初创公司尽调
尽调报告 infrastructure / devtools Post-Series C / acquired 2026-06-04

Windsurf

Windsurf / Codeium 平台转型的战略资产尽调

Windsurf 做出了具备战略价值的 AI 编程平台,也有真实企业牵引力;但独立公司事实上已被拆成团队、技术和后续产品几块,干净的新股权入口已经消失。

封面要素

Series C 融资额 01
150 USD M [CO014]
Series C 估值 02
1250 USD M [CO015]
Cognition 签约时 ARR 03
82 USD M [CO019]
签约时企业客户数 04
350 customers+ [CO020]
当前继任产品用户数 05
1000000 users+ [CO023]

公司概况

Windsurf 2021 年以 Exafunction 起步,后来以 Codeium 身份转向 AI 编程软件;2024 年底推出 Windsurf Editor 后,成为增长最快的 AI 开发者工具创业公司之一。它的产品逻辑把 IDE 原生辅助、代码库上下文、智能体工作流和企业部署控制结合起来,服务的是不满足于单纯插件式编程副驾驶的组织。 Windsurf 的独立创业故事实际上在 2025 年 7 月结束。Google 通过一笔大型人才加技术许可交易挖走关键创始人和技术负责人,Cognition 则收购了剩余产品、IP、品牌、团队和客户基础。到 2026 年本次运行日,windsurf.com 已跳转到 Devin Desktop,因此核心尽调问题不再是 Windsurf 能否融到下一轮干净融资,而是有多少战略价值和产品价值在新所有者接手后保留下来。

官网
windsurf.com
成立时间
2021-01-01
创始人
Varun Mohan, Douglas Chen
创立地点
Mountain View, CA
总部
Mountain View, CA
产品
历史上,Windsurf 先销售 AI 编程扩展,后来推出 AI 原生 IDE,核心是 Cascade 式智能体工作流、深度代码库上下文和企业级控制。当前继任形态通过 Devin Desktop 呈现这些能力,把本地 IDE 工作、云端智能体、广泛模型支持和企业治理功能合在一起。
客户
个人开发者、产品驱动团队,以及拥有复杂代码库和安全要求的企业工程组织
商业模式
免费增值自助采用,加上付费 Pro / Max / Teams 方案,以及带部署、管理和安全附加项的定制企业合同
阶段
Post-Series C / acquired
融资情况
2024 年 8 月以 $1.25B 估值完成 $150M Series C;2025 年 Google 和 Cognition 交易终结了干净的独立融资路径
[CO014, CO015, CO016, CO018, CO022]

执行摘要

主要优势

  • 2025 年所有权转换前,Windsurf 在 AI-native 编程上已体现强产品市场匹配,并拿到有意义的企业采用
  • Google 的许可 / 人才交易和 Cognition 收购剩余业务,都验证了它的战略价值
  • 产品差异化来自智能体工作流、代码库上下文和企业部署控制,不只是简单自动补全
  • 通过当前 Devin Desktop 这一后续界面,原有装机基础看起来得到保留,甚至可能扩大

主要风险

  • 2025 年拆分交易后,它已不再是干净的独立公司,也不再是可融资的私募股权故事
  • 推理成本和经审计财务未公开,毛利率和单位经济性仍看不清
  • Google 挖走核心创始人和技术领导层后,管理连续性被打断
  • 后续产品经济性、客户留存和员工整合,如今绑定 Cognition,而不是独立 Windsurf 的股权结构

未决问题

  • Google 许可 / 人才交易以及独立 Cognition 收购的最终经济条款仍未披露
  • 独立 Windsurf 没有公开经审计收入、毛利率、烧钱速度或现金余额数据
  • 2025 年拆分交易中的董事会构成、投资人权利和员工期权处理,没有完整公开文件

目录

Chapter 01

01公司概况

1.1 身份、品牌与当前运营外壳

Windsurf 的公司故事已经不是一条简单的 Series C 软件公司叙事。业务 2021 年以 Exafunction 起步,聚焦 GPU 优化基础设施,随后在 2022 年以 Codeium 名义转向 AI 编程辅助。2025 年 4 月官方改名文章说得很清楚:管理层把 2024 年 11 月 Windsurf Editor 发布视为决定性转折,公司想从「代码补全扩展」转变为「智能体开发环境」。到本次运行日,转型又向前走了一步。windsurf.com 和 codeium.com 现在都跳转到 Devin Desktop,线上首页也明确告诉老 Windsurf 用户,其方案、定价、扩展和设置会通过 OTA 升级延续。尽调上,独立 Windsurf 品牌在历史上仍重要,但面向客户的运营外壳现在已经是 Cognition 主导的更大技术栈里的继任产品界面。这实质性改变了投资框架,也改变了后续章节的可比对象。[CO001, CO002, CO003, CO004, CO022, CO025]

里程碑表
日期里程碑类别证据质量评论
2021Exafunction 为 GPU 优化而创立创立最初的基础设施逻辑早于编码助手产品
2022转型并更名为 Codeium战略标志公司从基础设施转向 AI 开发者工具
2024-11Windsurf Editor 发布产品第一个智能体 IDE 定位改变了外部认知
2025-04-04公司正式更名为 Windsurf品牌官方文章称,公司希望身份与编辑器愿景匹配
2025-07Google 与 Cognition 执行拆分交易所有权创始人离开加入 Google,Cognition 收购剩余部分
2026-06windsurf.com 现已跳转到 Devin Desktop当前状态历史品牌仍重要,但活跃界面已经迁移

本表混合了成长期、融资和当前状态里程碑,因为报告的核心尽调问题是 Windsurf 是否仍作为同一个可投资实体存在。

[CO001, CO002, CO003, CO004, CO017, CO018]
FO001: Windsurf 公司演进时间线

约五年里,公司从基础设施创业公司转向代码助手,再到智能体 IDE,最终成为 Cognition 内部后继产品的界面。

[CO001, CO002, CO003, CO004, CO017, CO018]

1.2 创始人、领导层变化与治理可见度

公开创始人归因比提示词最初给出的背景更窄。本次审阅的较高质量公开档案一致把 Varun Mohan 和 Douglas Chen 识别为创始组合:Mohan 担任 CEO,Chen 是关键技术联合创始人。Google 在 2025 年 7 月完成反向收购式挖人交易后,临时领导权转给 Jeff Wang;他的公开表态把随后几天描述成危机应对式过渡,而不是计划内接班。治理可见度仍然很差。公司自有页面和可访问的第三方资料页,都没有披露足够细的董事会名单、独立董事或股东控制机制,无法支撑治理质量判断。这个不透明度很关键,因为公司经历了一种极不寻常的拆分交易结局:Google 挖走顶层技术领导层却没有取得控制权,随后 Cognition 收购剩余运营业务。公共治理细节缺失,使外部很难复原交接期间是谁为小股东、留任员工或客户谈下了哪些保护。[CO005, CO006, CO007, CO008, CO009, CO010]

领导层与创始人表
人物 / 角色公开角色2025 年交易后状态来源质量尽调备注
Varun Mohan联合创始人兼 CEO2025 年 7 月加入 Google战略愿景制定者和外部代表;已不再运营 Windsurf
Douglas Chen联合创始人兼技术负责人2025 年 7 月加入 Google他离开后,技术连续性风险上升
Jeff WangGoogle 交易后的临时 CEOCognition 交接期间的桥接运营者暗示这是紧急过渡,而非有序继任
董事会 / 独立董事未公开披露Unknown治理尽调仍不完整

领导层表刻意只纳入可访问来源明确记录的公开人物;本章不推断未披露的董事会构成。

[CO007, CO008, CO009, CO010, CO011, CO039]

1.3 融资历史与 2025 年交易序列

作为一家独立风投支持公司,Windsurf 最干净的资本故事止于 2024 年 8 月 Series C:General Catalyst 领投约 $150M,累计融资增至约 $243M,估值约 $1.25B。此后,公司战略价值快速超出常规创业阶段标签。Sacra 和 TechCrunch 都描述了一条 2025 年 7 月序列:OpenAI 据称的收购兴趣到期,Google 支付约 $2.4B 挖走创始人和一部分研究团队,同时获得技术许可;随后 Cognition 签署最终协议,收购剩余业务。Cognition 官方公告值得注意,因为它披露了经营基本盘——$82M ARR、企业 ARR 环比翻倍、350 多家企业客户——同时强调被收购资产包包括产品、IP、商标和品牌。换句话说,Windsurf 不再是一家普通 Series C 创业公司,而是被拆分的战略资产;其经济价值不是通过一笔干净收购实现,而是通过人才、技术和客户关系转移兑现。[CO012, CO013, CO014, CO015, CO016, CO017]

利益相关方或投资人图谱
事件资本 / 对价日期关键参与方重要性
Series A$25M2022-04Greenoaks验证了转向 AI 编码基础设施的转型
Series B$65M2024-01投资方:Kleiner Perkins、Greenoaks、General Catalyst扩大企业商业化和产品投入
Series C$150M,估值 $1.25B2024-08投资方:General Catalyst、Kleiner Perkins、Greenoaks确立最后一个干净的独立私有市场估值
Google 反向收购式挖人 / 许可约 $2.4B 对价2025-07Google、Varun Mohan、Douglas Chen、部分团队成员在未转让公司控制权的情况下,将人才和技术变现
Cognition 确定性收购价格未披露2025-07Cognition 与剩余 Windsurf 业务转移产品、IP、品牌、ARR 基础和剩余团队

2025 年序列不是一次单一收购。Google 和 Cognition 执行了两笔独立交易,将人才与剩余运营业务拆开。

[CO012, CO013, CO014, CO015, CO016, CO017]
FO002: 当前状态与最后独立规模标记对比

本图把最后一轮干净独立融资标记,与 2025 年所有权转换期间及之后披露的运营指标作对比。

[CO015, CO016, CO019, CO020, CO023, CO024]
FO003: 2025 年资产拆分与后继控制流

关键尽调事件是一场拆分式结果:Google 招走领导层并取得技术授权,Cognition 收购剩余运营业务和品牌。

[CO017, CO018, CO019, CO020, CO022, CO025]

1.4 规模标记、当前界面与负面整合证据

Windsurf 最新的独立价值规模标记分两层。第一层是整合前运营规模:Sacra 估计到 2025 年 7 月 ARR 约 $82M、到 2025 年初活跃开发者超过 800,000、企业客户超过 1,000、每日处理 token 超过 100B。第二层是当前继任产品界面:Devin Desktop 首页声称用户超过 1M、企业客户超过 4,000,说明装机基础至少被保住了,也可能在新产品包装下继续扩张。不过,交易后证据并非全是正面。Economic Times 和 WinBuzzer 都描述了 Cognition 接手剩余 Windsurf 业务后的裁员、买断和极端工作预期。这些信号不否定被收购资产的质量,但确实抬高了整合和留存风险。尽调上更合理的结论是:Windsurf 打造了一个有分量、快速增长的企业开发者工具业务,但可投资实体已经转化为交易后的运营组件,不再是一家完整、准备融下一轮的私营公司。[CO023, CO024, CO026, CO027, CO028, CO030]

快照 KPI 与运营状态表
指标数值 / 状态截至时间置信度含义
独立估值$1.25B2024-082025 年交易序列前最后披露的风险投资估值
累计融资~$243M2025-07包括已披露的 A/B/C 轮,以及分析师档案中的未披露种子轮
Cognition 签约时 ARR$82M2025-07Cognition 官方公告披露了 ARR 数字
签约时企业客户350+2025-07Cognition 官方公告披露了该数字
签约时日活用户数十万2025-07Cognition 官方公告只披露了方向性规模
当前首页用户1M+2026-06继任产品页面显示,大规模安装基础仍在使用
当前首页企业客户4,000+2026-06继任产品页面暗示更广的商业覆盖面
当前阶段Series C 后 / 已被收购2026-06以当前形态看,并非独立融资候选标的
当前网站状态windsurf.com 跳转到 Devin Desktop2026-06表明品牌迁移到 Cognition 产品界面

将历史独立公司指标与当前继任产品观察合并,展示最后一轮风险投资与当前运营状态之间的差距。

[CO015, CO016, CO019, CO020, CO021, CO022]
Chapter 02

02市场分析

2.1 市场边界与现状替代品

应把 Windsurf 放在快速成形的 AI 编程工作站市场里分析,而不是放进更宽泛的通用 LLM 产品宇宙。当前 Windsurf 界面跳转到 Devin Desktop,但价值主张仍明显是智能体 IDE 形态:完整编辑器、本地和云端智能体、上下文共享 Spaces、ACP 驱动工具,以及集成审查。这让 Windsurf 最接近 Cursor 和其他 AI 原生环境所在的换 IDE 细分市场。它离经典自动补全工具更远,因为待完成任务已经不再是「建议下一行」,而是「在人工参与下规划、编辑、测试并审查多文件工作」。 因此,正确的市场边界包含三层支出。第一层是 AI 原生编辑器,要求买方改变主要编程界面。第二层是 GitHub Copilot 这类扩展优先的既有平台,留在现有编辑器和 GitHub 工作流里。第三层是 Claude Code、Codex 这类终端或云端智能体,它们不必成为主 IDE,也能完成有意义的工程工作。即使界面不同,这三层都在争夺同一笔开发者生产力预算。核心边界之外应排除通用聊天支出、广义 DevOps 或可观测性支出,以及面向非专业用户的无代码构建器。这些产品可能影响付费意愿,但并不直接解决 Windsurf 的核心专业软件开发任务。 这个边界很重要,因为 Windsurf 的经济性与编辑器切换绑定。若买方已经满足于 VS Code 或 JetBrains 里的 Copilot 扩展,迁移成本就低于选择 AI 原生工作站的买方。因此,Windsurf 必须在更深工作流编排足够重要、足以支撑替换既有编辑器习惯的场景里取胜,而不是只靠更便宜的补全。[CM001, CM002, CM003, CM004, CM005, CM006]

市场定义表
细分 / 类别纳入支出排除支出买方 / 付款方与 Windsurf 的相关性
AI 原生 IDE 与智能体工作站主编辑器席位、智能体额度、上下文管理、审查和自动化不含 AI 的通用 IDE 许可开发者、工程负责人、平台买方核心品类;这是 Windsurf 明确瞄准的市场
插件优先的编码助手嵌入现有 IDE 和 GitHub 工作流的 AI 助手席位独立编辑器迁移预算开发者、团队管理员、IT 买方主要存量替代品,因为它避免编辑器切换
终端或云端编码智能体任务委派、代码库感知编辑、测试或 PR 工作流通用聊天机器人支出高阶用户、平台工程师、自动化负责人邻近但有竞争性;无需新编辑器也可吸收高价值智能体工作流
私有或隔离环境代码助手安全部署、策略控制、私有模型路由无治理的商品化云 AI 额度CISO、受监管工程组织、采购在隐私敏感细分中有竞争性,而 Windsurf 仍需证明适配度
邻近软件开发 AI围绕编码工作流的测试、审查、迁移、安全附加功能可观测性、通用 DevOps、无代码构建器平台和工程负责人影响预算外沿,但不是最清晰的 Windsurf 核心市场边界

按工作流控制和预算所有权定义市场,而不是按模型家族定义;被排除的行是替代品或邻近领域,不是 Windsurf 品类核心收入。

[CM001, CM002, CM003, CM004, CM005, CM006]
FM001: Windsurf 相关市场层级

市场从广泛的 AI 编码试用,收窄到 Windsurf 必须赢下的切换 IDE 细分市场。

各层级混合了调查采用率、官方定价和公司规模信号;公开数据更适合约束市场视角,而不是给出单一经审计 TAM 数字。

[CM008, CM009, CM010, CM014, CM015, CM023]

2.2 通过采用率和收入代理指标测算规模

Windsurf 的 TAM / SAM / SOM 栈无法像成熟软件品类那样被公开清晰观察到。公开证据对采用率和商业化代理指标更强,对一个被普遍接受的美元规模估计更弱,所以应通过几组受约束的视角来测算市场。采用率视角已经很大:GitHub 国际调查称,超过 97% 的受访者曾在工作中使用过 AI 编程工具;JetBrains 则报告 69% 试过用 ChatGPT 编程,40% 试过 Copilot。常规使用率低于试用率,但仍然可观:JetBrains 看到 49% 常规使用 ChatGPT 编程,26% 常规使用 Copilot。再结合 GitHub 2024 年生成式 AI 项目和贡献的激增,说明这个品类已经越过新奇阶段。 商业化代理指标指向同一方向。GitHub 把 Copilot 定位为拥有数百万个人用户和数万商业客户的产品;Cursor 已经提出官方企业渗透声称,这在两年前对一家 IDE 创业公司来说还显得难以置信;Sacra 和 TechCrunch 都显示,在 2025 年结束前很久,投资者和付费客户就已经奖励了 Cursor 的收入爬坡。Windsurf 自身继任界面声称用户超过 1M、企业客户超过 4,000。这些数字不能证明所有用户都在付费,也不能证明该基数能干净归因于收购前 Windsurf 产品,但它们确实证明:智能体编辑器的可寻址市场已经以数百万用户和数千企业账户来衡量,而不是少数灯塔试点。 实际含义是,Windsurf 应按一条已验证的高价开发者软件预算线来估值。该品类已经大到足以支撑差异化自助、团队和企业定价,但仍年轻到精确市场份额数学比采用率和收入速度数据更嘈杂。[CM008, CM009, CM010, CM011, CM012, CM013]

规模测算与采用视角表
视角观察指标数值来源集合含义关键局限
工作场景试用视角曾在工作中使用过 AI 编码工具的开发者超过 97% 的 GitHub 调研受访者GitHub 调研品类认知和试用已经主流化基于调研;不是付费或常规使用指标
常规使用视角经常使用 ChatGPT 编码的开发者JetBrains 受访者的 49%JetBrains 调研常规编码 AI 行为已经普遍覆盖一个工具类别和调研受众,不代表整个专业市场
常规使用视角经常使用 GitHub Copilot 的开发者JetBrains 受访者的 26%JetBrains 调研付费助手行为在专业规模上已具备意义针对特定工具,并不直接等同于所有 AI 编码使用
生产力视角受控任务中的观察速度提升使用 Copilot 快 55%GitHub 研究给预算负责人一个简单 ROI 故事单一受控任务;不能普遍适用于每个工作流
AI 原生 IDE 变现视角Cursor ARR 和付费用户$200M ARR;截至 2025 年 3 月约 720k 付费用户Sacra显示 AI 原生编辑器可以在品类成熟前规模化变现分析师估计,并非经审计申报数据
继任产品界面视角当前 Windsurf / Devin 用户和企业覆盖面1M+ 用户;4,000+ 企业客户Windsurf / Devin 编辑器页面证明今天已存在大型智能体编辑器安装基础未区分老 Windsurf 用户、收购后增长或付费席位

这里使用采用与变现代理指标,因为公开证据在席位数、使用量和收入速度上远强于单一公认 TAM 数字。

[CM008, CM009, CM010, CM011, CM012, CM020]
FM002: 2026 年 AI 编码工具年化席位价格区间

从个人试用到重治理重度使用,公开定价已经给出清晰的低 / 基准 / 高锚点。

所有数值都由当前公开标价年化得出;除非明确定价,否则不计入企业定制合同和按用量计费的超额费用。

[CM014, CM015, CM016, CM017]

2.3 买方地图、预算归属与采用路径

AI 编程的买方分层,与其说由公司规模定义,不如说由谁承担迁移风险定义。个人开发者是摩擦最低的入口,因为决策本质上是:一款每月 $10 到 $20 的工具,能否带来足够的日常生产力或使用愉悦感,值得刷信用卡。团队方案出现于工程经理或创始人想要一致性、用量可见性、隐私默认值和统一账单时。企业买方又加一层:采购、安全、身份、审计、用量预算和模型治理政策都会成为产品的一部分,而不只是旁边的清单。Cursor 和 GitHub 的公开材料把这一点说得很明白;可变现产品不再只是模型访问,而是对智能体行为的管理控制。 这种结构有利于经典的自下而上再转自上而下动作。开发者会先于公司完全批准去试用 AI 工具,等 ROI 可见、安全问题可控后,团队再标准化。因此,当前定价锚点很关键。GitHub 低价个人方案拉动广泛试验,Cursor 的 $40 团队档捕获愿意为协作付费的专业团队,Tabnine 的高价私有部署面向受监管买方,AWS Q 则说明买方已经处在更广平台关系里时,用量型或池化经济性也可以接受。这个品类因此不是单一市场,而是一架付费意愿阶梯:从自助生产力到有治理的企业自动化。 对 Windsurf 来说,近期最强匹配是那些重视编辑器切换、想要更智能体优先工作流的细分。最弱匹配则是把既有平台分发、政府级证明或私有部署弹性置于 UX 新鲜感之上的买方。这些买方已经有可信替代品。[CM014, CM015, CM016, CM017, CM023, CM024]

买方 / 用户 / 付款方图谱
细分买方用户付款方 / 预算所有者采用触发因素Windsurf 为什么能赢或不能赢
个人专业开发者开发者开发者个人信用卡或报销工具以 $10-$20/月换取更快日常编码当智能体优先 UX 明显优于插件时,Windsurf 可以赢
初创公司或 SMB 工程团队创始人 / CTO / 工程经理全工程团队部门预算需要一致工作流、共享上下文和团队控制适配度好,因为编辑器切换成本低,工作流收益出现快
中端市场产品工程组织工程副总裁或平台负责人开发者和技术负责人工程生产力预算自下而上采用成熟为标准化和管理控制若 Windsurf 能对 Copilot 或 Cursor 证明治理深度和 ROI,就有竞争力
Fortune 500 企业CIO / CISO / 平台领导层大型工程组织IT 采购和平台预算安全审查、SSO、可审计性、策略控制和可衡量铺开更难,因为存量厂商已拥有身份、代码库和云分发
受监管或隐私敏感买方CISO / 合规负责人受限工程团队安全和合规预算需要私有部署、隔离环境和数据处理确定性当前较难,因为竞争对手更明确宣传私有部署
AWS 原生平台团队云平台负责人开发者和平台工程师云生产力预算希望把编码帮助与 AWS 故障排查和转型结合这是困难细分,因为 Amazon Q 嵌入周边平台

一旦治理、隐私和铺开控制进入决策,预算所有权就会从个人生产力支出转向正式的平台和合规预算。

[CM014, CM015, CM016, CM017, CM023, CM024]
FM003: Windsurf 买方准备度矩阵

工作流深度越重要,Windsurf 越强;既有分发或私有部署主导时,Windsurf 最弱。

评级综合公开定价、治理、安全和分发证据;它们是方向性判断,不是来自调查的分数。

[CM023, CM024, CM025, CM029, CM030, CM031]

