KB
10.5 KB
17.7 KB
14.0 KB
中文
超过 25 万家企业——从中小型企业到《财富》500 强组织——使用 monday.com 管理工作。十多年前，该公司推出时的核心产品是一套可视化界面，帮助团队自动化工作流并管理项目。如今，它已围绕人类与 Agent 协作模式，从底层重新架构产品，将 AI 融入工作的每一个层面。以 Claude 为核心，monday 的新平台负责处理技术复杂性，让客户能够在自己熟悉的工作流中使用前沿 AI。
“向 Agent-first 产品转型，是我们作为一家公司做出的最重要决策之一，”monday.com 首席产品与技术官 Daniel Lereya 表示，“这意味着要从根本上重新构想平台应当做什么，而不只是为现有工作流添加 AI。我们的愿景是让 monday 成为人们与 AI Agent 无缝协作的场所，而 Anthropic 和 Claude 一直是值得信赖的合作伙伴，帮助我们将这一愿景变为现实。”
撞上“AI 尘埃”天花板
monday 的重构经历了三个阶段。在第一阶段，随着前沿 LLM 技术逐渐成熟、客户热情不断上升，monday 团队开始将 AI 能力嵌入原有平台。这项工作在 2025 年 5 月达到高潮：公司举办了内部“AI 月”，连续四周专门用于在全公司范围内发布 AI 功能或产品。
采用情况很强劲，也带来了热情，但团队很快碰到了天花板。monday.com AI Works Platform 产品副总裁 Orly Stern Izhaki 表示：“我们当时是在构建‘AI 尘埃’——向现有工作流零散撒入自动化功能，却没有将其真正嵌入其中，也没有改变产品的根本价值主张。我们的功能能帮助用户总结文本和分类信息，但并未形成持续的使用模式。”
公司需要将重点从为产品功能添加 AI，转向将 AI 原生构建进平台之中。Izhaki 说：“采用 AI 功能，并不等于成为一家 AI 公司。一旦我们理解了这一点，一切就改变了。”
于是，Izhaki 的团队着手彻底重新构想 monday：将其从一款工作管理工具，转变为人类与 Agent 共同完成工作的场所。虽然改造产品的使命来自管理层，但每个团队和每位员工都必须将这一北极星目标转化为具体的产品选择，并构建自己的 Agent。
经过数月的密集工作，公司宣布了其历史上最重大的变化：围绕人类与 Agent 协作，重新构建整个产品体验，并利用已经内置的上下文、工作流、看板、权限和治理机制。自 2026 年 5 月上线以来，monday 的客户已在其平台上与 Agent 进行了超过 500 万次互动。
将 Agent 视为队友
除访问权限和限制外，每个 monday Agent 都拥有名称和头像，同事可以通过 monday 平台中的触发器和提及功能向 Agent 分配工作。
这种设计是有意为之，旨在解决 monday 在客户群中观察到的一种模式：许多企业希望让 AI 投入工作，但往往停留在一个与实际工作地点并行运行的 AI 聊天界面。将 Agent 直接嵌入工作流，并让人们像与同事互动一样与它们交互，使 Agentic AI 从一个抽象概念变成了具体、可执行的能力。
monday 为 Agent 规划的工作范围包括：IT 工单分流和知识库维护、候选人搜寻和面试排期、面向销售及营销的竞争情报简报，以及诸如会议准备、将决策转化为可跟踪任务等幕僚长工作。
| 用例 | 工作内容 |
|---|---|
| IT —— 从工单到解决方案 |
- 受理与分流 Agent —— 对工单分类，自动解决常见请求，并携带完整上下文升级处理
- 知识 Agent —— 识别知识缺口，起草新的知识库文章
- 事件 Agent —— 检测事件，建立应急协作室，触发事后复盘
HR —— 从职位发布到录用 |
- 简历筛选器 —— 为申请打分，筛选优秀候选人，发送拒信
- 面试排期 Agent —— 处理全部排期与确认事宜
- 招聘协调 Agent —— 在整个流程中让所有利益相关者及时了解进展，确保始终有人类参与其中
- 反馈管理 Agent —— 自动收集结构化面试官反馈
营销 —— 竞争情报 |
- 竞争情报 Agent —— 监测竞争对手，发现并分类信号，发送警报和每周简报
- 竞争卡片 Agent —— 根据已批准信号更新竞争卡片，立即通知销售团队
高管办公室 —— 幕僚长即服务 |
- 运营 Agent —— 安排会议、准备简报、将决策转化为可跟踪任务、监控优先事项
- 组织健康 Agent —— 扫描营收风险、成本泄漏和失败中的项目
- 战略顾问 Agent —— 识别增长机会，生成行动计划
四种常见工作流中的 Agent 团队及其工作内容。
在 monday 中运行 Claude 的四种方式
客户可通过四项能力，在 monday 平台内使用 Claude：
借助 monday Agents，团队可以使用提示词构建自定义 Agent，并选择 Claude 作为其模型。平台会为 Agent 提供名称、头像以及看板中的位置，任何人都可以向其分配工作。
Bring Your Own Agent（BYOA，自带 Agent） 让 Claude Managed Agents 能够加入平台。一个人构建的 Agent 一旦进入 monday 平台，就能成为整个团队可提及和分配工作的队友。