2.4 增长驱动、采用约束与 Windsurf 特定含义

最大市场驱动在于,AI 编程已经成为可衡量的工作流杠杆,而不是想象中的自动化剧场。GitHub 生产力研究给工程领导者一个简单 ROI 故事,Octoverse 和独立市场跟踪也显示开发者持续构建、测试并比较新的智能体工作流。竞争对手的产品路线图也收敛到同一终点:智能体能规划、编辑、运行命令、验证结果,并交回可审查的工作。这种收敛验证了 Windsurf 的原始逻辑:更好的编程界面可以围绕智能体来构建,而不是围绕菜单栏和点状工具。 约束同样真实。智能体会话变得长时运行且并行后,算力经济性正在收紧;GitHub 2026 年的暂停和用量上限直接提醒市场,在当前价格点下需求可能跑赢供给。法律和归因风险仍未解决,随着 AI 生成代码从试验进入企业受控交付流水线,这个问题更重要。分发也不对称:GitHub 可以把 AI 捆进许多团队已经使用的源码控制界面,AWS 可以把编程辅助压进更广的云工作流。最后,受监管买方仍可选择更强调私有部署、而非工作流新鲜感的工具。 对 Windsurf 来说,净含义是具体而非泛化的。公司不需要在每个基准测试或每个细分上击败所有对手。它需要拿下那些想要有治理、智能体优先开发环境,并相信更好编排值得改变主工具的团队所形成的换 IDE 楔子。如果这个楔子继续扩大,Windsurf 仍有战略价值。如果购买标准压向捆绑型既有平台、纯私有部署或中立终端智能体,Windsurf 获取高价份额的空间会很快变窄。[CM018, CM019, CM029, CM030, CM031, CM032]

增长驱动因素与采用约束表
因素类型时点证据对 Windsurf 的含义尽调要求
可衡量的生产力提升驱动因素当前GitHub 生产力研究支撑溢价席位定价和更快自下而上采用要求提供按细分拆分的 Windsurf 生产力和留存数据
开发者基础和项目增长驱动因素当前GitHub Octoverse 2024扩大所有编码 AI 产品的漏斗顶部测试 Windsurf 转化在成熟 AI 采用者还是首次采用者中最强
从助手转向智能体驱动因素当前至未来 24 个月Windsurf、GitHub、AWS、Anthropic 官方产品页面利好具备编排和审查界面的 AI 原生产品衡量 Windsurf 使用中来自智能体工作流而非补全的占比
企业治理需求驱动因素当前Cursor 企业版和安全、GitHub 文档、Tabnine 定价在个人订阅之外创造溢价预算层审查模型控制、可审计性和继任管理功能路线图
算力压力和用量上限约束当前GitHub 文档和 Developer-Tech在重度智能体负载下,可能压缩毛利率并损害用户体验要求提供按计划和智能体会话类型拆分的内部单位经济
版权与署名风险约束当前至 36 个月The Verge 诉讼报道可能拖慢企业采用,或推高赔偿压力要求提供 Windsurf 法务立场、赔偿条款和模型供应商合同
既有厂商分发优势约束当前GitHub 和 AWS 产品界面看重生态契合而非工作流深度的账户,Windsurf 更难拿下按细分市场量化对 GitHub 和 AWS 的竞争胜率
继任产品基盘不透明约束当前仅限 Windsurf / Devin 当前界面难以区分继承来的安装基盘和收购后动能要求拆分旧 Windsurf 用户、活跃付费席位和企业续约队列

将品类顺风与最可能压缩 Windsurf 转化率、定价权或企业扩张的具体约束配对。

[CM011, CM013, CM018, CM019, CM023, CM029]
FM004: 从个人试用到受控推广的采用路径

这一品类通常从个人试用扩展到团队标准化,再进入政策密集的企业推广阶段。

这条流程是结构性的,不是确定性的;不同细分市场会根据安全姿态和既有工具跳过或重复某些步骤。

[CM018, CM019, CM028, CM029, CM031, CM032]

2.5 图表

Chapter 03

03竞争格局

3.1 格局、竞争者类型与买方的第一选择

Windsurf 如今所在的市场里,买方可以通过几条非常不同的渠道解决同一件事。GitHub Copilot 和 Amazon Q Developer 是默认既有选项,因为它们借助现有开发者和云控制平面,而不是要求团队采用一个全新环境。GitHub 在 GitHub.com、IDE、终端、项目工具、聊天应用和 MCP 服务器中推广 Copilot,同时 GitHub 自身平台规模在 2024 年已越过 100M 开发者和 518M 项目。AWS 在云技术栈里做类似事情:Amazon Q 覆盖 IDE、CLI、AWS Console、Slack、Teams,甚至 GitHub 预览工作流,这让它成为已经在 AWS 身份和运维上标准化的团队最自然替代品。Cursor 是最重要的直接 IDE 原生对手,因为它把 VS Code fork 与云端智能体、Jira 工作流钩子,以及公司声称覆盖 64% Fortune 500 和 50,000 多家企业的渗透结合起来。Tabnine 从信任侧进攻,提供私有部署选项;Replit 从应用构建器侧进攻,提供并行智能体和浏览器原生产品构建。Claude Code 和 Devin 则进一步走向更高委托度的智能体工作流。战略结论是,Windsurf 不再与一个单一 AI 编程助手品类竞争;它同时面对拥有分发的既有平台、拥有工作流深度的挑战者,以及承诺比自动补全更高自主性的智能体原生工具。[CP001, CP003, CP007, CP009, CP013, CP020]

竞争对手画像表
竞争对手类别规模 / 分发目标细分市场关键差异化关键限制
GitHub Copilot既有工作流平台GitHub 开发者生态 100M+;数百万用户;$10-$100 自助阶梯广泛的专业开发者和企业凭 GitHub 原生分发覆盖 IDE、CLI、移动端和 GitHub.com消费者端训练信任口径与企业版分裂,且不支持 GitHub Enterprise Server
CursorIDE 原生挑战者Fortune 500 中 64% 使用 Cursor;50,000+ 家企业;据报道 ARR >$500M、估值 $9.9B专业工程师和工程团队VS Code 分叉,加上云智能体、Jira 工作流和强开发者偏好主张没有自研模型,企业定价仍需定制
Amazon Q Developer超大规模云厂商编码助手依托 AWS 分发,Pro 为 $19/user/mo,管理相邻场景广以 AWS 为标准的企业和重度云开发者覆盖 IDE/CLI/控制台,并接入 IAM 和运维上下文AWS 已是标准的组织里最有说服力
Tabnine隐私优先的企业利基$39 和 $59 标价档位;企业可选私有部署受监管或安全敏感组织VPC、本地、隔离部署和零留存姿态大众市场分发弱于 GitHub、AWS 或 Cursor
Replit云端应用构建替代品数百万开发者;免费至 $95 年付自助方案;企业定制初创公司、产品构建者、学生和小团队并行智能体、集成制品,以及在一个云工作区内快速出应用与成熟企业仓库治理的匹配度不那么直接
Claude Code前沿模型本地智能体Claude Pro/Max/Team 捆绑定价从 $20 到 $100+想要本地控制和前沿模型的重度用户与团队本地运行,使用测试和 PR,并要求明确批准比云优先企业平台更难集中管理
Devin / Windsurf 继任界面云智能体与继任平台个人定价从免费到 $200,另有团队和企业方案愿意委托端到端编码工作的团队云智能体,加上 GitHub、Slack、Jira、MCP 和专用部署基准优势已被压缩,Windsurf 品牌现在嵌入更宽的产品界面

表格聚焦同一任务场景中最值得关注的替代方案。规模一栏混合公开生态指标、公开定价,以及私营公司未披露财务数据时的可靠新闻报道。

[CP002, CP007, CP008, CP009, CP012, CP014]
FP001: 竞争定位图

基于证据的序位图,对比 Windsurf 主要替代方案的分发能力与自主深度。

坐标轴采用证据支撑的 0-10 序数评分。分发能力反映既有安装基础和采购邻近性;自主深度反映工具在不退回普通自动补全的情况下,能把规划、编辑、运行和验证推进多远。

[CP020, CP024, CP031, CP033, CP034, CP035]

3.2 各竞争者的能力、定价与信任姿态

仅看标价,Windsurf 已经没有明显价值楔子。GitHub Copilot 自助阶梯从 $10 的 Pro 起,到 $100 的 Max;Amazon Q Pro 是每用户每月 $19;Cursor Individual 是 $20,Teams 是每用户每月 $40;Claude Code 包在 Claude Pro 里,月付 $20 或年付折合 $17;Devin 当前继任界面从 $20 起,并有 $200 Max 档。Tabnine 的代码助手标价 $39,智能体平台标价 $59,更高,但溢价买的是另一套采购叙事:VPC、本地部署、全隔离部署,加上明确的零留存和不训练表述。Replit 的界面更偏用量形态,Starter 层有免费每日智能体额度,Core 年付时每月 $20,Pro 年付时每月 $95。能力宽度也已经收敛。GitHub 现在在 GitHub 和终端中提供智能体工作流,Cursor 从编辑器原生智能体延伸到云端智能体和 Jira,Amazon Q 把编程帮助与 AWS 运维捆绑,Claude Code 从本地机器运行测试并发起 PR,Replit 推并行应用构建智能体,Devin 则在 GitHub、Slack、Jira、MCP 和专用部署选项之上叠加云执行。从 Windsurf 视角看,剩余空白不是原始功能勾选框,而是买方是否足够重视深度自主执行,愿意接受比超大规模云厂商和隐私优先供应商更弱的信任或分发姿态。[CP002, CP004, CP008, CP010, CP011, CP014]

功能 / 能力矩阵
能力Devin / WindsurfCursorGitHub CopilotAmazon QTabnineReplitClaude Code
独立 IDE 或专用工作区部分 — 云工作区加桌面界面是 — VS Code 分叉否 — 扩展加 GitHub 界面否 — 扩展加 AWS 界面否 — IDE 扩展和平台是 — 浏览器工作区部分 — 终端、IDE 和 Web 钩子,而非完整 IDE
云端或后台智能体是 — 云智能体是 — 云智能体和自动化是 — 后台智能体工作流部分 — 支持智能体请求,但偏管理中心部分 — 智能体平台和无头附加组件是 — 并行智能体部分 — 本地优先,可选 GitHub Actions
终端原生工作流部分 — API 和智能体工作流,并非终端优先是 — shell、自动化、/loop是 — Copilot CLI 和终端工作流是 — CLI 补全和聊天是 — Tabnine CLI部分 — 不是主要界面是 — 终端是主要界面
GitHub / Jira / Slack 邻近性是 — 原生 GitHub、Jira、Slack、MCP是 — Jira 加更广自动化是 — GitHub 原生,加项目工具和聊天应用是 — GitHub 预览,加 Slack 和 Teams部分 — Jira 和企业集成部分 — 自有工作区内协作是 — GitHub、GitLab、Slack 和命令行
企业管理和策略控制是 — 企业管理控制和专用部署是 — SAML、SCIM、审计和访问控制是 — 集中管理、策略控制、DPA是 — IAM Identity Center 和 AWS 治理是 — 治理控制和分析部分 — 企业隐私控制,但编码专属治理细节较少部分 — Team/Enterprise 管理和 SSO,但本地优先用法仍由用户驱动
私有部署 / 最强信任姿态部分 — 专用部署和选择加入训练部分 — 隐私模式和零数据留存条款部分 — 企业保护强,但消费者训练选择退出口径分裂部分 — IAM 和 AWS 数据控制强,但非隔离部署是 — VPC、本地部署或完全隔离部分 — SOC 2 云姿态部分 — 付费团队界面默认不训练,但主要控制来自本机工作流
入门档经济信号$20 Pro$20 Individual$10 Pro$19 Pro$39 Code AssistantFree Starter / $20 Core 年付$20 月付 Pro

各行混合二元判断和定性单元格,因为产品都向宽能力族收敛,但运营模型和买家匹配差异很大。「部分」表示能力存在,但范围更窄、由用户驱动,或绑定不同部署模型。

[CP001, CP003, CP008, CP010, CP011, CP013]
定价 / 打包比较
厂商自助入门价团队 / 商业版信号企业版姿态打包细节Windsurf 含义
GitHub CopilotFree $0;Pro $10;Pro+ $39;Max $100Business 和 Enterprise 集中管理、额度池化企业 DPA、策略控制、GitHub.com 索引重度智能体使用也消耗 AI 额度,部分工作流消耗 GitHub Actions 分钟数最便宜的可信既有厂商,也是 GitHub 重度团队最容易加购的工具
CursorHobby 免费;Individual $20团队价:Teams $40/user/moEnterprise 定制高级模型、云智能体、Bugbot、团队市场、隐私模式Windsurf 已很难只靠价格竞争
Amazon Q Developer免费层零成本专业版:Pro $19/user/moIAM Identity Center 和 AWS 账户管理Transformation 超额按 LOC 计费,$0.003价格近乎持平,且更贴合超大规模云厂商采购
TabnineCode Assistant 价:$39/user/moAgentic Platform 价:$59/user/mo可私有部署更高标价搭配不训练和私有部署选择信任优先的买家可能愿意为避免纯云取舍支付溢价
ReplitStarter 免费,含每日 Agent 额度Core $20 年付 / $25 月付;Pro $95 年付 / $100 月付Enterprise 定制,含 SSO/SAML 和隐私控制额度驱动定价更贴合应用构建和运行时用量,而非按席位编码助手对从零搭建产品的团队是相关替代品,作为直接企业 IDE 替代则不够清晰
Claude CodeClaude Pro $17 年付 / $20 月付Team $20 年付 / $25 月付每席;Max 从 $100 起Enterprise 定制Claude Code 打包进更宽的 Claude 方案,而不是单独按席位销售对想要本地控制和前沿模型的重度用户,价值很强
Devin / Windsurf 继任产品Free $0;Pro $20;Max $200Teams $80/month,加 $40/mo / 完整开发者席位Enterprise 定制用量额度按日、按周刷新,额外用量按 API 定价出售继任定价在核心付费档位上已不比 Cursor、Amazon Q 或 Claude Code 便宜

公开标价不是企业实际成交价。表格展示买家在折扣、量级承诺、AI 额度超额或协商部署条款之前看到的价格。

[CP002, CP008, CP014, CP017, CP018, CP021]
FP002: 功能广度 / 能力图谱

按战略采购标准展示主要竞争对手各自最强位置的品类级图谱。

这些单元格概括公开证据,并不声称经过审计的量化评分。AWS 内强表示强势地位集中在 AWS 标准化环境内,而不是覆盖所有软件团队。

[CP010, CP013, CP016, CP019, CP020, CP023]

3.3 切换成本、分发权力与现状 / 内部自建替代品

分发和切换成本比基准测试截图更能解释这个品类。GitHub 能把 Copilot 卖进既有 GitHub 采购、代码库管理、移动端、CLI 和知识工作流;这让它成为许多团队评估 AI IDE 前会先加上的默认工具。Amazon Q 在高度依赖 AWS 的环境里也有类似优势,因为身份、计费、console 操作和编程支持都处在一个熟悉的管理界面里。Cursor 的切换成本不同:一旦团队在它的 VS Code fork、SAML、SCIM、共享规则、云端智能体、自动化和 Jira 交接上标准化,工具就会嵌入日常工程管理,而不是作为可选扩展坐在编辑器旁边。Tabnine 的粘性由采购驱动,而不是产品驱动:需要 VPC、本地或隔离部署的受监管团队,可能只有少数可用供应商,竞争场马上收窄。Claude Code 很强,但仍是本地优先、强调批准流程,因此容易与既有环境并行试点。Replit 在从零应用构建上很容易采用,但这让它更像快速产品创建的替代品,而不是企业代码库治理工具。因此,Windsurf 最重要的竞争者之一其实是现状本身:团队可以在 IDE、终端、MCP 工具、CI/CD API 和特定供应商智能体之间多平台并用,而不是承诺使用单一助手。在这种世界里,买方惯性利好既有平台,也提高了 Windsurf 的证明负担:它必须证明自己能成为权威系统,而不只是另一个 AI 标签页。[CP003, CP010, CP011, CP015, CP016, CP019]

3.4 从 Windsurf 视角看护城河耐久性与负面证据

负面证据让这个品类很难承销。GitHub 2026 年 3 月隐私更新划出企业账户和消费者 Copilot 层之间的清晰边界:Free、Pro、Pro+ 交互数据可能被用于训练,除非用户选择退出;这制造了信任分层机会,但只在边缘有效,因为 GitHub 保留了更强的企业保护。Devin 面对另一种负面模式:Cognition 最初技术报告把 13.86% SWE-bench 成绩确立为 2024 年的突破时刻,但官方 SWE-bench 生态后来公开了更新或开放智能体高得多的 Verified 分数。这不能证明 Devin 在生产中很弱,但确实说明基准测试驱动的差异化压缩得有多快。Cursor 最强的采用率和偏好数字来自公司自称,意味着买方仍需现场证明,看看这种热度有多少能穿过采购、合规和大代码库推广。Replit 的公开材料也展示了相邻替代品如何通过把设计、部署和智能体执行捆进一个云工作区,吸走经典 AI IDE 的注意力。从 Windsurf 视角看,净结果偏冷:它原本作为插件既有平台与完全自主智能体之间 AI 原生中间地带的承诺,已经变窄。持久优势现在取决于能否证明更好的信任、更深的工作流嵌入,或在真实客户任务上可衡量地执行更好,而不是简单匹配市场价格或重复昨天的基准测试头条。[CP004, CP005, CP006, CP012, CP023, CP029]

护城河耐久性 / 竞争风险登记表
Windsurf 护城河主张威胁严重度公开证据缓释 / 尽调问题
编码环境里的智能体执行足够形成差异化GitHub、Cursor、Amazon Q、Claude Code 和 Replit 现在都主打某种智能体工作流智能体界面已覆盖 IDE、终端、云智能体和项目工具在真实客户仓库上衡量任务级胜率,而不是依赖品类标签
价格仍能抢份额Copilot Pro $10 和 Amazon Q Pro $19 压低低端价格带,而 Cursor、Claude 和 Devin 聚在 $20 左右官方公开定价显示,多数严肃个人付费档位已挤在窄价格带里用转化率、周期时间或质量数据证明 ROI,而不是低价席位话术
信任方案足以通过企业采购Tabnine 提供 VPC / 本地 / 隔离部署;GitHub 和 AWS 提供更强的管理和合同界面竞争对手越来越把消费者试用和企业治理隔开记录部署模型、训练默认设置和客户可控管理界面
基准测试领先足以撑起品牌SWE-bench 头条变化快过产品口碑,开放智能体成绩已超过 Devin 早期发布分数官方基准生态里,Verified 结果已高于 Devin 最初 13.86% 的发布报告在客户私有代码库上跑受控对测,并公布方法论
仅靠集成就能形成锁定MCP、IDE 扩展、CLI 和 API 让很多团队很容易多工具并用多数对手现在都连接 GitHub、Jira、Slack、终端或外部工具的某种组合跟踪有多少客户将平台标准化,而不是把它当第二工具使用
品牌切换不会影响竞争位置公开 Windsurf 入口如今指向 Devin Desktop,容易模糊买家究竟在评估什么继任产品定位强化自动化,但削弱旧有独立 IDE 身份要求提供旧 Windsurf 账户的队列迁移、续约和扩张数据

严重度反映投资者相关性,不只是客户侧严重度。多行是综合性判断,结合了多个竞争对手的定价、基准测试和采购证据。

[CP004, CP029, CP030, CP032, CP033, CP034]
FP003: 护城河 / 就绪度 KPI

用简明竞争读数呈现最可能决定 Windsurf 能否保住持久差异化的力量。

这些 KPI 混合了公开标价、公开基准引用和定性分析师综合判断。它们旨在概括战略就绪度,不是经过审计的公司指标。

[CP029, CP030, CP032, CP033, CP035, CP036]

3.5 图表

Chapter 04

04财务情况

4.1 收入模型转型与结构可见度

Windsurf 的商业化故事更像三个阶段,而不是一张稳定 SaaS 方案表。2025 年 4 月,公司公开简化定价,取消 flow action 额度,转向 prompt 额度方案:Pro 每月 $15、Teams 每用户每月 $30、Enterprise 每用户每月 $60,并提供池化附加额度和企业部署选项。2026 年 3 月,公司再次调整,这次用每日和每周配额替代额度,保留 Free 和 Pro,新增 $200 Max 档,并把付费超额用量转为 API 标价。到 2026 年 6 月本次运行日,实时客户界面已不再以独立 Windsurf 品牌呈现,而是 Devin Desktop;Free、Pro、Max、Teams 和 Enterprise 仍存在,但团队包变成 $80 基础费加每个完整开发者席位 $40。这个时间线很重要,因为它显示公司在模型行为变化时,反复尝试让商业化同时贴合智能体成本和客户心理。 收入结构方向上可见,但没有量化披露。官方收购材料确认,到 2025 年 7 月 Windsurf 已是 $82M ARR 业务,企业 ARR 环比翻倍,企业客户超过 350 家;而 Sacra 更早的估计仍显示更广客户基础里有数十万用户和超过 1,000 家企业。合起来看,这与 PLG 到企业漏斗一致:免费和付费自助方案制造使用量与分发,随后团队和企业层把集中账单、分析、部署和安全需求变现。缺口在于精确拆分。没有公开来源说明 ARR 中有多少来自低 ARPU 自助席位、多少来自更大的企业合同,所以正确承销姿态是:商业化偏企业,但新客户创建仍实质依赖自助漏斗。[CI001, CI002, CI003, CI004, CI005, CI006]

收入流表
收入流机制单位当前数值 / 状态收入质量尽调问题
免费漏斗免费增值获客用户活跃但不产生收入;现在位于 Devin Desktop 内直接收入低,但管线价值高需要免费 MAU、激活率和免费转付费转化率
Pro 自助月订阅加额度现在 $20/mo;2025 年重置时 $15/mo2025 和 2026 方案里的核心个人付费档位经常性收入中高,但暴露于模型成本波动需要付费席位数、超额用量附加率和流失率
Max / 重度用户自助高阶月订阅2026 年 $200/mo;历史上部分重度用户购买附加包或使用 $60 Pro Ultimate新设的明确重度用户变现层中等经常性收入,但成本可能偏重需要 Max 附加率和重度用户队列毛利率
Teams带管理控制的按席位或组织套餐2025 年 $30/user/mo;2026 年 3 月 $40/user/mo;继任页面现在显示 $80 基础费 + $40/full dev seat从 PLG 走向受管组织支出的商业桥梁管理工具粘住后,经常性收入质量高需要团队转化率、平均席位数和实际 ARPU
Enterprise合同席位、部署、支持2025 年重置时 $60/user/mo;继任界面定制收购时 350+ 家企业客户;2026 年调整中保留既有计费协议合同能续约并扩张,收入质量很高需要 ACV、合同期限、续约率和部署组合
额外用量 / 附加包2025 年为 prompt 额度包;2026 年为按 API 定价的超额用量2025 年 $10/250 额度或 $40/1,000 额度;2026 年按 API 定价把高强度高级模型消耗变现中等,因为收入跟用量和成本绑定需要来自超额用量的 ARR 占比和成本转嫁毛利

表格区分漏斗、经常性订阅、企业合同和用量绑定变现;标价不是实际净收入。

[CI001, CI002, CI003, CI004, CI006, CI007]
定价 / 变现表
时期 / 方案标价用量模型变化含义来源
2025 年 4 月 Pro$15/mo500 个 prompt 额度;再买 250 个 $10用 prompt 计费取代旧的 flow-action 逻辑个人 ARPU 更清晰,但用户体验仍是有限额度Windsurf 定价 v2
2025 年 4 月 Teams$30/user/mo每用户 500 个 prompt 额度;1,000 个池化附加额度 $40基础额度不再池化;新增管理控制显示以管理功能向 SMB 团队变现和增售Windsurf 定价 v2
2025 年 4 月 Enterprise$60/user/mo每用户 1,000 个 prompt 额度;池化附加额度单一企业席位类型,并可选批量折扣 / 混合部署 / FedRAMP更高 ARPU 绑定安全与部署Windsurf 定价 v2
2026 年 3 月 Free$0/mo每日 / 每周额度取消积分额度仍保留 PLG 漏斗Devin 博客定价方案
2026 年 3 月 Pro$20/mo每日 / 每周额度;额外用量按 API 价格价格上调 $5;积分包改为配额 + 超额用量抬高重度用户变现上限Devin 博客定价方案
2026 年 3 月 Max$200/mo更高的日 / 周配额面向重度智能体用户的新档位明确切出高 ARPU 自助服务档位Devin 博客定价方案
2026 年 3 月 Teams$40/user/mo每席位标准配额配额制取代月度积分保留从 PLG 到团队的路径,但支出约束更强Devin 博客定价方案
报告日继任 Teams$80 基础费 + $40/完整开发席位团队方案 + 席位收费当前线上包装已超出 2026 年 3 月 Windsurf 公告显示收购后套餐仍在演进Devin 定价
报告日继任 Enterprise定制既有协议或谈判后的企业定价企业合同保留在自助服务配额迁移之外企业收入确认方式可能不同于自助订阅Devin 定价 + 定价方案博客

这是一张时间线表:它拆开 2025 年提示词积分重置、2026 年 3 月配额上线、以及 2026 年 6 月继任定价界面, 而不是假装全程只有一张静态价目表。

[CI001, CI002, CI003, CI004, CI005, CI006]
FI001: 收入模型桥接图

Windsurf 如何把免费使用转化为自助订阅、团队扩张、企业合同和用量挂钩的货币化。

[CI003, CI006, CI010, CI011, CI017]
FI002: 财务估计区间

公开可支撑的 ARR、标价和继任所有者融资标记区间。

融资和战略交易标记的中点,是在相互冲突或时间错位的公开数字之间放置的分析占位;低值和高值为有来源支撑的边界。

[CI004, CI008, CI025, CI026, CI027, CI028]

4.2 单位经济性、推理成本压力与 GTM 效率代理指标

公开证据对核心经济张力异常清楚:Windsurf 有真实商业需求,但据称毛利率很差。Sacra 的描述很直白:前沿模型推理成本超过了公司向用户收取的价格;考虑到高端编程模型的成本曲线,以及智能体会话会随着时间扩大上下文和工具用量,这很可信。OpenAI 的公开 API 价格表显示高端模型成本会如何随着 token 量快速放大;GitLab FY2025 10-K 则提供了有用对照:一家成熟开发者工具软件公司的毛利率为 89%,同时明确警告 AI 功能可能压低这一利润率。换句话说,Windsurf 据称的利润率压力不像普通软件分发低效,更像这个品类特有的成本:把重度依赖前沿模型的智能体工作流卖得太便宜、定价太简单。 定价重设计看起来就是在修复这种错配。2025 年从 flow-action 额度转向 prompt 额度,是在应对一个单条 prompt 可能触发远多于以往工具调用的环境。2026 年转向配额加 API 标价超额用量,则更进一步,让重度使用更明显地变成可变成本,同时把企业协议与自助计费分开。这应当改善重度用户商业化,但不能证明毛利率健康,因为基础配额仍补贴部分高端模型使用,公司也没有披露超额用量结构、模型路由结构或各档贡献利润率。 GTM 效率比量化更容易框定。Windsurf 显然用 PLG 做分发,再在漏斗上部通过企业控制和部署功能变现。Nubank 案例研究为企业销售提供了真正 ROI 证据,但公开记录仍缺 CAC、回本周期、NRR 和流失率。结果是:企业价值主张可信,销售效率数据不完整。[CI013, CI014, CI015, CI016, CI017, CI018]

单位经济性表
指标数值 / 公开信号可信度为何重要尽调需求
收购时 ARR$82M说明出售前已有可观收入规模需要按细分拆解、经审计或董事会批准的 ARR 桥表
2025 年初 ARR~$40M说明公司在收购前增长很快需要月度 ARR 历史及各档位贡献
收购时企业客户数350+证明公司已有不可忽视的自上而下业务,而不只是 PLG需要 ACV 分布和前 20 大客户集中度
毛利率据报明显为负长期经济性的核心问题需要按套餐和模型族拆分毛利率
高端模型成本敞口OpenAI 旗舰模型定价为每 1M tokens 输入 $2.50-5、输出 $15-30解释了为什么大量使用前沿模型会吞掉毛利需要模型路由组合和每次付费会话平均 tokens
超额用量变现2026 年额外用量按 API 定价计费可能降低对重度用户的补贴需要超额用量贡献的收入和毛利占比
传统软件基准GitLab FY2025 毛利率 89%显示 AI 原生推理经济性可能比经典软件范式低很多需要 AI 原生毛利率目标的内部可比公司组
CAC / 回本周期未披露评估 GTM 效率需要该指标需要按细分拆分的全口径 CAC、回本周期和销售周期长度
NRR / 流失未披露检验 ARR 持久性需要该指标需要按队列拆分的 logo 流失、总留存率和 NRR

这些类似空值的行是有意保留:本章点名公开承销最缺的精确指标,而不是用薄弱代理指标推断。

[CI007, CI008, CI013, CI014, CI015, CI017]
FI003: 单位经济模型桥接图

从订阅和超额收入到未披露毛利率结果的示意桥接;结果受模型组合和部署选择牵动。

由于公司不按套餐、模型或部署类型披露利润率,该图只表示方向,不做完整量化。

[CI013, CI014, CI017, CI018, CI041]

4.3 Series C 之后及 2025 年所有权变化后的资本充足性

Windsurf 干净的独立资本故事止于 Series C。公开摘要一致把累计融资放在约 $243M,其中 2024 年 8 月完成 $150M Series C,估值 $1.25B。这足以说明,在 2025 年交易序列之前,公司并非明显处在资本饥饿状态。但 2025 年 7 月从根本上改变了「现金跑道」的含义。Google 约 $2.4B 的人才加技术交易没有买下整家公司;随后 Cognition 同意收购剩余业务,包括产品、IP、品牌和运营收入基础,金额未披露。此后,相关偿付能力问题从 Windsurf 的独立烧钱,转向 Cognition 是否愿意、也是否有能力为整合和持续产品开发出资。 在这个更窄问题上,母公司看起来资金充足。多家高质量媒体报道 Cognition 在 2025 年 9 月以 $10.2B 估值融资约 $400M,另一篇报道则称同一轮接近 $500M、估值 $9.8B。差异本身是披露缺口,但两个版本都指向同一方向:买方在交易完成后不久拿到了新资金。VentureBeat 和 TechCrunch 也都重复称,Cognition 自成立以来净烧钱保持在 $20M 以下,同时 Devin ARR 从 2024 年 9 月的 $1M 增至 2025 年 6 月的 $73M,意味着融资前资本效率异常高。 这并不意味着收购后没有财务风险。裁员和九个月买断要约说明整合成本真实存在,也可能带来客户服务中断和员工留存压力。但风险在于整合质量和留存收入,而不是产品界面层面的近期流动性。[CI025, CI026, CI027, CI028, CI029, CI030]

资本充足性表
项目公开数值 / 状态可支持结论为何重要尽调需求
出售前累计融资~$243MSacra 与二级资料支持界定最后一个清晰的独立融资基准需要 Series C 后的股权结构表和现金余额
Series C$150M,估值 $1.25B可支持其为最后一轮清晰的独立融资最后一个风险投资估值锚需要资金用途备忘录和签约时剩余现金
出售时独立公司账上现金未披露未找到公开资产负债表数字无法计算干净的现金跑道需要 2025 年 7 月交易前最近月末现金
独立公司月度烧钱速度未披露未找到 Windsurf 直接公开的烧钱指标无法检验独立经营时的资本强度需要截至出售交割、按季度拆分的月度烧钱速度
独立现金跑道(月)未披露无法用公开记录计算缺失的关键偿付能力指标需要现金与烧钱桥表
债务 / 项目融资义务公开未披露已审阅来源未显示债务或项目融资证据未披露不等于没有债务需要债务明细、租约和表外承诺
Google 交易~$2.4B 反向人才收购 / 许可适用于创始人、部分研究员和技术许可,而非整家公司收购解释了后续融资路径为何终止需要按利益相关方类别拆分收益分配
Cognition 收购价格未披露;收购范围包括产品 + 品牌 + IP + 业务官方口径包括收入基础和运营业务资本充足性问题转到买方资产负债表需要购买价分摊和留任包成本
Cognition 收购后融资据报为 $400M、估值 $10.2B,或接近 $500M、估值 $9.8B二手报道对具体规模和估值不一致仍显示母公司融资能力很强需要已签融资文件或股权结构表更新
母公司烧钱披露2025 年 9 月融资前,自创立以来净烧钱低于 $20M多家媒体重复报道叠加新融资后,暗示现金跑道充足需要收购后烧钱速度和整合支出节奏

资本充足性有意拆成最终独立公司与出售后所有者两段,因为 2025 年 7 月交易序列改变了真正相关的偿付能力问题。

[CI025, CI026, CI027, CI028, CI029, CI030]
FI004: 资本强度 / 现金流图谱

从 Series A/B/C 融资到 2025 年拆分交易,再到资本更充足继任所有者的资金路径。

[CI025, CI026, CI027, CI028, CI029, CI030]

4.4 未披露财务与承销结论

最大分析约束不是需求发现,而是披露质量。公开来源足以说明 Windsurf 已具备战略价值、达到有意义 ARR,并发展出真实企业业务。但这些来源不足以告诉投资者,独立基础上的业务到底有多持久、利润有多好。公开记录缺失的是经典承销指标:GAAP 收入、递延收入、净留存和总留存、分细分流失率、CAC、回本周期、ACV 分布、合同期限、按方案或模型家族拆分的毛利率,以及收购后迁移留存。即便 Cognition 收购后的融资数据,也存在轮次规模互相冲突的报道。 这种缺口结构导向谨慎结论。收入质量真实但并不完美:业务显然找到了产品市场匹配和买方紧迫感,但公开记录无法拆开自助与企业结构,也无法告诉我们 $82M ARR 有多少穿越了模型访问和所有权冲击。利润率路径在设计上改善,但没有数字证明:从 flow-action 额度转向 prompt 额度,再转向配额和 API 标价超额用量,正是一家公司试图停止低价出售昂贵智能体会话时会做的事,但这些步骤都没有证明贡献利润率为正。资本充足性现在已不再是独立层面的主要问题,Cognition 内部的整合治理和留存才是。 总体看,Windsurf 应被视为一个战略重要的开发者工具资产,其需求信号强到足以吸引多个买方,但独立财务报表仍过于不完整,无法支撑高信心独立承销。正确的尽调下一步是拿到私有数据,而不是继续做公开外推。[CI035, CI036, CI037, CI038, CI039, CI040]

公开财务缺口表
缺失指标对承销判断的影响严重性使用的最佳公开代理指标精确尽调路径
自助服务与企业收入拆分妨碍清楚判断 ARR 质量和渠道组合$82M ARR + 350+ 企业客户 + 公开定价阶梯索取按细分、队列和套餐拆分的收入
GAAP 收入 / 递延收入ARR 可能高估近期确认收入仅 ARR索取经审计财务报表或月度结账包
按套餐 / 模型拆分毛利率卡住单位经济性承销Sacra 负毛利表述 + OpenAI 定价 + GitLab 可比索取按档位和模型路由拆分的毛利率瀑布
CAC / 回本周期无法检验 PLG 效率或企业销售生产率仅有 PLG 动作和客户案例索取全口径 CAC、回本周期和销售漏斗转化
NRR / 流失无法衡量被收购 ARR 基础的持久性没有稳健公开代理指标索取按细分拆分的总留存率、NRR 和流失率
ACV / 合同期限 / 部署组合企业收入质量和实施负担未知350+ 企业客户 + 定价档位索取按 ACV、合同期限、托管与混合部署拆分的客户分层
收购后留存和迁移流失无法知道模型访问中断和所有权变更后有多少 ARR 留存继任界面的用户数和企业数索取客户迁移队列和按月留存 ARR
独立公司现金 / 烧钱 / 债务妨碍清楚分析出售前现金跑道仅有融资历史索取出售前即时资产负债表、债务明细和现金流量表

本表列出承销最关键的私有指标,以及补齐每个缺口需要的具体文件或分析。

[CI019, CI034, CI035, CI036, CI039, CI042]

4.5 图表

Chapter 05

05产品与技术

5.1 产品连续性与当前界面范围

现在应把 Windsurf 看作连续性产品,而不是停摆产品。2026 年 6 月转型把 Windsurf 改名为 Devin Desktop,但 FAQ 明确说它在更大的 Devin 品牌伞下仍是同一个 IDE、同一个编辑器、同一组核心功能。这种连续性很重要,因为它保住了既有 Windsurf 用户工作流——扩展、快捷键、LSP 式编辑行为、本地规则和 JetBrains 插件支持——同时把界面重新居中到智能体编排。首页和 Agent Command Center 文档把定位讲得很清楚:产品不再只是 AI 辅助编辑器,而是本地和云端智能体的命令界面,用 Spaces 把会话、PR、文件和任务上下文放在一起。历史背景也强化了这个判断。Codeium 已在 2025 年把自己改名为 Windsurf,Cognition 后来的收购文章仍把 Windsurf 称作智能体 IDE。因此,收购后的动作是加法:Cognition 保留 IDE,并把它改造成更大 Devin 产品家族的桌面前端,而不是直接替换。[CE001, CE002, CE003, CE004, CE017, CE034]

产品模块 / 资产矩阵
模块 / 资产主要用户当前状态 / 成熟度差异化依赖 / 治理尽调缺口
Devin Desktop 外壳个人开发者 + 团队开发者正式可用(GA);2026-06-02 从 Windsurf 更名独立 IDE,如今是智能体控制界面,不只是聊天叠层依赖桌面端继续铺开,以及管理员接受更名需要当前桌面端与插件端活跃用户拆分
Cascade / Devin Local编辑器里的日常开发者正式可用(GA),但正在纳入 Devin Local 品牌本地智能体让编辑循环保持快速,也保留近似离线的工作流依赖本机资源、模型访问和策略限制需要按任务类别拆分的公开成功率数据
Desktop 中的 Cloud Devin转交长任务的重度用户和团队正在自助服务套餐中铺开本地规划后,可一键交给云端 VM 执行消耗共享配额,并要求连接身份与仓库访问需要从桌面会话转云端会话的衔接率数据
Agent Command Center + Spaces管理多个会话的用户Windsurf 2.0 / Devin Desktop 中的新主界面用统一 Kanban 视图管理本地和云端智能体,并共享上下文依赖用户采用多智能体工作流,而不只是单次聊天需要 Spaces 周活使用的公开证据
ACP / 第三方智能体高级用户和企业团队已有文档;Pro/Max/Teams 可用允许 Desktop 承载 Codex、Claude Agent、Junie、Gemini CLI 和自定义智能体需要按团队治理,并预装智能体二进制文件需要第三方智能体使用的采用率数据和定价组合
上下文栈:Fast Context、DeepWiki、Supercomplete所有需要基于代码库工作的用户成熟的核心功能集在一个 IDE 里拼合 RAG 检索、符号解释和感知编辑的补全质量取决于索引、提示路由和仓库规模需要针对超大仓库的独立准确率 / 幻觉基准

当前公开模块图由现行桌面端、文档和管理界面综合而来;缺口是尽调需求,不是推断出来的缺席。

[CE001, CE003, CE004, CE005, CE006, CE012]
FE001: 产品架构图谱

将 Devin Desktop 作为延续型 IDE 的分层视图;它现在叠在本地智能体、云端委托、上下文系统以及开放智能体 / 工具协议上。

[CE001, CE003, CE004, CE005, CE018, CE021]

5.2 Cascade 工作流、上下文栈与本地到云端交接

核心本地工作流仍是 Cascade,它不只是侧边栏聊天机器人。文档描述了分开的 Code 和 Chat 模式、后台规划智能体、长任务 Todo 列表、每条 prompt 最多 20 次工具调用、排队消息、感知 linter 的编辑,以及对用户已在编辑器中完成工作的实时感知。在这个本地循环之上,Devin Desktop 保留了一套差异化上下文栈:Tab 由内部模型驱动并针对流畅度优化,Supercomplete 不只建议下一 token 文本,还能建议删除和新增;Fast Context 加基于 RAG 的上下文引擎,旨在为任务找出确切文件和行。DeepWiki 增加符号级解释,并可直接提升到 Cascade。云端交接是另一个重要架构元素。Devin in Devin Desktop 让用户先在本地勾勒或调查,再把工作发送到云端 Devin VM;VM 会独立持续运行,并在同一工作区返回可审查输出。这种本地到云端连续性,正是 Windsurf 想要占住的产品行为。[CE005, CE006, CE007, CE008, CE009, CE010]

工作流 / 用例表
用户任务当前工作流公司方案可衡量收益 / 信号局限
在本地实现或重构代码在 Cascade Code 模式中提示,借助工具和 linter 修复迭代Cascade 规划任务、编辑文件,并可自动修复 lint 问题设计目标是让开发者不离开编辑器也能保持心流没有按工作流披露的公开通过率或错误率
浏览陌生代码库使用 Fast Context、DeepWiki 和 Chat / @-mentionsRAG 索引 + 符号级解释 + 固定上下文公开文档强调精确文件 / 行检索和悬停解释质量主张仍来自公司披露,而非独立基准测试
把更大的任务升级到云端先在本地规划,再一键交给云端 Devin笔记本合上后,云端 VM 继续运行,并在同一工作区返回可审阅产出共享配额让 Desktop、CLI 和云端用量支撑同一条工作流计量定价让真实单位经济性很难从公开资料审计
安全修复或失败检查修复从 GitHub、Jira、Slack、扫描器或 webhook 触发Devin 调查、打补丁、重跑测试 / 扫描器,并打开 PR官方页面称,指定扫描器和 Itaú 积压任务约 70% 可自动解决结果数据来自公司口径,且集中在精选案例
对比插件优先的助手留在独立 IDE 中,或通过 ACP 承载外部智能体专用外壳同时服务本地和云端智能体,而不是只有扩展 UIArtificial Analysis 仍把 Windsurf 作为独立 IDE 区分竞争对手在云端智能体、MCP 和企业控制上越来越接近

工作流行混合了官方流程描述与选择性公开结果案例;缺少广泛遥测本身就是尽调点。

[CE005, CE007, CE008, CE009, CE010, CE014]
FE002: 客户工作流 / 运营流程

典型 Windsurf 到 Devin 工作流:先在本地启动,用代码上下文扎根,必要时升级到云端 Devin,再通过既有控制完成审查和合并。

[CE005, CE006, CE008, CE009, CE014, CE021]

5.3 开放智能体托管、集成与商业化机制

Cognition 正试图让桌面外壳比任何单一模型选择更耐久。ACP 让编辑器成为智能体宿主,而不只是第一方助手界面:文档把 ACP 与 LSP 类比,并明确展示 Codex CLI、Claude Agent、Junie、Gemini CLI 和 OpenCode 作为兼容例子。MCP 随后扩展工具层,把 Devin 连接到监控、数据库、文档系统和其他外部服务;公开 API 则把组织级会话操作与企业级分析、审计、计费和基础设施控制拆开。定价变化与集成模型一样大。2026 年 3 月,Windsurf 自助业务从更早的额度或 ACU 框架转向配额加按需用量;当前公开方案栈为 Free、$20 Pro、$200 Max、Teams($80 最低费用加 $40 完整席位)和定制 Enterprise。重要的是,这些配额横跨 Desktop、CLI 和云端 Devin。它比早期 Devin 时代定位降低了入门价格,但也意味着产品已变成一个按量计费的多界面套装,其经济性取决于有多少工作从本地智能体转向云执行。[CE018, CE019, CE020, CE021, CE022, CE023]

技术 / 运营架构表
层 / 流程角色关键依赖公开证据主要风险
桌面端外壳承载编辑器、标签页、终端、设置和智能体界面桌面客户端铺开 + 二进制文件更名官网和 FAQ 用新品牌描述同一款 IDE如果白名单过期,企业设备管理可能拦截更名后的二进制文件
本地智能体框架运行带规划和工具调用的 Cascade 或 Devin Local用户本机资源和所选模型Cascade 文档描述 Code/Chat 模式、工具调用和实时感知本地执行质量会随仓库规模和机器约束而波动
上下文引擎索引代码、检索相关片段,并为聊天提供事实锚定仓库索引和检索质量上下文文档描述 RAG、远程仓库索引和 M-Query 检索没有公开的独立大仓库相关性基准
开放智能体层ACP 让 Desktop 从同一 UI 启动第三方智能体预装外部智能体二进制文件和注册表配置ACP 文档把协议与 LSP 对比,并列出示例智能体第三方计费、隐私和支持不在 Cognition 条款内
工具与数据层原生集成 + MCP 连接源代码管理、PM、通信和基础设施GitHub/Jira/Slack 账户、MCP 服务器治理集成文档展示原生 + MCP 的拆分MCP 可能创建内置安全监控之外的外部资源
云端执行与管理平面Devin VM、组织 / 企业 API、审计、用量和策略控制身份提供商、API key、VPC 或企业部署API 文档和管理员指南披露服务用户、审计日志、模型与终端策略没有面向客户公开的可用时间或 SLA 承诺

架构层来自当前文档,而非单一官方参考图;风险反映明确提示或尚未填补的公开披露。

[CE003, CE005, CE009, CE015, CE018, CE020]
路线图 / 发布 / 开发阶段表
日期 / 阶段功能 / 里程碑状态含义来源
2025-04-04Codeium / Exafunction 更名为 Windsurf已完成的历史里程碑显示产品在并入 Cognition 统一品牌前,已经扛过一次重大品牌更换SE022
2025-07-14Cognition 签署收购 Windsurf 的最终协议已完成的历史里程碑确认 IDE 纳入更广 Devin 平台战略SE023
2026-03自助按量付费计划取代此前自助结构已完成 / 当前定价从旧积分框架转向配额加超额用量SE016 + SE012
2026-06-02Windsurf 成为 Devin Desktop当前已完成里程碑品牌统一保留 IDE,同时抬升 Agent Command Center 的地位SE002 + SE001
2026-06 当前ACP 第三方智能体支持已面向付费用户写入文档当前活跃能力Desktop 定位成外部智能体的宿主壳层,不只承载 Cognition 智能体SE009
2026-06 当前自适应模型路由按固定 token 定价运行当前活跃能力表明 Cognition 在优化支出效率,不只是加卖前沿模型SE017

路线图行强调外部可见里程碑;未发布的内部路线图项目被有意排除。

[CE001, CE024, CE027, CE028, CE034, CE035]
FE003: 关键依赖图谱

依赖图显示 Devin Desktop 哪些环节依赖 Cognition 控制平面、第三方模型或智能体以及外部工具连接,而不是纯粹拥有自研技术栈。

[CE018, CE020, CE021, CE022, CE023, CE025]