monday Agents Store 中提供的 预构建 Agent，可将 Claude 插件转化为专业队友：法务团队可以在自己的工作流中将法务插件作为 Agent 运行，财务团队也可以对其插件采取同样方式。
Claude Coding 集成使团队能够在 monday 仪表板中连接 Claude，随后规划任务并向 Agent 分配任务。Claude Managed Agents 会在客户自己的环境中执行，结果和更新会在任务移交给下一位 Agent 或人类审核者之前回写到工单中。工作流可以从业务需求延伸到可运行代码，再回到业务用户手中。
无需离开看板：从简报到落地页
以下是一个端到端示例：营销团队在单个看板条目中运行一条营销活动生产线。营销人员和内容负责人在该条目中完善简报，就目标、受众、核心信息和渠道达成一致。使用 monday Agents 构建的策略 Agent，会将这些原始输入转化为结构化简报，涵盖营销活动目标、信息支柱、渠道拆分和成功指标。
随后，落地页构建器接手工作。它运行在公司自有环境中的 Claude Managed Agents 上，提取已获批准的简报，并生成现有落地页的新版本，使文案、结构和信息传达适应这次营销活动。输出会自动回写至 monday 条目。在页面进入审批前，品牌审核员——一个 Claude Managed Agent——会依据品牌指南和法律标准检查页面，并标记任何需要人工关注的问题。此后，营销经理只需做出一个决策：发布还是继续优化。
[视频/嵌入内容：展示营销团队在看板内从简报生成并审核落地页的端到端工作流]
处于 AI 前沿的家族企业
Cooke 是一家家族海产品企业，1985 年创立于新不伦瑞克省布莱克斯港。从最初一个拥有 5,000 条三文鱼的养殖场，发展为业务覆盖 16 个国家、全球最大的家族所有海产品公司。如今，Cooke 同时使用 Claude 和 monday 进行项目交付、资源管理和合同管理。产品经理使用 Claude 将已批准的项目章程和需求转化为初步项目计划，生成状态报告，并发现风险与问题；这些信息会直接汇入 monday 的 RAID 日志，覆盖约 200 个正在进行和拟议中的项目。Claude 还自动处理报告与数据准备工作，以确保 130 份合同的全生命周期工作流保持准确——这些维护工作过去既繁琐又依赖人工。
Cooke 战略总监 Patti Stevens 表示：“monday 和 Claude 结合后，能帮助我们了解团队产能，并做出更明智的资源分配决策。过去，Monday 是一个我们必须不断更新的平台；现在，我们直接在其中开展运营。”
monday.com 学到了什么
对于计划进行类似重构的公司，monday 团队分享了在将平台转型为 AI-first 产品过程中获得的五点经验：
- 改变心智模型比改变技术更难。 人们天然倾向于保护质量，并持续改善已经有效的事物。要让团队从“我们如何负责任地改进当前产品？”转向“我们如何负责任地为不同的未来重建它？”，所花时间比技术工作更长。
- 当一切都同时变化时，小团队行动更快。 方向、UX、技术、定价、信任模型以及公司对“好”的定义，都在同时变化。多层利益相关者会丢失大量细节，而职责清晰、拥有快速决策权的小团队则能始终贴近这些变化。
- 采用率既取决于信任，也取决于能力。 产品市场契合度取决于用户是否有信心、是否准备好让 Agent 参与实际工作。治理、权限、透明度和可靠性，决定了 Agent 是停留在试点项目中，还是能进入生产环境。
- 能力需要匹配的基础设施。 当 Agent 以实时项目数据、团队历史和结构化工作流为基础时，其表现会提升到另一个层次；而在企业规模下，后端必须承受得住。除 Agent 层外，monday 还投资建设 monday DB，使数据基础设施能够支持 Agent 跨组织运行所需的规模、速度和复杂性。
- 建立在已被验证有效的基础上。 monday 始终将自己描述为人们协作推动业务成果的场所，而这次 Agent-first 重构将这一承诺延伸给一种新型团队成员。人们依然来到 monday 实现目标，不同之处在于，如今与他们并肩工作的部分团队成员已成为 Agent。
English / 原文
More than 250,000 companies, from small and midsize businesses to Fortune 500 organizations, use monday.com to manage their work. When the company launched more than a decade ago, its core product was a visual interface that helped teams automate workflows and manage projects. Today, it has rearchitected its product from the ground up around a human-agent collaboration model where AI is woven into work at every level. With Claude at the core, monday’s new platform handles the technical complexity so customers can work at the AI frontier inside workflows they already know.