5.4 安全、合规与企业治理姿态

公开信任姿态比早期 Windsurf 品牌本身显示的要强很多,但仍部分是平台级而非产品特定。企业营销称客户数据留在客户控制环境中,可运行在 VPC 内,且不会用于训练。安全文档补充了传输中和静态加密、SOC 2 Type II 覆盖,以及企业承诺永不使用客户数据训练。当前安全工作流页面还描述了隔离的临时沙箱、短期凭证、密钥脱敏、域名白名单、人工控制合并,以及可导出到 SIEM 的会话记录。面向企业推广,管理指南增加了模型访问、终端自动运行等级、MCP 启用和其他团队设置的治理杠杆。主要细节在于,开放性也制造了新的治理工作。Cognition 自己的管理指南警告,MCP 可能在 Devin Desktop 安全监控之外创建基础设施资源,因此企业仍需要围绕外部服务器和第三方 ACP 智能体设置白名单与审批流程。公开文档在控制和流程上不错,但对客户可见可靠性指标和服务级别保证仍然偏薄。[CE029, CE030, CE031, CE032, CE033, CE046]

信任 / 质量 / 合规表
控制 / 认证状态范围来源支持的细节缺口
SOC 2 Type II公开表述Cognition 平台层面安全文档称 Cognition 于 2024 年 3 月取得 SOC 2 Type II未公开拆分 Desktop 专属审计范围
传输和静态加密公开表述平台层面安全文档称所有数据传输均在传输中和静态状态下加密未公开密钥管理架构细节
不用企业客户数据训练公开表述企业合同 / 企业使用文档称企业客户数据永远不会用于训练自助默认设置 / 选择加入行为仍需政策审查
VPC 部署公开营销企业部署企业页面称 Devin 可部署在客户横跨主要云的 VPC 中公开文档未展示实施引用或数据驻留认证
身份与访问控制公开文档企业与团队管理SSO、SCIM、RBAC、服务用户和最小权限指导见于安全与管理文档需要仅 Desktop 管理员与平台管理员的字段级权限矩阵
可审计性公开文档企业安全与合规工作流文档说明了可导出至 SIEM 的转录、审计日志和按会话归因未公开覆盖所有日志类型的留存期矩阵
MCP 治理公开文档,但有保留启用外部服务器的企业团队管理员指南支持团队 MCP 控制,并提醒外部资源不在监控范围内需要客户许可名单和供应商审查流程
可靠性 / SLA未公开披露面向客户的服务承诺文档展示了使用与审计仪表盘,但没有正常运行时间承诺需要 SLA、事故历史和支持严重级别响应承诺

状态反映截至 2026-06-04 的公开文档;私有企业合同的承诺可能强于外部可审计内容。

[CE029, CE030, CE031, CE032, CE033, CE046]

5.5 相对插件优先工具的差异化与剩余产品风险

Windsurf 现在最清晰的差异化是编排,而不是独家模型护城河。独立市场地图仍按部署形态把它与插件优先工具分开:Artificial Analysis 把 Windsurf 标为独立 IDE,GitHub Copilot 是扩展加云层,Claude Code 是本地或终端中心智能体并带 IDE 集成,Devin 本身则是纯云端。这确实形成了真实产品楔子。Windsurf 可以保留专用 IDE 外壳,在其上叠加本地智能体行为,再把工作升级到云端 Devin,同时不要求用户离开同一个工作区。ACP 让第三方智能体出现在同一命令中心,也强化了这个位置;这在战略上优于把产品押在单一自研模型家族上。问题是竞争场移动很快。GitHub 和 Cursor 现在都打包云端智能体、模型目录、MCP 式连接和企业控制,基准测试领先地位也早已远远越过 Devin 2024 年 SWE-bench 发布成绩。独立批评还指出 Cognition 设计中的取舍:云沙箱改善隔离,但可能增加集成摩擦,并让单任务经济性更不透明。尽调上,这意味着 Windsurf 应被按 Cognition 技术栈里的控制平面资产来承销,而不是按无可撼动的独立编程模型领导者来承销。[CE036, CE037, CE038, CE039, CE040, CE041]

FE004: 产品成熟度 / 能力图谱

定性成熟度矩阵显示,Windsurf 目前最强的位置是跨本地与云端界面的工作流编排;基准领先和透明可靠性证明弱于竞争对手营销所暗示的水平。

序数成熟度评估综合了当前文档、公开定价、基准历史和第三方市场地图。

[CE024, CE026, CE037, CE038, CE039, CE040]
Chapter 06

06客户情况

6.1 买方 / 用户 / 付款方分层

Windsurf 的客户地图最适合理解为三层买方 / 用户 / 付款方栈,而不是单一细分。历史上,Codeium / Windsurf 使用经典产品驱动漏斗:免费个人开发者先进入,团队负责人或工程经理在代码库内扩展使用,企业买方只有在身份、治理和部署控制变重要后才介入。Sacra 明确把这一动作描述为自下而上采用,把有机开发者使用转化为团队级、再到企业级部署。当前继任定价页仍反映同一阶梯:Free、Pro、Max、Team 和需联系销售的 Enterprise。这意味着主要用户仍是软件开发者,日常内部支持者通常是工程负责人或平台负责人;一旦 SSO、VPC 和管理控制进入讨论,经济买方就变成集中工程、IT 或采购预算负责人。 细分结构比泛泛的「开发者」更宽。历史 Windsurf 证明点集中在处理大型复杂代码库的企业开发团队,Sacra 点名 Dell、Zillow 和 Anduril 为公开客户,Anthropic 也描述了多年企业代码库工作。当前继任界面进一步拓宽了这一组合:Ramp 和 Bilt 代表金融科技 / 平台工程团队,Litera 代表法律科技 QA 与发布工程,Hamming AI 代表基础设施团队,AngelList 代表数据工程,Evinova 代表受监管临床软件团队。Gumroad 和 Linktree 也显示,一旦产品嵌入,产品、支持、营销等非工程职能也可以间接或直接使用智能体。这扩大了用户宽度,但也让预算归属更复杂,因为工具可以从工程起步,随后触达更广运营团队。[CU001, CU002, CU004, CU011, CU015, CU040]

客户分群表
细分买方主要用户付款方代表性用例证据 / 时期
个人开发者 / 爱好者自助个人用户开发者个人信用卡或免费层代码补全、轻量级智能体使用、实验历史与当前均有;漏斗顶部证据强,付费转化披露弱
创业公司 / SMB 工程小组工程经理或创始人小型开发团队团队负责人 / 创业公司预算共享代码库工作、原型开发、bug 修复当前继任产品定价与案例研究证据
大型企业平台团队工程副总裁 / 平台负责人开发者、SRE、QA、数据工程师中央工程 / IT 采购现代化改造、测试、技术债削减、SDLC 自动化历史 Windsurf 以及当前继任产品案例
受监管企业CTO / 重视合规的工程管理层开发者、QA、产品、安全审核人员企业采购 / 风险委员会可审计的金融服务和生命科学交付当前继任产品证据最强(Itaú、Evinova)
服务商 / 多代码库交付组织业务负责人 / 交付高管实施工程师部门或客户服务预算并行现代化改造与多代码库迁移工作当前继任产品证据强于历史 Windsurf
跨职能邻近岗位借工程团队部署使用的产品、支持或营销负责人产品、支持、营销和运营人员通常仍由工程团队预算承担bug 分诊、内容修改、文档、轻量级自动化仅当前继任产品;历史 Windsurf 未验证

各行区分谁发起使用、谁最终付款。历史 Windsurf 的证据在开发者主导的企业采用上最强;跨职能使用主要是继任产品时期现象。

[CU001, CU002, CU011, CU015, CU017, CU040]
FU001: 客户旅程图

公开记录支撑的旅程是:从开发者主导的试用开始,经团队和安全控制正式化,再扩展到更广的工程工作流。

该图综合公开定价、合作伙伴评论和客户案例;它描述公开材料中的主导路径,而不是每个账号必须遵循的单一路径。

[CU001, CU002, CU015, CU040, CU041, CU044]

6.2 采用轨迹与历史 / 继任边界

公开采用指标必须拆成三个阶段看。第一阶段是早期上线证据,Anthropic 的客户故事显示出异常快的产品驱动增长:Windsurf 两天拿到 10,000 名用户,第二周结束时已达数十万,管理层称日留存非常高。Anthropic 页面还写明有 800,000+ 活跃用户和 1,000+ 企业客户;Sacra 也独立称,到 2025 年初 Codeium 已扩展到 800,000+ 活跃开发者和 1,000+ 家企业。第二阶段是收购时点的口径明显更窄:Cognition 2025 年 7 月的收购公告和 TechCrunch 都说,Windsurf 带来 350+ 企业客户、数十万日活用户、$82M 年经常性收入(ARR),且企业 ARR 环比翻倍。由此看,历史上的「1,000+ 企业」口径与后来「350+ 企业客户」口径并不是可直接比较的分母。 第三阶段是今天的继任产品界面,明确已经是交易后的口径。当前 Windsurf/Devin Desktop 首页声称有 1M+ 开发者和 4,000+ 企业客户,Cognition 后续融资文章还说,老 Windsurf 与收购前 Devin 的企业客户重叠率低于 5%,合并后的企业 ARR 在交割后前七周增长超过 30%。这些都是重要商业信号,但不能倒填进历史 Windsurf。最干净的解读是:交易前的历史 Windsurf 已证明很强的 PLG 采用能力;更大的 1M+/4,000+ 覆盖面,则是当前合并平台的营销口径。[CU003, CU004, CU005, CU006, CU007, CU008]

客户增长 / 采用轨迹表
指标公开数值日期 / 时期来源质量置信度含义缺失分母
发布两天内用户数10,000发布期(历史)合作伙伴客户故事显示即时 PLG 拉力未拆分付费 / 免费
第二周结束时用户数数十万发布期(历史)合作伙伴客户故事确认自助采用速度异常快未拆分 WAU / MAU
活跃用户基数800,000+ 活跃用户 / 开发者2025 年初合作伙伴证明 + 分析师收购前历史规模上限不清楚是仅免费、月活还是日活
企业客户 / 企业1,000+ 企业客户 / 1,000+ 企业2025 年初合作伙伴证明 + 分析师表明收购前企业触达较广企业账户定义不清
收购快照:企业客户基础350+ 企业客户2025 年 7 月官方 + 主要媒体交易完成时披露最清晰的客户数可能排除此前计入的小型企业
收购快照:使用量数十万日活跃用户2025 年 7 月官方 + 主要媒体显示收购时有真实活跃使用未披露精确 DAU
收购快照:ARR$82M ARR;企业 ARR 环比翻倍2025 年 7 月官方 + 媒体 + 分析师确认商业牵引未拆分客户集中度
交叉销售重叠交易完成前 Devin 与 Windsurf 企业客户重叠 <5%2025 年交易完成后官方 + 两家新闻源交叉销售空白大未披露重叠计算方法
合并企业 ARR 增长交易完成后七周 >30%2025 年交易完成后官方 + 两家新闻源即时商业协同强未拆分价格与客户增长
当前继任产品口径1M+ 开发者和 4,000+ 企业客户当前 2026 口径官方首页当前平台覆盖面大当前合并口径,并非历史 Windsurf 单独口径

本表有意拆分历史 Windsurf 指标与当前继任产品口径指标,避免把 2026 年合并平台数字误读为历史 Windsurf 的独立证据。

[CU003, CU004, CU005, CU006, CU007, CU008]
历史 Windsurf 与当前继任平台证据边界表
产品界面 / 时期公开说法或证据数值 / 标杆客户证据质量未证明什么结论
发布期合作伙伴故事早期 PLG 采用2 天内 10,000 用户;第二周末达到数十万未显示付费账户或企业客户结构有力证明即时自助拉力
2025 年初的历史 Windsurf历史规模800,000+ 活跃用户 / 开发者;1,000+ 企业客户 / 企业各来源对分母的定义不一致历史采用上限信号
2025 年 7 月收购快照交易完成时规模350+ 企业客户;数十万 DAU;$82M ARR不反映交易完成后的合并平台最干净的历史交易快照
历史具名标杆客户具名客户Zillow、Dell、Anduril低-中不证明生产深度、席位数或续约历史证据偏标杆客户展示
交易完成后合并平台经济性交叉销售释放客户重叠率 <5%;七周内企业 ARR 增长 >30%无法隔离原 Windsurf 单独增长商业协同看起来真实
当前继任平台营销界面当前覆盖面1M+ 开发者;4,000+ 企业客户当前合并平台,不是历史 Windsurf 单体只作为继任平台证据使用
当前量化案例研究层详细证据客户案例:Nubank、Ramp、Litera、Gumroad、Hamming、AngelList、Evinova不一定代表完整客户群合并后证据深度明显提升

本表是本章的边界图:把历史 Windsurf 证据与当前继任平台界面证据分开,避免跨时期重复计算采用规模。

[CU003, CU004, CU005, CU006, CU007, CU012]
FU002: 采用 / 部署流程

公开证据显示路径从开发者自助采用走向企业标准化;交易完成后的 Devin 组合包扩大了可触达账号集合。

该流程结合历史 Windsurf 证据和交易完成后 Cognition 披露;节点展示公开商业顺序,不是私人 CRM 漏斗。

[CU003, CU006, CU013, CU015, CU040, CU041]

6.3 具名客户证明质量

具名客户证明最能看出历史 Windsurf 与当前 Cognition/Devin 继任产品之间的证据质量差距。历史 Windsurf 的公开证明大多停留在 logo 层面:Sacra 点名 Zillow、Dell 和 Anduril 是客户,但公开材料没有说明这些账户的部署阶段、团队规模或量化结果。相比之下,当前继任产品界面发布了生产环境案例研究。Nubank 是最强例子:Devin 参与了一项 6M+ 行 ETL 迁移,触及约 100,000 个数据类,工程工时效率提升 12x、成本节省超过 20x,并帮助 Data、Collections 和 Risk 业务单元在数周内完成迁移。Ramp 报告每周最多合并 80 个 PR,每月节省 10,000+ 小时;Litera 报告测试覆盖率提高 40%、回归周期加快 93%、30 天内 90+ 用户上线、三周发布五个产品;Gumroad 称 Devin 四个月合并 1,583 个 PR,合并率 >85%;Hamming 称 Devin 贡献了总代码量的 25%;AngelList 借助 20 个并行智能体,将 Redshift 到 Snowflake 的迁移速度提高 5.2x;Evinova 则给出受监管医疗场景证明,GxP 文档速度提升 8x,前 10 天自主修复 66% 的 bug。 但这不代表当前所有具名 logo 都同样有力。Cognition 融资文章及后续报道列出 Goldman Sachs、Citi、Cisco、Palantir、Santander、U.S. Army 和 U.S. Navy,但这些是客户名单披露,不是已发布的结果型案例研究。因此,投资人应把证明分成三层:历史 Windsurf 仅有 logo、当前合并后的 logo 名单、当前已量化的案例研究。[CU011, CU014, CU018, CU019, CU020, CU021]

具名客户验证表
客户细分部署 / 用例生产 vs 试点结果限制
Zillow(历史 Windsurf)消费互联网 / 房地产科技公开列名为客户未知 / 仅 logo无公开量化结果仅历史 logo 提及;无公开部署细节
Dell(历史 Windsurf)企业 IT公开列名为客户未知 / 仅 logo无公开量化结果仅历史 logo 提及;无公开部署细节
Anduril(历史 Windsurf)国防科技公开列名为客户未知 / 仅 logo无公开量化结果仅历史 logo 提及;无公开部署细节
Nubank金融科技银行跨业务单元的大规模 ETL 迁移与重构生产效率提升 12x,成本节省 >20x,周期从数月 / 数年缩至数周公司发布案例研究;无独立复盘
Ramp金融科技 / 支出管理技术债清理与内部开发者工具生产每周 80 个 PR,每月节省 10,000+ 小时公司发布案例研究;经济性未经审计
Litera法律科技QA / 测试智能体与发布提速生产测试覆盖率增加 40%,回归周期加快 93%,30 天内 90+ 用户公司发布案例研究
Gumroad创作者商业跨职能 PR 生成、支持分诊、营销修改生产4 个月合并 1,583 个 PR,合并率 >85%公司发布案例研究
Hamming语音 AI QA智能体主导开发与客户问题解决生产总代码量 25% 来自 Devin公司发布案例研究
AngelList金融科技 / 数据工程Redshift 到 Snowflake 迁移生产速度快 5.2x;20 个智能体在 3 周迁移 14,000 张卡片公司发布案例研究
Evinova生命科学 / 受监管软件GxP 文档、bug 修复、迁移、测试自动化生产GxP 文档速度快 8x;前 10 天 66% bug 自主修复公司发布案例研究

历史 Windsurf 行有意偏重 logo,因为公开部署细节稀缺。当前继任产品行更强,因为 Devin 官方网站现在发布了详细案例研究。

[CU011, CU014, CU018, CU019, CU022, CU024]
FU003: 客户证明矩阵

最强的公开证明现在来自继任者时代的量化案例研究;历史 Windsurf 证据薄得多,更偏 logo 展示。

评分为序数,1 表示公开证明弱,5 表示公开证明量化且强。该矩阵比较证明层级,不比较客户价值。

[CU004, CU006, CU014, CU037, CU045]

6.4 耐久性、扩张、集中度与采购摩擦

耐久性、扩张和采购风险仍是客户故事中最缺证明的部分。无论历史 Windsurf 还是当前继任产品界面,都没有披露 NRR、GRR、logo 续约率、头部客户集中度,或免费用户转化为团队及企业合同的情况。公开材料里最好的耐久性代理指标不是财务指标,而是运营指标:Anthropic 提到上线后日留存非常高,Gumroad 从单个小组扩展到全公司使用,Bilt 在 106 名工程师中铺开 Devin 且超过一半每周使用,Litera 在 30 天内将 Devin 活跃使用量提高一倍以上。这些使用信号令人鼓舞,但不能替代付费续约数据。 反向证据也真实存在。存档的 G2 覆盖评论深度太浅,无法提供购买洞察;Trustpilot 评论则记录了围绕定价调整、退款争议、额度耗尽以及长期用户称 2026 年 3 月大改后取消订阅的强烈反弹。TechCrunch 还报道,Anthropic 2025 年 6 月切断 Claude 直接访问后,部分 Windsurf 客户转向竞争对手。在采购层面,今天的企业版报价强于历史公开记录,因为它宣传 VPC 部署、SOC 2 Type II、teamspace 隔离以及 SSO/管理员控制。但合并后的实体仍需协调 Windsurf 的按席位定价与 Devin 基于使用量的算力经济、按 API 定价的超额用量。这会带来预算和续约摩擦,而集中度风险恰恰最难被观察。[CU036, CU037, CU038, CU039, CU041, CU042]

留存 / 重复使用 / 满意度表
指标公开数值细分置信度尽调请求
发布后日留存“非常高”的日留存,未给数字曲线历史自助用户按发布队列索取 30/90/180 天留存用户曲线
独立评论深度G2 评论不足,无法提供采购洞察独立评论渠道索取原始 G2 / Gartner / 客户推荐导出数据
定价反噬信号Trustpilot 出现取消、退款纠纷和额度耗尽投诉自助与团队账户衡量 2026 年 3 月定价大改前后的 logo 流失
公开 NRR未披露企业Unknown按 ACV 档位、历史 Windsurf 与合并平台队列索取 NRR
公开 GRR / logo 流失未披露企业Unknown按季度索取年度总留存率和 logo 流失
Gumroad 运营耐久性4 个月 1,583 个 PR,合并率 >85%当前继任产品生产账户确认初始上线后高合并质量能否持续
Bilt 重复使用106 名工程师中超过半数每周使用 Devin当前继任产品企业账户索取每周活跃席位趋势和付费席位利用率
Litera 扩张代理指标30 天内 90+ 活跃用户,且每日增长当前继任平台企业账户索取 6 个月和 12 个月席位扩张及续约数据

因公开材料未披露 NRR、GRR 和客户续约情况,耐久性主要只能靠使用代理指标衡量。

[CU009, CU025, CU027, CU031, CU036, CU037]
扩张与集中度风险表
驱动因素 / 风险类型描述影响尽调路径
从免费到团队再到企业的阶梯扩张驱动因素历史与当前证据都指向一条 PLG 路径:开发者先免费使用;当控制能力和共享工作流变得重要,账户再向上扩张。按套餐层级获取队列转化和扩张数据
交易完成时客户重叠率 <5%扩张驱动因素原 Devin 与 Windsurf 客户重叠度低,在 IDE 和自主智能体两个界面留下交叉销售空白。按原有客户群索取交叉销售管线、赢单率和附加率
历史集中度不透明集中度风险$82M ARR 之外,公开可查的历史部署案例很薄,仍可能由少数大客户贡献大部分收入。索取头部客户收入占比和 ACV 区间
定价复杂度和超额费用采购摩擦当前套餐混合席位费、基础费和按 API 定价的超额费用;相比简单的按席位订阅,预算更难做。索取使用量分布、超额费用发生率和采购异议
整合驱动的套餐变更留存 / 采购风险第三方报道说,Cognition 只打算在整合前的初始过渡窗口保留当前支持和定价。索取交易完成后的定价路线图和老客户保留政策
模型访问依赖切换风险TechCrunch 报道,Windsurf 暂时失去 Claude 直接访问权后,部分客户转向其他产品。复核供应商集中度、备用模型,以及中断期间的客户流失
历史证据深度与估值不匹配证据质量风险历史 Windsurf 标杆客户,不能像当前继任平台的量化案例研究那样证明生产环境确定性。索取收购前客户访谈和部署阶段证据
当前企业安全控制采购缓释因素当前企业版材料展示 SSO、VPC、SOC 2 Type II 和管理员控制,能压低部分买方异议。中等正向验证控制措施、审计范围,以及受监管垂直行业客户背书

同一套商业设计带来扩张上行空间,也让集中度、预算可预测性和续约质量变得不透明。

[CU038, CU039, CU040, CU041, CU042, CU043]
Chapter 07

07风险

7.1 所有权过渡与控制风险

Windsurf 最高严重度的风险在于,公司已不再是一个干净、由创始人领导的运营单元。Google 挖走了 CEO、联合创始人和关键研究负责人,Cognition 则签约接手剩余产品、品牌、IP 和客户基础。资产最需要连续性的时刻,核心人才所有权与运营业务所有权发生了分离。剩余业务并不小:Cognition 披露约 $82 million ARR、超过 350 家企业客户,以及数十万日活用户。换言之,整合失败不只是拖慢上行空间;它可能损害一个庞大的安装基础和真实收入流。公开缓释措施只是不完整的。Cognition 承诺尊重式处理和即时连续性,但也表示会大举投资,把 Windsurf 折入自己的技术栈。随后关于裁员、买断和更强硬工作文化的负面报道,提高了产品、支持和商业化连续性在客户看到承诺产品增益前就恶化的概率。投资含义很直接:这是一个影响关键的风险,缓释成熟度只有部分到位;尽调应把留存和客户连续性指标视为一阶闸门,而不是背景性 HR 噪音。[CR001, CR002, CR003, CR004, CR005, CR006]

人员 / 执行风险登记表
角色 / 职能依赖或缺口可能性严重性缓释措施尽调路径
原 Windsurf 工程和产品负责人文化不匹配、裁员和买断,可能在整合期间打断知识转移严重Cognition 称所有员工都获得经济参与,且已规划即时连续性索取按职能划分的流失情况、留任负责人和知识转移里程碑
客户成功和企业支持迁移期间任何扰动,都可能先反映在 350+ 企业客户的续约和支持质量上大规模存量客户和当前企业工具仍在运行复核头部账户支持覆盖、工单 SLA 趋势和续约日历
安全和合规运营对受监管部署就绪度,没有公开负责人、人员配置计划或详细保证包当前安全材料记录了工作流级控制索取明确的合规负责人、保证日历和信任中心资料包
董事会和治理监督没有公开董事会名单,也没有清楚说明谁治理了拆分交易及当前监督模式公开不可见索取当前董事会名单、观察员权利和重大交易后治理契约
FinOps 和模型供应商管理尽管存在按 API 定价转嫁风险,但没有公开对冲供应商价格或配额冲击的安排多模型可用性带来战术路由灵活性索取供应商组合、毛利率桥接和合同通知期

执行风险偏向信息不对称:管理层可能已有缓释措施,但公开证据还不够强,不能按表面说法承销。

[CR003, CR005, CR006, CR007, CR008, CR041]
FR003: 依赖与控制权交接图

Windsurf 当前风险集中在 Cognition 集成、创始人语境流失和外部模型 / 平台依赖交汇处。

[CR001, CR002, CR003, CR010, CR015, CR041]

7.2 模型供应商依赖、安全与监管风险

下一层风险来自上游模型依赖与智能体安全暴露的叠加。Windsurf 自己的合作伙伴故事称,Claude 支撑 Cascade 的核心推理和规划;当前 Devin 定价则宣传可访问 OpenAI、Claude 和 Gemini 前沿模型。这有助于分散风险,但不能消除供应商权力;产品被嵌进一组不断变化的第三方定价、配额和政策制度中。Anthropic 和 OpenAI 都保留广泛的访问权利,Anthropic 自己的产品页面也说,限额和价格变更可由其酌情发生。与此同时,Devin 的集成和 MCP 支持放大了错误的影响半径:文档承认幻觉和不安全代码仍可能出现,即便营销材料强调沙箱、按工具限定范围、默认拒绝出站流量以及人工控制合并。OWASP 和 CISA 也强调,提示注入、不安全插件、过度自主性和糟糕的安全部署实践,是智能体 AI 的核心风险而不是边缘案例。监管暴露真实存在,但可管理。EU AI Act 现在适用 GPAI 透明度和版权义务,美国版权政策仍未定型。正确结论是:只有管理层能用独立证据证明企业级控制,而不只是产品文案,并且能证明收购后运营仍满足受监管买方隐含要求的标准,Windsurf 的安全和监管姿态才具备可投资性。[CR009, CR010, CR011, CR012, CR013, CR014]

监管 / 法律风险登记表
规则 / 案件司法辖区状态可能性严重性缓释措施剩余暴露尽调路径
EU AI Act GPAI 透明度和版权义务EU2025 年 8 月起适用;更多指南仍在 2026 年陆续推出发布模型和数据摘要;把供应商义务映射到产品界面获取 AI Act 合规工作计划、明确负责人和训练数据摘要流程
AI Act 高风险部署控制EU / 受监管行业取决于用例人工审核、日志记录、监督、网络安全和部署范围界定验证是否有客户工作流落入高风险或受监管客户类别
训练数据版权审查美国政策调查活跃,诉讼仍在进行记录来源、授权和排除控制索取法律顾问关于训练数据来源和许可处理的备忘录
政府 secure-by-design 预期美国公共部门指引有效让部署和事件流程对齐 secure-by-design 指引复核公共部门控制矩阵、事件响应手册和客户背书
收购后合同与政策一致性多司法辖区公开未披露统一 Windsurf 与 Devin 的 DPA、隐私条款和企业承诺复核现行企业合同中使用的最新 DPA、SCC 和事件通知语言

本登记表按严重性排序,列出 Cognition 交易后最可能影响 Windsurf 当前智能体编码业务的公开法律与监管风险向量。

[CR024, CR025, CR026, CR027, CR028]
运营 / 质量 / 安全风险登记表
失效模式可能性严重性缓释成熟度剩余暴露未解决缺口
智能体工作流中的提示注入和过度自主严重部分没有公开独立证据显示,当前控制在大规模对抗测试下能撑到什么程度
集成工作流中的 MCP 或外部工具滥用部分公开文档描述了集成,但工具审批治理和影响半径遥测大多没有文档说明
幻觉代码或不安全代码合入生产路径部分公司文档仍依赖代码审查和分支保护,而不是在源头消除模型错误
收购后界面的安全保证未披露部分合并界面看不到公开漏洞赏金、正常运行时间历史或当前独立渗透测试摘要
模型或算力冲击导致体验下降或账单冲击超额使用按 API 定价转嫁,上游供应商控制配额、条款和容量

失效模式按剩余严重性排序,而非按时间排序;缓释成熟度只反映公开文档和第三方评论中可见的内容。

[CR016, CR017, CR018, CR019, CR020, CR021]
FR001: 风险热力图——剩余可能性与影响

最高残余风险集中在控制权交接、模型供应商依赖以及安全 / 治理失效上。

[CR001, CR007, CR015, CR019, CR025, CR039]

7.3 竞争、分发与单位经济风险

竞争不只是激烈,而是结构性不对称。GitHub 可以把 Copilot cloud agents 与集中式策略控制、广泛模型访问打包,Cursor 已经在营销企业安全、零数据留存控制以及有意义的 Fortune 500 渗透。公开证据也显示,该品类的算力经济并不稳定:GitHub 2026 年 4 月暂停新的 Copilot 注册,外部报道称智能体工作负载迫使用量上限和模型降级。这一点很关键,因为 Windsurf 当前定价相对品类并不低,却仍允许客户按 API 定价购买额外用量,从而把供应商成本波动压进毛利率或客户账单。Cursor 的规模——2025 年 4 月 ARR 为 $200 million,2025 年 6 月超过 $500 million——说明竞争对手有预算同时补贴价格、企业销售和安全营销。OpenAI 与 Anthropic 更棘手,因为它们能从技术栈两端进攻:上游调整模型条款,下游推出 Codex 或 Claude Code。公开基准和对比网站让买方更容易看清赛场,也压缩了仅靠叙事维持差异化的半衰期。这里的剩余暴露很高:即便竞争对手继续更快改进、更强打包,或更聪明地配给算力,Windsurf 也必须证明自己能守住企业留存和实际 ROI。[CR029, CR030, CR031, CR032, CR033, CR034]

合作伙伴 / 依赖风险登记表
依赖对手方角色集中度失效场景严重性缓释措施剩余暴露
推理和规划模型供应商AnthropicClaude 支撑核心规划流程,也通过 Claude Code 直接竞争竞品持续改进的同时,定价、限额或访问条款收紧严重多模型支持,并已在前沿模型之间分散
模型供应商和直接工作流进入者OpenAI模型供应商,也在重叠工作流中运营 Codex政策变化、访问限制或直接产品捆绑压缩 Windsurf 灵活性多模型路由,客户可在供应商之间选择
代码仓库和企业分发平台GitHub / MicrosoftPR 工作流、代码仓库集成和 Copilot 企业分发API 或产品政策变化偏向 Copilot,或让依赖工作流更不可靠维持多界面工作流,并强调本地加云端的灵活性
云基础设施主干AWS当前生产环境和企业部署底座宕机、定价变化或更严格的 AI 服务条款扰乱服务质量隔离、出站控制和企业部署选项
历史产品知识和创始人背景被 Google 聘走的创始人 / 研究团队相当一部分路线图背景和技术叙事留在被收购运营边界之外知识分离拖慢产品整合,或削弱模型选择优势IP 转让,加上 Cognition 内留存的 GTM、工程和产品人员

当对手方本身也是直接或相邻竞争者、且有自己的编码智能体议程时,剩余暴露仍然高。

[CR002, CR010, CR011, CR012, CR013, CR014]
FR002: 风险传导图

供应商、竞争和安全冲击会传导到续约风险、利润率和估值质量。

[CR015, CR031, CR032, CR035, CR039, CR040]

7.4 缓释措施、监控指标与投资逻辑失效触发器

公开缓释措施确实存在,但成熟度不均。最清晰可见的控制在工作流层面:隔离执行、人工控制合并、分支保护、代码库和工具范围限定,以及审计轨迹。最薄弱的是交易后的运营透明度。没有公开的董事会层面说明谁治理这项分拆资产,没有披露供应商合约对冲,没有公开留存仪表盘,也没有单位经济视图告诉投资人,如果模型成本或竞争压力恶化,定价还有多少空间。因此,监控纪律必须承担主要工作。最有用的指标包括老 Windsurf 运营人员的自愿流失、企业客户留存、供应商条款变化、额外用量变现,以及任何会打破“人工审查能控制智能体风险”这一前提的安全事件。投资逻辑也应接受市场信号压力测试:如果 Copilot 或 Cursor 持续拿下企业份额,而 Windsurf 必须靠折扣留住账户,那么当前收入基础的质量可能低于账面 ARR 所暗示的水平。就尽调而言,本章的淘汰标准刻意设定得很具体:把控制权过渡、供应商、安全和竞争风险翻译成可监控事件,一旦触发就能立即重置承销假设,而不是让风险漂移躲在品类动能背后。[CR018, CR020, CR032, CR039, CR040, CR043]

缓释与终止标准表
风险可监测触发因素阈值 / 事件行动含义
所有权过渡失败原 Windsurf 人员流失、支持中断或整合滑坡连续两个季度出现重大原团队流失,或企业支持明显退化不再把过渡视为临时噪音;重新承销客户耐久性和产品路线图可信度
供应商访问或政策冲击Anthropic 或 OpenAI 的条款、限额、定价或模型目录变化暂停访问、硬性削减配额,或供应商成本上升到足以在没有抵消项时损害当前定价立即重做利润率模型,并在备用权利得到证明前下调信心
安全 / 治理失效已确认的提示注入事件、密钥泄露,或在控制存在下仍进入生产的不良合并任何公开事件表明,人工审核未能遏制重大客户工作流中的智能体失效在事件响应复盘和客户流失影响分析完成前,按投资逻辑破裂处理
竞争性分发或定价挤压大型企业竞争失单,或被迫实质降价需要大幅下调有效价格,同时 ARR 质量或续约速度转弱将基准情景下调到更低利润率、更短护城河周期;不要按高溢价倍数承销
监管市场准备度缺口企业尽调失败、发生执法事件,或无法为目标细分市场出示合规证据任何大型监管行业或公共部门机会,都因控制缺失或合规责任未厘清而受阻在拿出控制证据前,把监管市场上行空间移出基准情景

这些阈值是决策规则,不是泛泛的观察项;一旦触发,每一条都会改变承销判断。

[CR017, CR018, CR020, CR031, CR032, CR039]
Chapter 08

08估值

8.1 建议与入场框架

从公司质量看,Windsurf 显然曾打造出有价值的资产。最后一个干净基准是 2024 年 8 月 Series C,估值 $1.25B;随后 2025 年 7 月的交易过程显示出更强的战略需求:据报道 OpenAI 的 $3B 报价失效后,Google 支付约 $2.4B 挖走创始人和部分研究人员并获得技术许可,Cognition 另行签约收购剩余产品、IP、品牌和业务。但这些事实反而削弱了新的独立投资逻辑。它们证明资产有价值,并不证明外部投资人仍能买到一家权利清晰、公司形态完整的独立公司。公开证据显示,创始人主导的控制中心迁至 Google,运营剩余部分转向 Cognition,也没有披露新的独立融资。若以价格敏感视角看,正确建议是回避:可比问题不是 Windsurf 当年好不好,而是现在是否仍有一项权利可理解、入场价格可支撑的干净 Windsurf 证券。公开证据尚未达到这个门槛。[CV001, CV002, CV003, CV007, CV008, CV009]

建议摘要表
维度评估证据基础决策含义
建议回避单独进入 Windsurf战略需求确实存在,但公开证据已看不出一份干净的独立股权证券目前不要按新的风险投资仓位承销
置信度交易事实有较好交叉印证,但各利益相关方之间的经济分配仍不透明用区间和法律尽调替代单点估值承销
风险评级所有权拆分、未记录的瀑布分配,以及交易后执行风险,可能造成严重减值假设任何剩余权利的下行可接近全损
估值立场昂贵 / 无法作为干净入口合成的 29x–37x 战略参照,高于 Cursor 和多数公开开发者工具可比公司区间重新打开前,必须有大幅折扣和明确权利
重启条件可转让证券,对应穿透价值大约 $1.0B 或更低这样才能让进入价格更接近基准情景区间,并部分补偿不透明度只有到那时,才从回避转向重尽调的重新评估

决策表只反映单独的 Windsurf;它并不承销对合并后 Cognition 的另一笔投资。

[CV015, CV016, CV017, CV018, CV047, CV049]
投资逻辑 / 反向逻辑表
视角乐观情景证据反向逻辑证据改变观点的证据
资产质量82M ARR、350+ 企业客户和数十万 DAU,说明产品市场匹配真实存在拆分后客户留存没有公开披露一份交易后队列桥,证明企业 ARR 得以留存
战略需求OpenAI、Google 和 Cognition 在数日内都追逐过 Windsurf三份报价瞄准的是不同价值切片,而不是一家干净的独立公司证明新投资者可以买到同样的经济敞口
可比支撑Cursor 和合并后 Cognition 显示,完整的 AI 编码平台可以冲上约 20x 收入区间Windsurf 已不再完整,也不再由创始人主导,因此这些倍数需要打折关于控制权、权利和持久客户归属的证据
执行韧性团队大多数人没有进入 Google,保留下来的时间足以让 Cognition 买到一项仍可运转的业务文化重置报道和价格压力,抬高了留存与利润率风险经验证的留存、稳定定价,以及交易后健康的产品迭代速度

本表把公司质量和可投资性拆开看;Windsurf 前者可以得分不低,但后者仍可能不过关。

[CV003, CV005, CV006, CV007, CV009, CV013]
FV001: 建议逻辑

从资产质量和战略需求,经由所有权不透明,推导到回避建议的决策路径。

这张流程图把多项估值事实压缩成一条投委会(IC)式决策链,而不是确定性的财务模型。

[CV003, CV007, CV009, CV018, CV033, CV055]
FV004: 投资 KPI

投委会(IC)式评分卡,把强资产需求和弱独立可投资性拆开看。

评分卡是定性的投委会(IC)总结,不是机器打分模型;它有意把所有权清晰度的权重放得高于产品热度。

[CV025, CV032, CV041, CV049, CV055, CV056]

8.2 交易桥接与可比背景

交易数学进一步支持谨慎。以 Cognition 披露的 $82M ARR 作为共同分母,最后一个干净的 Series C 基准约为 15.2x ARR,据报道 OpenAI 兴趣对应约 36.6x,Google $2.4B 反向收购式挖人与许可交易约为 29.3x。Google 这个数字不应被视为正常的整家公司估值,因为 Google 既没有买下控制权,也没有取得股权;它买的是关键人才访问和非独家技术许可。Cursor 是最接近且仍完整的私有可比公司:TechCrunch 报道其 2025 年 6 月估值 $9.9B、ARR 超过 $500M,约为 19.8x ARR。按 Sacra 2026 年 5 月 ARR 估计和 CNBC 2025 年 9 月估值轮,合并后的 Cognition 约为 20.7x。公开开发者工具可比公司的倍数低得多,或至少更窄:GitLab 2026 年 6 月市值隐含约 6.9x FY2025 收入,JFrog 约为 23.7x FY2024 收入。Copilot 具有战略意义,但不是干净的股权可比,因为 GitHub 将其作为更大平台里的打包产品销售,覆盖个人、商业和企业计划。因此,估值信号很清楚:完整平台可以拿到溢价倍数,但所有权不透明的分拆资产应打折交易,而不是按最贵的战略成交价定价。[CV015, CV016, CV017, CV018, CV019, CV020]

可比估值表
可比对象指标估值 / 倍数相关性局限
Windsurf Series C 轮(2024 年 8 月)最后一轮干净的私募融资$1.25B 名义估值 / 以后续 $82M ARR 计算约 15.2x最后一个无争议的独立股权基准使用后续 ARR 分母,只作方向性比较
据报道 OpenAI 表达兴趣(2025 年 7 月)未成功的战略报价约 $3.0B / 以 $82M ARR 计算约 36.6x显示战略控制权兴趣如何给稀缺性定价报价已过期且从未成交
Google 反向收购式招聘 / 授权(2025 年 7 月)人才加非独家授权约 $2.4B / 以 $82M ARR 计算约 29.3x证明创始人和技术具有巨大战略价值不是整家公司股权收购
Cursor(2025 年 6 月)私有 AI 编码融资轮$9.9B / >$500M ARR / 约 19.8x最接近的完整私有 AI 编码可比公司收入基数大得多,控制权也完整
合并后 Cognition(2025 年 9 月 vs 2026 年 5 月 ARR 估计)收购后平台$10.2B / 约 $492M ARR 估计 / 约 20.7x显示买入 Windsurf 后,整合平台所处的倍数区间不是 Windsurf 单独价值,且使用分析师 ARR 估计
GitLab(2026 年 6 月)公开 DevSecOps 可比公司$5.22B 市值 / $759.2M FY2025 收入 / 约 6.9x规模化开发者平台的公开市场下沿增速更慢,且受公开市场纪律约束
JFrog(2026 年 6 月)公开开发者工具可比公司$10.15B 市值 / $428.5M FY2024 收入 / 约 23.7x高质量开发者工具资产的公开市场上沿不是 AI 原生,利润率画像也不同

合成的 Windsurf 倍数用交易名义价值除以 Cognition 披露的 $82M ARR,以保持比较口径一致;本表经过筛选,并非穷尽列表。

[CV015, CV016, CV017, CV018, CV019, CV021]
FV002: 估值敏感性

Windsurf 在不同基准倍数下的合成价值结果,锚定公开和私有市场参照。

柱状图是以 $M 计的合成输出,有意把口径不同的参照放在同一个 $82M ARR 分母上比较,显示标称价值会多快拉开差距。

[CV015, CV016, CV017, CV018, CV019, CV029]

8.3 合成乐观、基准与悲观情景

乐观、基准和悲观情景应被读作合成的穿透估值,而不是报价市场价格。乐观情景下,Windsurf $82M ARR 基础的大部分经济价值留存在 Cognition 买下的剩余业务中,客户关系证明有粘性,未来买方按接近 Cursor 或高端公开可比公司的倍数承销该资产。这可以支撑约 $1.6B 到 $2.2B 的价值。基准情景更保守,在我看来也更有决策价值:对已披露 ARR 适用 10x 到 15x 区间,再因创始人离开、企业客户基础留存未经验证,以及公开持有人不了解存续分配顺位而打折。由此得到约 $0.8B 到 $1.2B。悲观情景不是理论推演;如果经济价值已经通过 Google 的薪酬和许可包、以及未披露的 Cognition 购买分配实现,剩余股东只留下高度次级或实际上被清零的权利,那么这就是逻辑后果。在这一框架下,穿透价值可能下探至 $0 到 $0.4B。由于没有公开证据显示新的融资流程或干净二级证券,这组情景推演导向入场纪律,而不是热情。[CV018, CV033, CV041, CV042, CV043, CV044]

乐观 / 基准 / 悲观情景表
情景核心假设穿透价值基于 $82M ARR 的隐含倍数概率信号投资含义
乐观企业账户保持完整,战略需求维持高位,未来买方支付接近 Cursor 或公开可比公司高端的倍数$1.6B-$2.2B19x-27x只有真实可转让证券仍然存在时才值得关注
基准对 $82M ARR 采用 10x-15x,并因创始人离开、所有权不透明和留存不确定性打折$0.8B-$1.2B10x-15x这是唯一可能支撑纪律化重新进入的区间
悲观剩余持有人被劣后,客户迁移,或 2025 年两笔交易已拿走大部分经济价值$0-$0.4B0x-5x中-高即便产品仍有用,新投资者仍可能几乎损失全部资本

情景区间是合成穿透价值,不是市场报价;原因是没有公开披露新的独立 Windsurf 融资轮。

[CV048, CV049, CV050, CV055, CV056]
FV003: 估值 / 回报区间

悲观、基准、乐观和重新开放入场情景下的低位、基准、高位穿透价值区间。

这些区间不是公开估值标记或融资轮价格;它们是在所有权信息不完整时用于决策的合成价值。

[CV048, CV049, CV050, CV055, CV056, CV057]

8.4 尽调重点与投资逻辑失效触发器

对估值章节来说,尽调议程异常偏法律和结构,因为普通运营尽调不够。关键缺失文件包括 Google 许可和人才协议、Cognition 对剩余业务的购买协议,以及一张当前股权结构表,显示哪些类别参与了 2025 年分拆交易的每一步。没有这些文件,就无法严谨建模稀释或清算优先权负担。运营尽调仍然重要:投资人需要看到 Cognition 内部 Windsurf 企业账户的队列留存,证明定价压力没有永久压缩单位经济,并证明收购后文化扰动没有伤害产品速度或账户覆盖。如果客户留存疲弱、剩余普通股或优先股持有人事实证明几乎没有剩余经济权利,或者 AI 编码价格战把出清倍数推向公开可比公司区间,投资逻辑会迅速失效。反过来,只有当尽调证明存在可转让证券、透明权利,且价格不高于大致基准情景合成区间时,这个判断才可能改善。在此之前,正确姿态不是等待情绪好转,而是避免把战略热度误认为可投资的独立股权机会。[CV034, CV035, CV036, CV037, CV038, CV039]

投资逻辑破裂与终止触发表
触发项阈值如何传导到投资逻辑行动含义
剩余证券不可转让,或被深度劣后律师审查显示,外部持有人几乎没有或完全没有经济请求权机会从可投资资产变成法律尾部权利立即回避
企业留存破裂存续 Windsurf 客户基础出现重大流失或负 NRR乐观和基准情景失去收入分母估值下调到悲观区间
价格战加速席位或用量价格下调,且没有匹配的成本抵消溢价倍数支撑向公开可比公司区间压缩下调价值区间并扩大折扣
整合产品推进停滞Cognition 内部看不到交叉销售、留存证明或一致路线图战略溢价不再可重复将 2025 年成交价视为一次性事件价值
人员或 IP 冲击出现进一步人员流失、法律纠纷,或证据显示授权 IP 无法干净商业化执行和法律悬而未决风险同时上升问题解决前暂停承销

终止触发项按可观察事件来写,这样无需依赖叙事惯性,也能重新审视建议。

[CV038, CV039, CV041, CV045, CV050, CV052]
最终尽调请求表
主题缺失证据重要性负责人或尽调路径
当前股权结构表和瀑布分配证券类别、优先权、参与权、期权处理决定任何存续股权是否仍有意义律师认证的股权结构表加瀑布模型
Google 协议分配流向公司实体的现金、员工薪酬和授权价值之间的分配决定 $2.4B 中有多少真正可由投资者变现要求提供已签署协议或董事会备忘录
Cognition 购买协议名义对价、托管、扣留款,以及按证券类别的分配需要知道剩余资产出售到底值多少钱获取已签署的 SPA 或合并协议
客户留存桥Cognition 内部交易前 Windsurf 队列的 ARR 和账户留存检验 $82M 基础在经济上是否真正存续要求提供队列、NRR 和前 20 大账户状态
交易后员工留存组织架构图、按职能划分的流失率,以及销售覆盖连续性检查被收购资产是否仍能执行与管理层开展 HR 和运营审查
老股证券可得性是否仍能买到任何可转让的独立权利或剩余请求权没有证券就没有入口,无论内在价值如何与律师、经纪人和公司财务联系人确认