"The shift to an agent-first product was one of the most significant decisions we've made as a company," said Daniel Lereya, chief product and technology officer at monday.com. "It meant fundamentally reimagining what the platform should do, not just adding AI to existing workflows. Our vision is for monday to be the place where people and AI agents work together seamlessly and Anthropic and Claude have been trusted partners in helping us bring that vision to life."
Hitting the “AI dust” ceiling
monday’s rebuild unfolded in three phases. In the first phase, as frontier LLM technology matured and customer excitement grew, monday teams worked on embedding AI capabilities into its original platform. The effort culminated in May 2025 with an internal “AI month,” with four weeks dedicated to shipping AI features or products across the company.
Adoption was strong and generated excitement, but soon the team hit a ceiling. “We were building ‘AI dust’, sprinkling automations onto existing workflows without embedding them within or changing the product’s fundamental value proposition,” says Orly Stern Izhaki, VP of Product, AI Works Platform at monday.com. “Our features helped users summarize text and categorize information, but they weren’t creating sustained usage patterns.”
The company needed to shift focus from adding AI to product features to building it natively into the platform. "Adopting AI features is not the same as becoming an AI company," Izhaki says. "Once we understood that, everything changed."
That’s how Izhaki’s team set out to reimagine monday completely: from a work management tool to a place where people and agents get work done together. While the mandate to transform the product came from the top, it was up to each team and each employee to translate that north star into concrete product choices and build their own agents.
After months of intense work, the company announced the most significant change in its history, rebuilding its entire product experience around humans and agents working together, using already built-in context, workflows, boards, permissions, and governance. Since launching in May 2026, monday’s customers have had more than 5 million interactions with agents on its platform.
Agents as teammates
Along with access permissions and restrictions, each monday agent is given a name and an avatar, and colleagues can assign agents work through triggers and mentions in the monday platform.
This design was intentional, addressing a pattern monday noticed across its customer base: many enterprises want to put AI to work, but often stall at an AI chat that runs parallel to where they actually do the work. Embedding agents directly into workflows and enabling people to interact with them like they would with their colleagues turned agentic AI from an abstract concept to a concrete, actionable one.
The jobs monday has mapped for agents range from IT ticket triage and knowledge-base upkeep to candidate sourcing and interview scheduling, competitive-intelligence briefings for sales and marketing, and chief-of-staff work like meeting prep and converting decisions into tracked tasks.
| Use case | Jobs |
|---|---|
| IT — From ticket to resolution |
- Intake & Triage Agent — classify tickets, auto-resolve common requests, escalate with full context
- Knowledge Agent — detect knowledge gaps, draft new KB articles
- Incident Agent — detect incidents, open war rooms, trigger post-mortems
HR — From job post to hire |
- Resume Screener — score applications, surface top candidates, send rejections
- Interview Scheduler — handle all scheduling and confirmations
- Hiring Coordinator — keep all stakeholders updated throughout the process, so there is always a human in the loop
- Feedback Manager — collect structured interviewer feedback automatically
Marketing — Competitive intelligence |
- Competitive Intelligence Agent — monitor competitors, detect and categorize signals, send alerts and weekly briefings
- Battlecard Agent — update battlecards on approved signals, notify sales immediately
Executive Office — Chief of Staff as a Service |
- Operator Agent — book meetings, prep briefings, convert decisions into tracked tasks, monitor priorities
- Org Health Agent — scan for revenue risks, cost leaks, and failing initiatives
- Strategy Consultant Agent — identify growth opportunities, generate action plans
Agent teams and their jobs for four common workflows.