这些请求是重新审视回避建议的前置条件,不是可选的后续事项。

[CV036, CV038, CV041, CV049, CV051, CV056]

免责声明

本报告是基于公开证据的尽调快照,不构成投资建议。重要的财务、法律、技术和合同事实仍未公开;做出任何投资决定前,应直接向管理层和原始文件核验。

证据索引

结论
编号陈述可信度来源
CO001 Windsurf began as Exafunction in 2021 and originally focused on optimizing GPU workloads for deep learning. SO001, SO011
CO002 The company pivoted into AI coding software in 2022 and rebranded from Exafunction to Codeium. SO001, SO009
CO003 Windsurf launched the Windsurf Editor in November 2024 and described it as the first agentic IDE. SO001
CO004 The company formally rebranded from Codeium to Windsurf on April 4, 2025. SO001
CO005 CB Insights lists Windsurf as founded in 2021 and based in Mountain View, California. SO011
CO006 CB Insights lists Windsurf’s headquarters at 990 Villa Street, Mountain View, California 94041. SO011
CO007 ToolJunction identifies Varun Mohan and Douglas Chen as Windsurf’s founders. SO010
CO008 ToolJunction describes Varun Mohan as co-founder and CEO before he joined Google DeepMind. SO010, SO008
CO009 ToolJunction describes Douglas Chen as co-founder before he joined Google in 2025. SO010, SO008
CO010 Jeff Wang became Windsurf’s interim CEO after Google hired away the founding CEO and co-founder. SO008
CO011 Windsurf’s public board composition is not disclosed in the sources reviewed for this chapter. SO010, SO011, SO025
CO012 Windsurf raised a $25 million Series A in April 2022. SO009, SO010
CO013 Windsurf raised a $65 million Series B in January 2024. SO009, SO010
CO014 Windsurf raised a $150 million Series C in August 2024 led by General Catalyst with participation from Kleiner Perkins and Greenoaks. SO009, SO010
CO015 The August 2024 Series C valued Windsurf at $1.25 billion. SO009, SO010
CO016 Sacra and ToolJunction both place Windsurf’s total capital raised at approximately $243 million including its undisclosed seed. SO009, SO010
CO017 Google agreed to pay roughly $2.4 billion in July 2025 to hire Varun Mohan, Douglas Chen, and other leaders while licensing certain Windsurf technology without taking control of the company. SO008, SO009
CO018 Cognition signed a definitive agreement on July 14, 2025 to acquire Windsurf’s IP, product, trademark, brand, and operating business. SO007, SO008
CO019 Cognition said the acquisition brought in $82 million of ARR for Windsurf and enterprise ARR that was doubling quarter-over-quarter. SO007, SO008
CO020 Cognition said Windsurf had more than 350 enterprise customers at signing. SO007, SO008
CO021 Cognition said Windsurf also had hundreds of thousands of daily active users at signing. SO007, SO008
CO022 The current Windsurf homepage resolves to Devin Desktop rather than a standalone Windsurf-branded landing page. SO003, SO022
CO023 The current homepage says the product is trusted by over 1 million users worldwide. SO003
CO024 The current homepage also says the product serves more than 4,000 enterprise customers. SO003
CO025 The current FAQ says existing Windsurf users upgrade to Devin Desktop through a standard over-the-air update and keep their plan, pricing, extensions, and settings. SO003
CO026 Current pricing shows a free plan, a $20 per month Pro plan, a $200 per month Max plan, a Teams plan with an $80 monthly base plus $40 per full developer seat, and custom enterprise pricing. SO004, SO018
CO027 Current enterprise packaging includes SAML or OIDC SSO, centralized enterprise admin controls, dedicated account management, and dedicated deployment options. SO004, SO005
CO028 The security material says every session produces a transcript exportable to a SIEM and describes least-privilege controls such as per-repository and per-tool access scoping. SO006
CO029 Sacra estimates Windsurf reached about $40 million ARR in February 2025 and about $82 million ARR by July 2025. SO009
CO030 Sacra says Codeium expanded from roughly 10,000 users in early 2023 to more than 800,000 active developers by early 2025. SO009
CO031 Sacra says the platform processed more than 100 billion tokens daily across more than 70 programming languages by early 2025. SO009
CO032 Sacra says more than 1,000 businesses used Codeium, including Zillow, Dell, and Anduril. SO009
CO033 ToolJunction places Windsurf’s public stage as Series C historically but notes the company was acquired by Cognition in 2025. SO010
CO034 TechCrunch described Windsurf as having roughly 250 employees at the time Google hired away its leadership team. SO008
CO035 ToolJunction gives a broader public employee range of roughly 200 to 350 people. SO010
CO036 Cognition said 100 percent of Windsurf employees would participate financially in its transaction and receive accelerated vesting for their work to date. SO007
CO037 Economic Times reported post-acquisition layoffs and intensified work expectations after Cognition bought the remaining Windsurf business. SO015
CO038 WinBuzzer reported that Cognition offered nine-month buyouts to Windsurf staff while demanding 80-hour workweeks for those who stayed. SO016
CO039 Neither the primary official sources nor the higher-quality third-party profiles reviewed here clearly document Neil Agarwal as a formal co-founder. SO001, SO010, SO011
CM001 Windsurf’s current product surface is an AI-native editor that presents itself as an agent command center rather than a simple autocomplete plugin. SM001
CM002 GitHub Copilot now spans the editor, terminal, GitHub, chat apps, and MCP-connected workflows. SM006
CM003 Amazon Q Developer is positioned across IDE, CLI, console, collaboration, and code-transformation workflows. SM002
CM004 Claude Code is sold as a local or IDE-adjacent coding agent that can read code, write changes, run tests, and open pull requests. SM004
CM005 The relevant Windsurf market includes AI-native IDEs, extension assistants, and terminal or cloud coding agents. SM001, SM002, SM004, SM006, SM010, SM025
CM006 Relevant spend includes seat subscriptions, AI credit pools, governance controls, and agent execution features. SM003, SM008, SM009, SM019, SM020
CM007 Generic AI chat, adjacent DevOps automation, and no-code builders are substitutes or adjacency rather than core Windsurf market spend. SM002, SM006, SM016, SM025
CM008 GitHub’s 2,000-respondent international survey found that more than 97% of respondents had used AI coding tools at work at some point. SM013
CM009 JetBrains reported that 69% of developers had tried ChatGPT for coding and 49% used it regularly. SM014
CM010 JetBrains reported that 40% of developers had tried GitHub Copilot and 26% used it regularly. SM014
CM011 GitHub’s controlled experiment found that developers using Copilot completed a coding task 55% faster than the control group. SM006, SM011
CM012 GitHub’s productivity research also reported that 73% of users stayed in flow and 87% preserved mental effort during repetitive tasks. SM011
CM013 GitHub’s 2024 Octoverse reported a 59% surge in contributions to generative AI projects and a 98% increase in project count. SM012
CM014 Cursor prices its individual plan at $20 per month and its team plan at $40 per user per month. SM009
CM015 GitHub prices Copilot Pro at $10 per month, Pro+ at $39 per month, and Max at $100 per month. SM006, SM008
CM016 Tabnine prices its code assistant at $39 per user per month and its agentic platform at $59 per user per month on annual contracts. SM019
CM017 Amazon Q Developer uses a free tier with 50 monthly agentic chats and a paid upgrade path with pooled limits and per-line overage pricing for transformation workloads. SM002, SM003
CM018 GitHub paused new sign-ups for Copilot Pro, Pro+, Max, and Student on April 20, 2026, and paused new self-serve Business sign-ups on April 22, 2026. SM008
CM019 Developer-Tech reported that GitHub added session limits and weekly token caps because agentic workflows were overwhelming compute infrastructure. SM022
CM020 Sacra estimated that Cursor doubled ARR from $100 million to $200 million in three months and had about 720,000 paying users by March 2025. SM015
CM021 TechCrunch reported that Cursor’s revenue rose from $4 million in annualized recurring revenue in April 2024 to $4 million a month by October 2024. SM018
CM022 Opsera reported that Cursor reached 1 million users and 360,000 paying customers within 16 months of launch. SM017
CM023 Cursor says 64% of Fortune 500 companies use Cursor and more than 50,000 enterprises build with it. SM020
CM024 Cursor says its privacy mode is enabled by default for team members and that it offers SOC 2 Type II, annual penetration testing, and model blocklists. SM021
CM025 GitHub says Business and Enterprise data are not used to train models and that administrators control Copilot policies and budgets. SM006, SM008
CM026 Windsurf’s successor surface claims more than 1 million users and more than 4,000 enterprise customers. SM001
CM027 Windsurf therefore competes in the switch-IDE and agent-command-center segment rather than only as a low-friction add-on to an existing editor. SM001, SM005, SM010
CM028 The buyer path in this market usually begins with individual experimentation before organizations standardize seats or policies. SM013, SM020
CM029 Enterprise buyers pay for governance features such as SSO, SCIM, pooled usage, audit logs, model controls, and privacy mode in addition to model quality. SM009, SM019, SM020, SM021
CM030 Regulated or privacy-sensitive buyers can select private deployment options such as SaaS, VPC, on-premises, or fully air-gapped operation from competitors like Tabnine. SM019
CM031 AWS-heavy teams have a natural substitute in Amazon Q because it is embedded in AWS workflows and collaboration surfaces as well as coding environments. SM002, SM003
CM032 GitHub retains a major distribution advantage because Copilot works inside GitHub, popular IDEs, terminals, and enterprise policy surfaces without requiring an editor switch. SM006, SM008
CM033 Artificial Analysis shows that leading coding agents now span standalone IDE, IDE extension, CLI or local, and cloud deployment models. SM010
CM034 OpenAI’s Codex repository shows that a coding agent can be sold as a terminal tool with optional IDE and cloud experiences rather than as an AI-native editor. SM025
CM035 SWE-bench evaluates models on real-world GitHub issues and the Verified subset contains 500 human-filtered solvable instances. SM007, SM024
CM036 The MSR 2026 study found that no single coding agent led every task category and that task type explained more outcome variance than simple global rankings. SM005
CM037 Benchmark heterogeneity means Windsurf is more likely to win where buyers value multi-file fix, refactor, and agent-orchestration workflows than where they want one universal benchmark winner. SM001, SM005, SM010
CM038 The Verge’s Copilot lawsuit coverage shows that copyright and attribution risk remains an unresolved adoption constraint for code-generating models trained on public repositories. SM023
CM039 The shift from autocomplete to agents is broad because Windsurf, GitHub, AWS, and Claude Code all market multi-step task execution with edits, tests, or pull-request-style outputs. SM001, SM002, SM004, SM006
CM040 Windsurf’s best wedge is teams willing to switch editors for an agent-first workflow but unwilling to lock themselves entirely into GitHub or AWS as the control plane. SM001, SM002, SM006, SM020
CM041 Government-grade and heavily regulated deployments remain a hard segment for Windsurf because its public successor surfaces emphasize enterprise controls but do not document sector-specific accreditations or a clean successor-base mix. SM001, SM021
CP001 GitHub markets Copilot as working across GitHub, IDEs, project tools, chat apps, custom MCP servers, and the command line. SP005
CP002 GitHub's public self-serve Copilot ladder is Free $0, Pro $10 per user per month, Pro+ $39 per user per month, and Max $100 per user per month. SP005, SP006
CP003 GitHub says Copilot Business and Enterprise add centralized management, policy control, and deeper GitHub.com knowledge integration beyond the individual plans. SP005, SP006
CP004 GitHub's March 2026 privacy update says Copilot Free, Pro, and Pro+ interaction data may be used for training unless users opt out, while Business and Enterprise accounts are not affected. SP010, SP005
CP005 GitHub reported that developers using Copilot completed a controlled HTTP server task 55% faster than developers without Copilot. SP007
CP006 GitHub's 2024 global survey said more than 97% of respondents had used AI coding tools at work at some point and that organizations that actively encourage AI report simpler toolchains. SP008
CP007 GitHub says its platform passed 100 million developers and 518 million projects in 2024, giving Copilot the broadest public distribution adjacency in the category. SP009
CP008 Cursor's public pricing page lists a free Hobby tier, a $20 per month Individual tier, a $40 per user per month Teams tier, and enterprise custom pricing. SP001
CP009 Cursor's enterprise page says 64% of Fortune 500 companies use Cursor, 50,000-plus enterprises build with it, and more than 100 million lines of enterprise code are written per day with the product. SP002
CP010 Cursor's enterprise and security pages promise zero data retention, no training on customer data, SAML, SCIM, and SOC 2 Type II. SP002, SP003
CP011 Cursor now markets longer-running agent workflows such as auto-review, /loop, automations, shared canvases, and Jira integration instead of just inline assistance. SP004, SP002
CP012 TechCrunch reported that Cursor surpassed $500 million in ARR and raised capital at a $9.9 billion valuation in 2025. SP029
CP013 Amazon Q Developer is marketed across the IDE, CLI, AWS Console, Slack, Teams, and GitHub preview workflows, and AWS says its agentic capabilities can document, test, review, refactor, and transform code. SP011
CP014 Amazon Q's official pricing surface positions a zero-cost Free tier against a Pro tier at $19 per user per month, with Java upgrade overages priced at $0.003 per line of code. SP012
CP015 Amazon Q Developer Pro subscriptions are designed for personal Builder ID users and IAM Identity Center workforce users, which makes the product easy to embed in AWS identity governance. SP013, SP011
CP016 Amazon Q's data-protection guidance centers on AWS shared responsibility, IAM, MFA, TLS, CloudTrail, and encryption rather than a standalone coding-tool security model. SP014, SP011
CP017 Tabnine's Code Assistant platform lists public pricing at $39 per user per month for chat and completions across major IDEs. SP015
CP018 Tabnine's Agentic Platform lists public pricing at $59 per user per month and adds agentic workflows, MCP tool use, a terminal CLI, and a Context Engine. SP015
CP019 Tabnine's public materials say customers can choose SaaS, VPC, on-premises, or fully air-gapped deployment with zero code retention and no training on customer code. SP015, SP016
CP020 Replit positions itself as an AI app-and-site building workspace with parallel agents, multiple artifacts, and team-oriented execution. SP018
CP021 Replit's public pricing page uses a credit-led structure: Starter is free with daily Agent credits, Core is $20 per month billed annually versus a $25 monthly list, and Pro is $95 per month billed annually versus a $100 monthly list. SP017
CP022 Replit Enterprise adds SSO/SAML and advanced privacy controls, but the public surface still frames pricing around credits and app-building usage rather than classic coding-assistant seats. SP017
CP023 Replit's security documentation says the platform is used by millions of developers and that Replit itself has a SOC 2 Type 2 attestation on top of Google Cloud certifications. SP019
CP024 Claude Code works from the terminal, IDE, Slack, and the web, runs tests and opens pull requests from the local machine, and does not modify files without explicit approval. SP020
CP025 Claude pricing bundles Claude Code into Pro at $17 annual or $20 monthly, Max from $100 per month, and Team at $20 per seat per month billed annually with SSO and no model training on customer content by default. SP021
CP026 Devin pricing positions the successor Windsurf surface at Free $0, Pro $20, Max $200, Teams $80 plus $40 per full developer seat, and enterprise custom. SP022
CP027 Devin docs emphasize native GitHub, Slack, and Jira integrations, broad MCP connectivity, API access, and CI/CD integration rather than only in-editor assistance. SP023
CP028 Devin's security docs say Cognition has SOC 2 Type II, requires MFA, and does not use customer data for model training unless users explicitly opt in. SP024
CP029 Cognition's original SWE-bench technical report said Devin resolved 13.86% of the tested SWE-bench instances. SP025
CP030 SWE-bench's official materials describe a reproducible benchmark for resolving real GitHub issues and note that mini-SWE-agent reached 65% on SWE-bench Verified in 2025. SP026, SP027
CP031 Artificial Analysis classifies Cursor as a standalone IDE with local, CLI, and cloud modes; GitHub Copilot as an IDE extension plus cloud agent; Claude Code as IDE extension, local, CLI, and cloud; Devin as cloud; and Amazon Q as an IDE-native assistant. SP028
CP032 Better Stack's Windsurf comparison preserves the historical list-price ladder at roughly $15 Pro, $30 Teams, and $60 Enterprise, which means Windsurf's old price wedge has largely disappeared under Devin's current surface. SP030, SP022
CP033 GitHub is the hardest incumbent to dislodge because it combines low entry pricing with distribution across GitHub.com, IDEs, the terminal, Spaces, and a 100M-plus developer ecosystem. SP005, SP006, SP009
CP034 Amazon Q is the strongest substitute inside AWS-standardized teams because pricing, identity, operations help, and coding assistance all sit inside an existing AWS control plane. SP011, SP012, SP013, SP014
CP035 Cursor is the most dangerous IDE-native rival to Windsurf's original thesis because it pairs a VS Code fork with cloud agents, enterprise admin features, and meaningful enterprise adoption. SP001, SP002, SP003, SP004, SP029
CP036 Tabnine is the clearest trust-led niche threat because it can satisfy buyers who need VPC, on-premises, or fully air-gapped deployment and explicit no-training promises. SP015, SP016
CP037 Replit is a substitute for greenfield app building and prosumer teams more than for large regulated codebases, so it pressures Windsurf on accessibility and speed rather than on enterprise repo governance. SP017, SP018, SP019
CP038 Claude Code competes on frontier-model quality and local workflow control, but its explicit-approval and local-machine orientation makes it easier to pilot alongside an IDE rather than replace a managed enterprise platform outright. SP020, SP021
CP039 Devin and the successor Windsurf surface still differentiate on higher-delegation cloud agents plus GitHub, Slack, Jira, MCP, and dedicated deployment options rather than plain autocomplete. SP022, SP023, SP024
CP040 Windsurf's clean public price-value wedge is weaker in 2026 because Cursor and Devin both start at $20, Amazon Q is $19, Claude Code rides a $20 Pro plan, and only Copilot undercuts meaningfully at $10. SP001, SP005, SP012, SP021, SP022
CP041 The market is bifurcated between hyperscaler incumbents that win on distribution and governance, IDE-native challengers that win on workflow depth, and agent-native tools that win on autonomy. SP005, SP011, SP020, SP028
CP042 Benchmark differentiation is compressing quickly because Devin's 13.86% 2024 headline score now sits against an official SWE-bench ecosystem where newer or open agents post much higher Verified results. SP025, SP026, SP027
CP043 Multi-homing remains a real substitute because GitHub, Cursor, Claude Code, Devin, Tabnine, and Amazon Q all expose some mix of IDE, terminal, MCP, API, or workflow surfaces instead of forcing a single closed environment. SP004, SP005, SP011, SP015, SP020, SP023, SP028
CP044 From Windsurf's perspective, durable advantage now depends less on raw model access or benchmark screenshots and more on enterprise trust, integration depth, and workflow embed. SP003, SP010, SP015, SP023, SP026, SP029
CP045 The competitive risk is highest where Windsurf once looked strongest: buyers can now get agentic assistance from incumbents without abandoning GitHub or AWS, while Cursor and Claude Code satisfy the power-user segment that once sought a new AI IDE. SP005, SP011, SP020, SP028, SP029
CI001 As of the 2026-06-04 run date, the live successor pricing surface lists Free at $0 per month, Pro at $20 per month, and Max at $200 per month. SI001, SI002
CI002 The same live successor surface sells Teams at $80 per month plus $40 per full developer seat, while Enterprise is custom priced. SI001
CI003 The March 2026 Windsurf pricing update replaced credit balances with daily and weekly quotas and introduced Max as a $200 per month power-user tier. SI002, SI001
CI004 Under the April 2025 reset, Pro was $15 per month for 500 prompt credits, Teams was $30 per user per month for 500 prompt credits per seat, Enterprise was $60 per user per month for 1,000 prompt credits, and add-on credits cost $10 for 250 or $40 for 1,000 pooled credits. SI003, SI015, SI025
CI005 Windsurf eliminated flow action credits because newer reasoning models were using more tool calls per user prompt and making the old credit ratios economically unstable. SI003, SI025
CI006 The 2026 quota system keeps enterprise accounts on existing billing agreements while allowing paid self-serve users to buy extra usage that is consumed at API pricing. SI002, SI001
CI007 Cognition said Windsurf came with $82M ARR, enterprise ARR doubling quarter over quarter, 350+ enterprise customers, and hundreds of thousands of daily active users at signing. SI004, SI005
CI008 Sacra estimated Windsurf grew from roughly $40M ARR in February 2025 to $82M ARR by July 2025. SI006
CI009 Sacra estimated that more than 1,000 businesses used Codeium/Windsurf and more than 800,000 active developers were on the platform by early 2025. SI006
CI010 Public evidence supports a blended revenue stack in which free and paid self-serve plans drive usage while teams and enterprise contracts capture higher-value monetization, but the exact self-serve versus enterprise revenue split is not disclosed. SI001, SI002, SI004, SI006
CI011 Windsurf’s go-to-market motion is product-led into enterprise, using free and Pro plans for adoption before upselling centralized billing, analytics, security, and deployment features to teams and larger accounts. SI001, SI002, SI003, SI006
CI012 The current Devin Desktop homepage says legacy Windsurf plans and pricing carry over through the product migration and claims more than 1 million users and more than 4,000 enterprise customers on the successor surface. SI022
CI013 Sacra characterized Windsurf’s gross margins as materially negative because frontier-model inference costs were exceeding what the company charged users. SI006
CI014 OpenAI’s public API price card lists flagship coding-capable models from GPT-5.4 at $2.50 input and $15 output per 1M tokens up to GPT-5.5 at $5 input and $30 output per 1M tokens. SI014
CI015 GitLab’s FY2025 Form 10-K reported 89% gross margin and warned that AI-powered features can create high computing costs that pressure margin. SI013
CI016 The 2026 quota redesign explicitly says flat credits were mispricing simple versus complex requests because longer sessions now involve dozens of model calls and much wider variance in usage per prompt. SI002
CI017 Charging extra usage at API price is a monetization repair mechanism that should reduce subsidy on heavy users, but realized gross margin still depends on how much premium-model usage remains included inside base quotas. SI001, SI002, SI014
CI018 Windsurf’s enterprise packaging repeatedly pairs higher ARPU with hybrid or dedicated deployment, FedRAMP-related options, and access-control features that can shift part of the delivery burden away from the default self-serve cloud path. SI001, SI003, SI006
CI019 No public source reviewed for this run discloses Windsurf’s CAC, sales efficiency, or payback period, so GTM efficiency must be inferred rather than measured directly. SI006, SI020, SI021
CI020 The Nubank case study reports 8-12x engineering-time efficiency gains and over 20x cost savings on a large refactoring program, giving a concrete enterprise ROI story for commercial sales. SI012
CI021 TechCrunch reported that Anthropic cut Windsurf’s direct Claude access in June 2025 and that some customers switched to rival tools that still offered Claude. SI005
CI022 The March 2026 pricing announcement says enterprise customers remain on existing billing agreements during the transition, implying contract economics differ from self-serve monthly plans. SI002
CI023 The April 2025 pricing reset monetized admin features and heavier usage separately, including pooled add-on credits for teams and enterprise and a planned $10 per user access-control upsell for Teams. SI003, SI025
CI024 The live pricing surface monetizes enterprise expansion through centralized billing, analytics, priority support, dedicated deployment, and account management on top of seat pricing. SI001
CI025 Before the 2025 sale process, Windsurf had raised about $243M in total capital, capped by a $150M Series C at a $1.25B valuation in August 2024. SI006, SI019
CI026 The July 2025 ownership transition split value into two transactions: Google paid about $2.4B to hire the founders and selected researchers while licensing technology, and Cognition separately agreed to acquire the remaining product, brand, IP, and operating business for an undisclosed price. SI005, SI006
CI027 Several September 2025 outlets reported Cognition raised $400M at a $10.2B valuation shortly after acquiring Windsurf. SI008, SI009, SI024
CI028 Tech Funding News instead described the same post-acquisition financing as nearly $500M at a $9.8B valuation, so the exact round size and valuation remain inconsistent across secondary reports. SI018
CI029 VentureBeat and TechCrunch both said Cognition’s net burn had remained under $20M since founding while Devin grew from $1M ARR in September 2024 to $73M ARR in June 2025. SI008, SI024
CI030 Because standalone Windsurf was sold in July 2025, the run-date capital-adequacy question is not another independent round but whether Cognition has enough capital to keep the integrated product funded. SI004, SI008, SI024
CI031 The combination of $82M acquired ARR and Cognition’s post-acquisition financing suggests the successor product surface is not in near-term solvency distress, even though Windsurf’s standalone cash on hand and monthly burn were never publicly disclosed. SI004, SI018, SI024
CI032 Layoffs and voluntary buyouts after the acquisition imply real integration costs and retention risk even if parent-company capital is ample. SI010, SI011
CI033 A standalone next-round underwriting case for Windsurf is largely obsolete because the asset’s value was already realized through the Google licensing and acquihire transaction plus the Cognition asset purchase rather than through a new venture round. SI004, SI005, SI006
CI034 No public evidence reviewed for this run disclosed debt facilities, project finance, or other hard balance-sheet encumbrances for Windsurf. SI006, SI019
CI035 Publicly missing metrics include GAAP revenue, deferred revenue, gross margin by plan or model, net revenue retention, gross and logo churn, CAC, payback, and post-acquisition integration cost. SI004, SI006, SI007
CI036 Revenue quality looks real but only medium grade because ARR and enterprise-customer signals are strong while the public record does not say how much of the $82M base was self-serve versus enterprise or how it held after the ownership shock. SI004, SI005, SI006
CI037 The sequence from flow-action credits to prompt credits to quotas plus API-priced overages is strong evidence that Windsurf has been actively repairing monetization to keep pace with rising model-cost variance. SI001, SI002, SI003
CI038 Capital adequacy looks adequate through the final standalone period and stronger after the sale because Series C had already provided meaningful venture backing and the successor owner then added a large financing round. SI006, SI018, SI024
CI039 Without cohort retention, contract length, and ACV data, investors cannot determine whether enterprise ARR doubling was driven by durable expansions, new logos, or temporarily favorable contract timing. SI004, SI006
CI040 Model-supplier concentration remains the key financial risk because access to premium models affects both product demand and cost of service. SI005, SI013, SI014
CI041 GitLab’s public-filing baseline suggests high gross margins are normal for mature developer-tools software, so Windsurf’s margin problem is category-specific to AI inference rather than a normal software-distribution issue. SI006, SI013
CI042 The correct financial verdict is that Windsurf proved enough demand and strategic value to attract buyers, but the public record is still too thin on mix, margin, and retention to underwrite the business as a clean standalone compounder. SI004, SI006, SI018, SI024
CE001 On June 2, 2026, Windsurf was renamed Devin Desktop while remaining the same underlying IDE and editor surface. SE001, SE002
CE002 The rename keeps extensions, keybindings, workflows, and legacy Windsurf rules working, with .devin/ preferred and .windsurf/ kept as fallback. SE002
CE003 Agent Command Center is a Kanban view for both local Cascade sessions and cloud Devin sessions. SE003, SE004
CE004 Spaces group agent sessions, pull requests, files, and context for a task or project into a single view. SE003, SE001
CE005 Devin in Desktop lets a user hand a local plan to cloud Devin, which runs on its own VM and can keep working after the laptop is closed. SE004, SE001
CE006 Users can continue using Devin Desktop with local-only agents and do not need cloud Devin to preserve the core IDE workflow. SE002, SE004
CE007 Cascade operates in separate Code and Chat modes. SE005
CE008 Cascade uses a specialized planning agent and Todo list to structure longer tasks. SE005
CE009 Cascade can make up to 20 tool calls per prompt and supports queued follow-on messages plus auto-continue. SE005
CE010 Cascade has real-time awareness of user actions and can continue without the user re-supplying prior context. SE005
CE011 Devin Desktop Tab is powered by a custom in-house model trained for speed and flow awareness. SE006
CE012 Supercomplete can suggest deletions and additions around the cursor instead of only next-token insertions. SE006, SE001
CE013 Tab uses recent terminal activity, Cascade chat history, recent code changes, and optionally clipboard state as context. SE006
CE014 Fast Context is marketed as finding the exact files and lines an agent needs in milliseconds. SE001
CE015 Devin Desktop uses a RAG-based context engine that indexes local codebases and can index remote repositories for Teams and Enterprise. SE007
CE016 DeepWiki adds symbol-level explanations beyond classical hover cards and can send those explanations directly into Cascade. SE008
CE017 The Windsurf JetBrains plugin is explicitly stated to continue working under its existing name after the June 2026 rename. SE002
CE018 ACP is an open agent protocol that Devin Desktop compares to LSP for standardizing editor-to-agent integration. SE009
CE019 Documented ACP example agents include Codex CLI, Claude Agent, Junie, Gemini CLI, and OpenCode. SE009
CE020 ACP is available on Pro, Max, and Teams, while enterprise admins must separately enable third-party agents. SE009
CE021 Native integrations cover GitHub, Slack, and Jira, while MCP Marketplace extends the product to monitoring, databases, and documentation tools. SE010
CE022 The public API is split between org-scoped session and knowledge operations and enterprise-scoped analytics, audit logs, billing, and infrastructure endpoints. SE014, SE015
CE023 Enterprise API docs emphasize service users, least privilege, audit-log retrieval, and multi-organization management for automation. SE014, SE015
CE024 Current public self-serve pricing is Free, Pro at $20 per month, Max at $200 per month, Teams at $80 per month minimum plus $40 per full seat, and Enterprise custom. SE011, SE012
CE025 Current pricing includes cloud-agent access on Pro, larger quotas on Max, and SAML or OIDC plus dedicated deployment on Enterprise. SE011, SE018
CE026 Self-serve docs say Pro and Max quotas span Devin sessions, Devin CLI, and Devin Desktop, while Teams flex seats do not include Desktop access. SE012, SE004
CE027 Desktop billing docs say Windsurf introduced new self-serve usage-based plans in March 2026, and self-serve docs describe migration from legacy ACU-based plans to quotas and on-demand credits. SE012, SE016
CE028 Adaptive automatically routes prompts across models and uses fixed per-token pricing to help preserve usage allowance. SE017
CE029 Enterprise marketing says customer data stays in the customer-controlled environment, is never used for training, and can be deployed in a VPC. SE018, SE019
CE030 Cognition security docs say data is encrypted in transit and at rest, Cognition obtained SOC 2 Type II in March 2024, and enterprise customer data is never used for model training. SE019
CE031 Devin for Security markets isolated ephemeral sandboxes, short-lived credentials, secrets redaction, domain allowlists, and SIEM-exportable transcripts with human-controlled merge. SE020
CE032 The admin guide gives enterprise admins team-level control over models, terminal auto-run levels, MCP servers, app deploys, and conversation sharing. SE021
CE033 The admin guide warns MCP can create infrastructure resources outside Devin Desktop security monitoring, so approved-server governance matters. SE021
CE034 The April 2025 rebrand announcement shows the company had already moved from Codeium or Exafunction to Windsurf before the later Devin Desktop rename. SE022
CE035 Cognition’s July 2025 acquisition post describes Windsurf as the agentic IDE, indicating that the IDE remained a strategic asset after the acquisition. SE023
CE036 Cognition’s SWE-bench technical report said Devin resolved 13.86% of sampled SWE-bench issues with a 45-minute runtime limit at launch. SE024
CE037 Current SWE-bench leaderboards list materially higher agent scores than Devin’s launch-era result, so benchmark leadership has shifted since 2024. SE024, SE025
CE038 Artificial Analysis classifies Windsurf as a standalone IDE, GitHub Copilot Coding Agent as IDE extension plus cloud, Claude Code as IDE extension plus local or CLI plus cloud, and Devin as cloud-only. SE026
CE039 GitHub Copilot’s current positioning is to work inside existing editors, GitHub, project tools, chat apps, and custom MCP servers rather than replacing the primary IDE shell. SE027, SE028
CE040 Cursor’s paid plans now bundle cloud agents, shared team context, agentic review, SAML or OIDC, and usage analytics, reducing simple packaging differentiation versus Windsurf. SE029
CE041 Claude Code’s current positioning emphasizes local-machine execution, terminal-centric workflow, and reviewable pull requests rather than a separate IDE shell. SE030
CE042 GitHub’s 2026 survey found more than 97% of respondents had used AI coding tools at work at some point and said organizations still needed clearer adoption roadmaps. SE031
CE043 JetBrains’ 2024 ecosystem survey found 69% had tried ChatGPT for coding, 49% used it regularly, 40% had tried Copilot, and only 11% of organizations completely prohibited third-party cloud AI tools. SE032
CE044 A critical independent review argues Devin’s cloud sandbox is a security advantage but also creates integration friction and opaque per-task cost. SE033
CE045 The Devin Desktop homepage recasts the product as the home for coding agents and the only IDE designed for you to close your laptop. SE001, SE004
CE046 The FAQ says the June 2026 rename does not change customer pricing, but enterprise IT teams may need to update MDM allowlists for the renamed Devin app. SE002
CE047 Current public docs expose admin consumption, analytics, audit logs, and usage tracking, but they do not publish customer-facing uptime or SLA metrics for Devin Desktop or cloud Devin. SE013, SE015, SE021
CE048 Post-acquisition product evolution centers on continuity of the Windsurf IDE plus tighter integration with Devin Cloud, Devin CLI, and Devin Review rather than a clean discontinuity. SE001, SE002, SE004, SE023
CU001 Windsurf’s commercial motion is product-led: free individual developers enter first, team plans formalize shared usage, and enterprise contracts arrive once governance and deployment controls matter. SU002, SU006
CU002 In the current offer, software developers remain the primary users, while engineering leaders and IT administrators act as internal champions and centralized budget owners once SSO, VPC, and admin controls are required. SU002, SU017
CU003 The current Windsurf / Devin Desktop homepage claims 1M+ developers and 4,000+ enterprise customers on the successor surface. SU001
CU004 Anthropic’s customer story says Windsurf had 800,000+ active users and 1,000+ enterprise customers. SU005
CU005 Sacra independently says Codeium / Windsurf had over 800,000 active developers and over 1,000 businesses by early 2025. SU006
CU006 Cognition’s acquisition announcement and TechCrunch both said Windsurf came with 350+ enterprise customers and hundreds of thousands of daily active users at acquisition. SU003, SU004, SU015
CU007 The acquisition materials also described Windsurf as an $82M ARR business whose enterprise ARR was doubling quarter-over-quarter. SU003, SU004, SU006
CU008 Anthropic’s customer story says Windsurf reached 10,000 users within two days of launch and climbed into the hundreds of thousands by the end of week two. SU005
CU009 Windsurf management told Anthropic that daily retention was very high soon after launch, but no numeric retention curve was publicly disclosed. SU005
CU010 Anthropic’s customer story also quotes Windsurf saying that at installed companies, nearly half of all new committed code is written by its AI. SU005
CU011 Sacra names Zillow, Dell, and Anduril as Windsurf / Codeium customers, but the public evidence surfaced there is logo-level rather than a disclosed deployment case study. SU006
CU012 Cognition’s September 2025 funding post says combined enterprise ARR rose more than 30% in the seven weeks after buying Windsurf. SU009, SU010, SU011
CU013 Cognition’s September 2025 funding post says enterprise customer overlap between legacy Windsurf and pre-acquisition Devin was under 5 percent. SU009, SU010, SU011
CU014 Cognition’s post-close roster names Goldman Sachs, Citi, Dell, Cisco, Ramp, Palantir, Nubank, and Mercado Libre as customers on the combined platform. SU009, SU010, SU012
CU015 The current pricing ladder remains freemium: Free, Pro at $20 per month, Max at $200 per month, Team at $80 per month plus $40 per full user seat, and custom-priced Enterprise. SU002
CU016 The current offer still frames plan continuity as an over-the-air upgrade from Windsurf to Devin Desktop, preserving settings, plan, and extensions through the transition. SU001
CU017 The Devin customer roster spans migrations and modernizations, code maintenance, documentation, incident triage, testing, QA automation, rapid prototyping, and tech-debt reduction. SU007
CU018 Nubank used Devin on a core ETL migration spanning more than 6 million lines of code, roughly 100,000 data class implementations, and work that otherwise touched more than 1,000 engineers. SU007, SU008
CU019 Nubank reported 12x engineering-hour efficiency improvement and more than 20x cost savings on the migration scope delegated to Devin. SU007, SU008
CU020 Nubank said Data, Collections, and Risk business units completed migrations in weeks instead of months or years after deploying Devin. SU008
CU021 Nubank’s case study says fine-tuning doubled task completion scores and reduced typical sub-task time from roughly 40 minutes to 10 minutes. SU008
CU022 Ramp says a small team of Devin-savvy engineers now drives up to 80 merged PRs per week and saves more than 10,000 hours per month on rote technical-debt work. SU018
CU023 Ramp says one Devin workflow removed 150 complex feature flags in a month and saved over 1,000 engineering hours on that use case alone. SU018
CU024 Litera says Devin increased test coverage by 40 percent and accelerated regression cycles by 93 percent. SU019
CU025 Litera says it more than doubled active Devin usage within 30 days to 90+ engineers and launched five major products in three weeks. SU019
CU026 Gumroad expanded Devin from an initial engineering pod to broad use across engineering, product, support, and marketing within weeks. SU020
CU027 Gumroad says Devin merged 1,583 pull requests in four months at an above-85 percent merge rate, making it the top contributor on key repositories. SU020
CU028 Hamming says Devin contributes 25 percent of total code volume and ranks alongside its top one to two human engineers by output. SU021
CU029 AngelList says Devin completed a Redshift-to-Snowflake migration 5.2x faster, with 20 agents in parallel migrating 14,000 cards in three weeks. SU022
CU030 Itaú says Devin helped it migrate 59 services from .NET to Java 6x faster, migrate 800 database objects 5x faster, automatically remediate 70 percent of security vulnerabilities, and raise test coverage from roughly 50 percent to above 90 percent. SU023
CU031 Bilt says Devin is embedded across 106 engineers, with more than half of them using it every week and an average of 117 PRs per week over the last 30 days. SU025
CU032 Linktree says Devin authored roughly 300 PRs and merged roughly 100 in one month, much of it on customer-reported bugs, small features, and platform integrations. SU024
CU033 Crossmint says Devin became the number-one contributor to its open-source GOAT SDK by PR count and could merge some integrations with only about an hour of human involvement instead of days of manual work. SU026
CU034 Evinova says Devin cut GxP documentation from 35–40 hours to under 5 hours, produced structured first drafts at roughly 90 percent accuracy, and fixed 66 percent of bugs autonomously in the first 10 days. SU027
CU035 Evinova says Devin reduced a legacy component migration from about five days to roughly a day and a half across three components and compressed test-automation timelines from quarters to weeks. SU027
CU036 Public NRR, GRR, logo-renewal rates, and formal cohort retention metrics for Windsurf are not disclosed on the historical or current successor surfaces. SU003, SU004, SU014
CU037 The archived G2 page says Windsurf did not have enough reviews to provide buying insight, which means independent review depth is still thin relative to the size of the marketed customer base. SU014
CU038 Archived Trustpilot reviews show backlash around pricing changes, refund disputes, exhausted limits after a handful of prompts, and cancellations by long-time users. SU013
CU039 TechCrunch reported that some Windsurf customers switched to services such as Cursor after Anthropic cut Windsurf’s direct Claude access in June 2025. SU004
CU040 Historical Windsurf’s enterprise sales motion appears to begin with free individual usage, then team-wide adoption, and finally enterprise-wide deployment. SU002, SU006
CU041 The post-close opportunity is large because Cognition says legacy Windsurf and Devin enterprise customers overlapped by less than five percent before the transaction. SU009, SU010, SU011
CU042 Business News Today said Cognition planned to keep existing Windsurf support and subscription structures for at least six months, but that workflow and pricing integration was still expected. SU015
CU043 CorpDev argues the combined company still has to reconcile Windsurf’s per-seat licensing with Devin’s usage-based compute pricing, which adds procurement and budgeting friction for enterprise buyers. SU016
CU044 Today’s enterprise offer advertises SSO, centralized admin controls, teamspace isolation, SOC 2 Type II, and VPC deployment, which mitigates some procurement friction for large buyers compared with the thinner historical public record. SU002, SU017
CU045 Current successor evidence is materially stronger than historical Windsurf proof: historical public proof is mostly aggregate metrics and logos, whereas the current Devin surface publishes numerous quantified case studies. SU006, SU007, SU008, SU018, SU019, SU020, SU021, SU022, SU027
CU046 Because Windsurf showed $82M ARR with only a thin set of publicly documented historical deployments, investors cannot rule out meaningful concentration in a handful of large enterprise accounts. SU003, SU004, SU006
CR001 Cognition’s July 2025 deal gave it Windsurf’s IP, product, trademark, and brand, so the remaining investable asset is a post-acquisition business inside Cognition rather than a standalone company. SR002, SR006
CR002 TechCrunch reported that Google’s reverse-acquihire took Windsurf’s CEO, co-founder, and research leaders just before Cognition signed for the remaining business, meaning control of talent and control of the operating company split across buyers. SR006
CR003 Cognition disclosed that the acquired Windsurf business included about $82M ARR, 350+ enterprise customers, and hundreds of thousands of daily active users. SR002, SR006
CR004 Cognition said the Windsurf team would continue operating immediately after signing but that it would invest heavily in integrating Windsurf capabilities and IP into Cognition’s products over the following months. SR002
CR005 Economic Times reported that Cognition laid off about 30 former Windsurf employees and openly acknowledged an extreme work culture shortly after the acquisition. SR007
CR006 WinBuzzer reported that Cognition offered nine-month buyouts to roughly 200 incoming Windsurf employees while resetting expectations around very long work hours. SR008
CR007 The combination of layoffs, buyouts, and a hard culture reset implies material retention risk across product, support, and go-to-market functions during the integration window. SR007, SR008
CR008 Harvard Business School’s Windsurf case says that by April 2025 the company still faced open questions on customer focus, moat, and growth motion in an increasingly crowded AI code assistant market. SR029
CR009 Claude’s customer story says Windsurf used Claude for reasoning and planning inside Cascade and processed more than 100 million tokens per minute. SR009
CR010 Devin’s current pricing page says paid plans provide access to OpenAI, Claude, and Gemini frontier models, confirming that the current Windsurf/Devin product relies on multiple external foundation-model suppliers. SR003, SR009
CR011 Anthropic now sells Claude Code as its own agentic coding product, making one of Windsurf’s model suppliers a direct workflow competitor. SR010, SR021
CR012 Anthropic’s Claude Code page says usage limits apply and that price and plans are subject to change at Anthropic’s discretion. SR010, SR011
CR013 Anthropic’s terms reserve the right to modify, suspend, or discontinue services or access, and to suspend or terminate access without notice in some circumstances. SR011
CR014 OpenAI’s usage policies say users may lose access for breaking safeguards and that OpenAI reserves rights to withhold access when it thinks protection of the service or users requires it. SR012
CR015 Because the current Devin/Windsurf product both depends on third-party frontier models and competes with Anthropic and OpenAI’s own coding agents, model supply is controlled by economically interested counterparties. SR003, SR009, SR010, SR012, SR030
CR016 Devin’s integrations documentation says the product can connect via native integrations, secrets manager, API workflows, and MCP access to hundreds of external tools and data sources. SR004, SR005
CR017 Devin’s security documentation explicitly warns that the system can hallucinate, introduce bugs, or suggest insecure code or procedures and recommends human code review and branch protections. SR005, SR001
CR018 The same security docs describe mitigations including isolated ephemeral sessions, default-deny network allowlists, per-repository and per-tool scoping, human-controlled merges, and session transcripts exportable to a SIEM. SR001, SR005
CR019 OWASP identifies prompt injection, insecure output handling, supply-chain vulnerabilities, insecure plugin design, excessive agency, overreliance, and model denial of service as top security risks for LLM applications and agentic systems. SR025, SR026
CR020 CISA curates guidance specifically on careful adoption of agentic AI services and secure deployment and development of externally developed AI systems, signaling that strong operational controls are becoming a baseline expectation. SR026, SR025
CR021 MintMCP’s Windsurf security review argues that MCP tool usage, terminal execution, and secrets exposure expand blast radius unless admins enforce allowlists and human approval for side-effecting actions. SR031, SR004
CR022 Palo Alto warns that organizations relying on external AI models to generate production code are effectively outsourcing security decisions to systems they do not control and can propagate vulnerabilities faster than normal review processes. SR028, SR025
CR023 Forbes contributors say AI-assisted coding can introduce excessive permissions, insecure code, insecure plugin use, provenance opacity, and leaked proprietary information if teams rely on generated code without review. SR032, SR025
CR024 The EU AI Act says GPAI-provider rules became applicable in August 2025 and include transparency and copyright-related obligations, with further 2026 guidance still being rolled out. SR023