Four ways to run Claude in monday
Customers use Claude inside the monday platform through four capabilities:
With monday Agents , teams can build custom agents using prompts, and choose Claude as its model. The platform gives the agent a name, a face, and a place on the board where anyone can assign it work.
Bring Your Own Agent (BYOA) makes it possible for Claude Managed Agents to join the platform. Once on the monday platform, an agent one person has built can become a teammate the whole team can mention and assign work to.
Pre-built Agents , available in the monday Agents Store, turn Claude plugins into specialized teammates: a legal team can run a legal plugin as an agent inside its own workflows, and finance teams can do the same with theirs.
The Claude Coding integration enables teams to connect Claude in the monday dashboard, then plan and assign agents tasks. Claude Managed Agents executes in the customer's own environment, and results and updates land back on the ticket before the task hands off to the next agent or to a human for review. The work runs from business need to working code and back to the business user.
From brief to landing page without leaving the board
One end-to-end example: a marketing team runs a campaign production line inside a single board item. The marketer and content lead shape the brief on the item, aligning on goal, audience, key message, and channels. A Strategist Agent built with monday Agents turns that raw input into a structured brief covering the campaign objective, messaging pillars, channel breakdown, and success metrics.
From there, a Landing Page Builder takes over. Running on Claude Managed Agents in the company's own environment, it pulls the approved brief and generates a new variant of an existing landing page, with copy, structure, and messaging adapted to the campaign. The output lands back on the monday item automatically. Before the page reaches approval, a Brand Reviewer, a Claude Managed Agent, checks it against brand guidelines and legal standards and flags anything that needs human attention. The marketing manager then makes one decision: publish or refine.
VIDEO
A family business at the AI frontier
Cooke, a family seafood business founded in 1985 in Blacks Harbour, New Brunswick, has grown from a single farm site with 5,000 salmon into the world's largest family-owned seafood company, operating in 16 countries. Today, Cooke runs project delivery, resource management, and contract management on Claude and monday together. Product managers use Claude to turn approved charters and requirements into initial project plans, generate status reports, and surface risks and issues that feed straight into their monday RAID logs, across roughly 200 active and proposed projects. Claude automates the reporting and data prep that keep lifecycle workflows accurate across 130 contracts—upkeep work that used to be tedious and manual.
“Together, monday and Claude help us read team capacity and make smarter allocation calls,” says Patti Stevens, director of strategy at Cooke. “Monday used to be a platform we had to update. Now we operate from it.”
What monday.com learned
For companies planning a similar rebuild, monday’s team shares five lessons they learned as they transformed their platform into an AI-first product:
- The mental model is harder to change than the technology. People naturally want to protect quality and keep improving what already works. Moving teams from "how do we responsibly improve the current product?" to "how do we responsibly rebuild it for a different future?" took longer than the technical work.
- Small teams move faster when everything is changing at the same time. Direction, UX, technology, pricing, the trust model, and the company's own definition of good were all in motion at the same time. Layers of stakeholders would lose that much detail, but small teams with clear ownership and fast decision rights stayed close to it.
- Adoption depends on trust as much as it does on capability. Product-market fit depends on user confidence and preparedness to let agents into how work actually gets done. Governance, permissions, transparency, and reliability determine whether agents move beyond pilot programs and into production.
- Capability needs infrastructure to match. Agents perform at a different level when they're grounded in live project data, team history, and structured workflows, and at enterprise scale the backend has to hold. Alongside the agent layer, monday invested in monday DB so the data infrastructure could support the volume, speed, and complexity of agents operating across an organization.
- Build on what already works. monday has always described itself as the place where people team up to drive business outcomes, and the agent-first rebuild extends that promise to a new kind of team member. People still come to monday to achieve their goals, the difference is that some of the team members working alongside them are now agents.