CR025 The AI Act says high-risk AI systems need risk management, documentation, logging, human oversight, and strong cybersecurity before they can be put on the market. SR023, SR026
CR026 The U.S. Copyright Office says it is actively examining copyright in AI training and AI-generated works through a multipart report process, showing that training-data law remains unsettled. SR024
CR027 The Verge reported that the GitHub Copilot class action alleged the product relied on “software piracy on an unprecedented scale” because of how it was trained on public code. SR027
CR028 Taken together, the Copyright Office’s ongoing inquiry and the Copilot litigation show that code-model training provenance and license handling remain live legal risks for AI coding vendors. SR024, SR027
CR029 GitHub’s Copilot plans page says business and enterprise tiers include Copilot cloud agent, broad model access, monthly AI-credit pools, and centralized policy control. SR013, SR014
CR030 GitHub’s official plans page says new sign-ups for multiple Copilot plans and self-serve business sign-ups were temporarily paused in April 2026. SR013
CR031 Developer-Tech reported that GitHub added session limits, weekly caps, and model downgrades because agentic workflows were overwhelming Copilot’s compute infrastructure. SR015, SR013
CR032 Devin’s current pricing is $20 per month for Pro and $200 per month for Max, while Cursor public pricing and TechCrunch reporting frame Cursor around a much lower $20 Pro and $40 business price ladder. SR003, SR016, SR019
CR033 Cursor’s enterprise page claims 64% of Fortune 500 companies use Cursor and over 50,000 enterprises build with it, indicating strong enterprise distribution pressure. SR017, SR019
CR034 Cursor’s security page says Privacy Mode uses zero-data-retention terms with model providers and offers SOC 2 Type II, annual penetration testing, and MCP-specific security guidance. SR018, SR017
CR035 TechCrunch and Sacra say Cursor had already crossed $200M ARR by April 2025 and $500M ARR by June 2025, indicating a rival with unusual growth speed and budget for price and sales pressure. SR019, SR020
CR036 Artificial Analysis lists Windsurf alongside Cursor, Claude Code, GitHub Copilot Coding Agent, Codex, Amazon Q, and other agentic coding products, confirming that the category is already crowded. SR021, SR014
CR037 Public benchmark surfaces such as SWE-bench and coding-agent comparison sites make performance easy to compare, which weakens narrative-only differentiation for Windsurf. SR021, SR022
CR038 OpenAI’s Codex repo says Codex is available in VS Code, Cursor, and Windsurf as well as CLI and cloud surfaces, showing that OpenAI is shipping directly into the same workflow territory its models can also power. SR030, SR012
CR039 Devin’s pricing says extra usage is purchased at API pricing, so provider-cost inflation can flow straight into either gross margin compression or customer bill shock. SR003, SR012
CR040 Anthropic’s product and terms pages plus OpenAI’s policies show that model-provider prices, limits, and access rights can change unilaterally, while Windsurf does not publicly disclose a long-term hedge or proprietary-model fallback. SR010, SR011, SR012
CR041 The acquisition, leadership split, and post-deal culture reports mean Cognition must integrate a large ARR and enterprise base while rebuilding knowledge continuity and retaining key operators. SR002, SR006, SR007, SR008
CR042 Harvard Business School’s case says Windsurf already had more than 700,000 users, 1,000 enterprise customers, and over $40M ARR by April 2025, so a failed transition would destroy substantial existing enterprise value rather than just future optionality. SR029, SR009
CR043 The public mitigation stack is credible but incomplete because the docs do not disclose provider contracts, post-acquisition retention metrics, audited uptime history, or a public bug-bounty program for the combined Windsurf/Devin surface. SR001, SR003, SR005, SR018
CR044 The thesis should break if provider access tightens materially, secure human-reviewed workflows fail in production, or competitor distribution and pricing outrun Windsurf’s ability to hold enterprise customers after integration. SR003, SR007, SR008, SR015, SR019
CV001 Windsurf’s last disclosed fundraising round was a $150 million Series C at a $1.25 billion valuation in August 2024. SV001, SV002
CV002 Windsurf had raised about $243 million of total capital before the 2025 split transaction. SV001, SV002
CV003 Cognition said Windsurf came with $82 million of ARR at signing. SV006, SV007
CV004 Cognition said Windsurf’s enterprise ARR was doubling quarter over quarter at signing. SV006
CV005 Cognition said Windsurf served more than 350 enterprise customers at signing. SV006
CV006 Cognition said Windsurf had hundreds of thousands of daily active users at signing. SV006
CV007 TechCrunch reported that OpenAI’s roughly $3 billion offer for Windsurf expired before Google intervened. SV003, SV007
CV008 Google hired Windsurf CEO Varun Mohan, co-founder Douglas Chen, and selected research staff for DeepMind. SV003, SV004, SV030
CV009 Google paid about $2.4 billion for the hiring-and-licensing arrangement around Windsurf. SV003, SV004, SV005
CV010 Google did not take an equity stake or control of Windsurf in the $2.4 billion deal. SV003
CV011 Google received a non-exclusive license to certain Windsurf technology rather than a full acquisition of the company. SV003
CV012 Jeff Wang became Windsurf’s interim CEO after the founders left for Google. SV003, SV005
CV013 Cognition signed a definitive agreement to acquire the remaining Windsurf business within days of the Google deal. SV005, SV006, SV007
CV014 Cognition said the acquisition included Windsurf’s IP, product, trademark, brand, and business. SV005, SV006
CV015 Benchmarking the August 2024 $1.25 billion Series C against Cognition’s later $82 million ARR disclosure implies about a 15.2x ARR multiple. SV001, SV002, SV006
CV016 Benchmarking the reported $3 billion OpenAI interest against $82 million of ARR implies about a 36.6x ARR multiple. SV003, SV006, SV007
CV017 Benchmarking Google’s $2.4 billion figure against $82 million of ARR implies about a 29.3x ARR multiple. SV003, SV004, SV006
CV018 The Google multiple overstates residual equity value because the deal bought talent and a license rather than whole-company control. SV003, SV004, SV005
CV019 TechCrunch reported that Cursor’s June 2025 round valued the company at $9.9 billion on more than $500 million of ARR, or roughly 19.8x ARR. SV018
CV020 Cursor shows that intact AI-coding leaders can clear high-teen to about 20x ARR multiples in private markets. SV018, SV019
CV021 Sacra estimated combined Cognition ARR at about $492 million in May 2026 after the Windsurf acquisition. SV012
CV022 Cognition’s $10.2 billion September 2025 valuation implies roughly 20.7x on Sacra’s $492 million May 2026 ARR estimate. SV008, SV012
CV023 Combined Cognition therefore screens closer to Cursor’s high-teen or low-20s multiple band than Windsurf’s reported $3 billion OpenAI interest did. SV003, SV012, SV018
CV024 Microsoft agreed to acquire GitHub for $7.5 billion in 2018, showing that strategic buyers can pay multibillion-dollar prices for developer platforms. SV013
CV025 GitHub Copilot is sold across individual, business, and enterprise plans inside GitHub’s larger platform. SV020, SV021
CV026 Because GitHub does not disclose a standalone Copilot equity value, Copilot is a strategic benchmark rather than a clean valuation multiple comp for Windsurf. SV020, SV021
CV027 GitHub’s research found that developers completed coding tasks up to 55% faster with Copilot in controlled tests. SV022
CV028 GitLab reported $759.2 million of total revenue for fiscal 2025. SV014
CV029 GitLab’s June 2026 market capitalization of $5.22 billion implies roughly 6.9x revenue on fiscal 2025 sales. SV014, SV015
CV030 JFrog reported $428.5 million of total revenue for fiscal 2024. SV016
CV031 JFrog’s June 2026 market capitalization of $10.15 billion implies roughly 23.7x revenue on fiscal 2024 sales. SV016, SV017
CV032 Public developer-tooling comps therefore span roughly high-single-digit to low-20s revenue multiples before any control or AI premium. SV014, SV015, SV016, SV017
CV033 Windsurf’s stale 2024 Series C benchmark sits inside that broad public and private comp band, while the two 2025 strategic prints sit above it on an $82 million ARR denominator. SV001, SV003, SV004, SV006, SV014, SV015, SV016, SV017, SV018
CV034 Sacra estimated that about half of Google’s $2.4 billion package went to investors and about half went to compensation packages for the employees who joined Google. SV001
CV035 Public sources do not provide an official purchase-accounting bridge confirming that estimated Google allocation. SV001, SV003, SV005
CV036 Public evidence does not disclose the cap table, liquidation preferences, or security-class treatment after the Google and Cognition transactions. SV005, SV006
CV037 CNBC reported that all Windsurf employees would participate financially in the Cognition deal and receive vesting relief for time served. SV005
CV038 Even with employee-protection language, the absence of disclosed purchase allocation prevents clean underwriting of residual common-holder outcomes. SV005, SV006
CV039 WinBuzzer reported that Cognition offered nine-month buyouts and demanded 80-hour weeks from remaining Windsurf staff after the deal. SV026
CV040 Economic Times separately reported layoffs and long-hour expectations after Cognition bought the remnants of Windsurf. SV027
CV041 Post-acquisition culture disruption raises a plausible customer-retention and execution discount versus treating the asset as intact. SV026, SV027
CV042 TechCrunch reported that Windsurf ARR had reached about $100 million by April 2025, showing momentum immediately before the split transaction. SV003
CV043 TechCrunch also reported that most of Windsurf’s 250-person team remained outside Google after the reverse-acquihire. SV003
CV044 That residue allowed Cognition to buy a functioning product and customer base, but not the original founder-led company investors had backed. SV003, SV006, SV007
CV045 Artificial Analysis and SWE-bench show that AI-coding benchmarks move quickly across vendors, reducing confidence that a momentary strategic premium should be treated as durable margin power. SV024, SV025
CV046 Microsoft’s FY25 Q3 investor press release framed cloud and AI as essential inputs for every business, reinforcing why strategic buyers will still pay up for AI-coding distribution. SV028
CV047 DeepLearning.AI and TechSpot both described Google and Cognition as carving up Windsurf after the failed $3 billion bid. SV029, SV030
CV048 A synthetic bull case of roughly $1.6 billion to $2.2 billion is only defensible if customer relationships stayed intact and a future buyer paid near Cursor-like or high-end public-comp multiples. SV006, SV018, SV028
CV049 A synthetic base case of roughly $0.8 billion to $1.2 billion better fits $82 million of ARR, public comp bands, and a discount for ownership opacity. SV006, SV014, SV015, SV016, SV017, SV018
CV050 A synthetic bear case of roughly $0 billion to $0.4 billion is plausible if residual equity is heavily subordinated or if most economic value was already allocated through the two 2025 deals. SV003, SV005, SV006, SV026, SV027
CV051 The most decision-changing diligence items are the cap table, Google allocation, Cognition purchase agreement, customer-retention bridge, and secondary-security availability. SV005, SV006
CV052 The clearest thesis-break triggers are no transferable security, material customer attrition, pricing-led multiple compression, and legal or retention shocks. SV003, SV005, SV024, SV025, SV026, SV027
CV053 GitHub paused some new Copilot self-serve sign-ups in 2026, showing that frontier demand can outrun delivery economics and that monetization conditions are still being actively managed. SV020, SV021
CV054 GitHub’s reported productivity and satisfaction gains show the AI-coding market remains valuable, but those demand signals do not create a clean standalone Windsurf entry. SV020, SV022, SV023
CV055 The public record supports avoid as the only disciplined recommendation for standalone Windsurf today. SV003, SV005, SV006, SV018
CV056 The call could improve only if diligence proves a transferable security with clear rights and a price at or below roughly the base-case synthetic range. SV005, SV006, SV014, SV015, SV016, SV017, SV018
CV057 Recommendation confidence should remain medium because the structural conclusion is strong while the stakeholder economics remain partially hidden. SV005, SV006
来源
编号出版方标题引文
SO001 Windsurf The Next Chapter: Renaming to Windsurf Today, we are proud to officially rebrand our company to Windsurf.
SO002 Windsurf Windsurf Blog Windsurf is now Devin Desktop.
SO003 Windsurf / Devin Devin Desktop How do I upgrade to Devin Desktop from Windsurf?
SO004 Windsurf / Devin Plans and Pricing Pro $20 per month.
SO005 Windsurf / Devin Enterprise Dedicated deployment option.
SO006 Windsurf / Devin Security Every session produces a transcript exportable to your SIEM.
SO007 Cognition Cognition’s acquisition of Windsurf We’re excited to share that Cognition has signed a definitive agreement to acquire Windsurf, the agentic IDE.
SO008 TechCrunch Cognition, maker of the AI coding agent Devin, acquires Windsurf Google hired away Windsurf’s CEO Varun Mohan, co-founder Douglas Chen, and research leaders in a $2.4 billion reverse-acquihire.
SO009 Sacra Codeium revenue, valuation & funding Windsurf (formerly Codeium) raised approximately $243M in total funding since its 2021 founding.
SO010 ToolJunction Windsurf — Funding, Revenue & Investors (2026) Windsurf has raised a total of $243M in funding and is currently at the Series C (acquired by Cognition in 2025) stage.
SO011 CB Insights Products, Competitors, Financials, Employees, Headquarters Locations It was founded in 2021 and is based in Mountain View, California. In July 2025, Windsurf was acquired by Cognition.
SO012 CNBC Cognition valued at $10.2 billion two months after Windsurf purchase Cognition valued at $10.2 billion two months after Windsurf purchase.
SO013 TechFundingNews Cognition raises $500M at nearly $10B valuation following Windsurf acquisition Cognition raises $500M at nearly $10B valuation following Windsurf acquisition.
SO014 VentureBeat Cognition follows Windsurf acquisition with $400M fundraise, showing strong backing for enterprise AI coding vision Cognition follows Windsurf acquisition with $400M fundraise.
SO015 Economic Times Work-life imbalance: After buying remnants of Windsurf, Cognition lays off 30, tells rest to work long hours After buying remnants of Windsurf, Cognition lays off 30, tells rest to work long hours.
SO016 WinBuzzer Cognition AI’s Culture Reset: Offers Nine-Month Buyouts to Windsurf Staff, Demands 80-Hour Weeks Cognition AI’s Culture Reset: Offers Nine-Month Buyouts to Windsurf Staff, Demands 80-Hour Weeks.
SO017 Devin Customers From scanner finding to merged pull request.
SO018 Devin Plans and Pricing Each paid plan comes with a usage allowance that refreshes automatically on a daily and weekly basis.
SO019 Devin Docs Introducing Devin Devin is designed to be a conversational user interface, and allows you to follow and take over Devin’s development process in the embedded IDE.
SO020 Better Stack GitHub Copilot vs. Cursor vs. Windsurf GitHub Copilot vs. Cursor vs. Windsurf.
SO021 TechCrunch Cognition AI defies turbulence with a $400M raise at $10.2B valuation Cognition AI defies turbulence with a $400M raise at $10.2B valuation.
SO022 Codeium / Windsurf Windsurf — The AI Code Editor Devin Desktop is the home for coding agents to do your best work.
SO023 Devin Devin Devin is an autonomous AI software engineer that can write, run and test code.
SO024 Devin Customers | Nubank Within weeks of Devin’s launch, Nubank identified a clear opportunity to accelerate their refactor at a fraction of the engineering hours.
SO025 PitchBook Company profile verification page Performing security verification
SM001 Windsurf / Devin Devin Desktop Devin Desktop is the home for coding agents to do your best work.
SM002 Amazon Web Services Amazon Q Developer Amazon Q Developer agentic capabilities can autonomously perform a range of tasks–everything from implementing features, documenting, testing, reviewing, and refactoring code.
SM003 Amazon Web Services Pricing The Amazon Q Developer perpetual Free Tier gives you 50 agentic chat interactions per month.
SM004 Anthropic Claude Code It reads code in your local environment, writes changes, runs tests, and opens a PR.
SM005 MSR 2026 / arXiv Comparing AI Coding Agents: A Task-Stratified Analysis of Pull Request Acceptance No single agent performs best across all task types.
SM006 GitHub GitHub Copilot · Your AI pair programmer Assign tasks to agents like Copilot, Claude by Anthropic, and OpenAI Codex, and let them plan, explore, and execute work autonomously in the background.
SM007 SWE-bench SWE-bench Leaderboards Each entry reports the % Resolved metric, the percentage of instances solved.
SM008 GitHub Docs Plans for GitHub Copilot Starting April 20, 2026, new sign-ups for Copilot Pro, Copilot Pro+, Copilot Max, and student plans are temporarily paused.
SM009 Cursor Cursor · Pricing Individual $20 / mo. Teams $40 / user / mo.
SM010 Artificial Analysis Coding Agents Comparison: Cursor, Claude Code, GitHub Copilot, and more Coding Agents Comparison: Cursor, Claude Code, GitHub Copilot, and more.
SM011 GitHub Blog Research: quantifying GitHub Copilot’s impact on developer productivity and happiness Developers who used GitHub Copilot completed the task significantly faster–55% faster than the developers who didn’t use GitHub Copilot.
SM012 GitHub Blog Octoverse: AI leads Python to top language as the number of global developers surges In 2024, there was a 59% surge in the number of contributions to generative AI projects on GitHub and a 98% increase in the number of projects overall.
SM013 GitHub Blog Survey: The AI wave continues to grow on software development teams More than 97% of respondents reported having used AI coding tools at work at some point.
SM014 JetBrains Software Developers Statistics 2024 - State of Developer Ecosystem Report 69% of developers have tried, and 49% regularly use, ChatGPT for coding and other development-related activities. The second most popular AI tool for developers, GitHub Copilot, has been tried by 40% and is regularly used by 26% of our respondents.
SM015 Sacra Cursor at $200M ARR Sacra estimates that Cursor doubled its ARR from $100M to $200M in just three months, with approximately 720,000 paying users.
SM016 The Verge Why tech is racing to adopt AI coding, with Cursor’s Michael Truell Our intention with Cursor is to have it be the best way to build software and, specifically, the best way to code with AI.
SM017 Opsera Cursor AI Adoption Trends: Real Data from the Fastest Growing Coding Tool Within 16 months of launch, it amassed 1 million users, with 360,000 paying customers.
SM018 TechCrunch Exclusive: Benchmark, Index, others are in a wild unsolicited bidding war over Anysphere, maker of Cursor Its developer, Anysphere, has seen its revenue grow from $4 million annualized recurring revenue (ARR) in April to $4 million a month as of last month.
SM019 Tabnine Plans & Pricing | Tabnine: The AI code assistant that you control Flexible deployment options – SaaS, VPC, on-premises, or fully air-gapped. You choose where your code lives.
SM020 Cursor Cursor for Enterprise — Trusted by 64% of Fortune 500 companies 64% Fortune 500 companies using Cursor. 50,000+ Enterprises choose to build with Cursor.
SM021 Cursor Cursor · Security Privacy Mode is available to anyone (free or Pro) and is enabled by default for members of a team.
SM022 Developer-Tech GitHub restricts Copilot as agentic AI workflows strain infrastructure GitHub has paused new sign-ups for its Copilot Pro, Pro+, and Student individual plans.
SM023 The Verge The lawsuit that could rewrite the rules of AI copyright The companies’ creation of AI-powered coding assistant GitHub Copilot relies on “software piracy on an unprecedented scale.”
SM024 SWE-bench / Princeton NLP GitHub - SWE-bench/SWE-bench: SWE-bench: Can Language Models Resolve Real-world Github Issues? SWE-bench is a benchmark for evaluating large language models on real world software issues collected from GitHub.
SM025 OpenAI GitHub - openai/codex: Lightweight coding agent that runs in your terminal Codex CLI is a coding agent from OpenAI that runs locally on your computer.
SP001 Cursor Cursor Pricing
SP002 Cursor Cursor for Enterprise — Trusted by 64% of Fortune 500 companies
SP003 Cursor Cursor Security
SP004 Cursor What's New in Cursor — Latest Updates & Release Notes
SP005 GitHub GitHub Copilot · Your AI pair programmer
SP006 GitHub Docs Plans for GitHub Copilot
SP007 GitHub Blog Research: quantifying GitHub Copilot’s impact on developer productivity and happiness
SP008 GitHub Blog Survey: The AI wave continues to grow on software development teams
SP009 GitHub Blog Octoverse 2024
SP010 GitHub Changelog Updates to our Privacy Statement and Terms of Service: How we use your data
SP011 AWS Amazon Q Developer
SP012 AWS Amazon Q Developer Pricing
SP013 AWS Docs Amazon Q Developer Pro subscriptions
SP014 AWS Docs Data protection in Amazon Q Developer
SP015 Tabnine Tabnine Plans & Pricing
SP016 Tabnine Docs Deployment Options
SP017 Replit Pricing
SP018 Replit Replit – Build apps and sites with AI
SP019 Replit Docs Information Security
SP020 Anthropic Claude Code by Anthropic
SP021 Claude Plans & Pricing
SP022 Devin Plans and Pricing
SP023 Devin Docs Integrations Overview
SP024 Devin Docs Security
SP025 Cognition SWE-bench technical report
SP026 SWE-bench SWE-bench Leaderboards
SP027 SWE-bench GitHub - SWE-bench/SWE-bench
SP028 Artificial Analysis Coding Agents Comparison: Cursor, Claude Code, GitHub Copilot, and more
SP029 TechCrunch Cursor's Anysphere nabs $9.9B valuation, soars past $500M ARR
SP030 Better Stack GitHub Copilot vs. Cursor vs. Windsurf
SI001 Devin Plans and Pricing Each paid plan comes with a usage allowance that refreshes automatically on a daily and weekly basis.
SI002 Devin Introducing our new Windsurf pricing plans The new plans replace the current credit-based system with industry-standard quotas.
SI003 Windsurf An Update to Our Pricing Eliminating flow action credits, so you only pay per user prompt.
SI004 Cognition Cognition’s acquisition of Windsurf $82M of ARR and a fast-growing business, with enterprise ARR doubling quarter-over-quarter.
SI005 TechCrunch Cognition, maker of the AI coding agent Devin, acquires Windsurf Cognition did not announce the price it acquired Windsurf for; however, the company says Windsurf reached $82 million in ARR.
SI006 Sacra Codeium revenue, valuation & funding Gross margins are materially negative, as the cost of running the product — driven primarily by frontier model inference costs — exceeds what the company charges users.
SI007 Sacra Cognition revenue, valuation & funding The addition of Windsurf expanded Cognition’s customer mix from bottoms-up developer subscriptions into larger enterprise contracts.
SI008 VentureBeat Cognition follows Windsurf acquisition with $400M fundraise, showing strong backing for enterprise AI coding vision Before acquiring Windsurf, Cognition’s Devin ARR grew from $1M ARR in September 2024 to $73M ARR in June 2025, with total net burn under $20M across the company’s entire history.
SI009 CNBC Cognition valued at $10.2 billion two months after Windsurf purchase Cognition valued at $10.2 billion two months after Windsurf purchase.
SI010 The Economic Times Work-life imbalance: After buying remnants of Windsurf, Cognition lays off 30, tells rest to work long hours This move by Wu has drawn criticism, sparking a debate about extreme work expectations, employee wellbeing, and the push-and-pull between the so-called hustle culture and work-life balance within the tech industry.
SI011 WinBuzzer Cognition AI’s Culture Reset: Offers Nine-Month Buyouts to Windsurf Staff, Demands 80-Hour Weeks The offer, which gives the roughly 200 affected employees until August 10 to decide, follows the layoff of approximately 30 other Windsurf employees just days prior.
SI012 Devin How Nubank refactors millions of lines of code to improve engineering efficiency with Devin Delivering an 8-12x faster migration, lifting a burden from every engineer, and slashing migration costs by 20x.
SI013 SEC EDGAR GitLab FY2025 Form 10-K Gross margin decreased by 1% to 89% for fiscal year 2025 compared to fiscal year 2024.
SI014 OpenAI OpenAI API Pricing GPT-5.4 ... $2.50 / 1M tokens input ... $15.00 / 1M tokens output.
SI015 PriceTimeline Windsurf Pricing Changes - PriceTimeline
SI016 Verdent Windsurf Pricing: Plans & Quotas Pro: $20/mo → usage quota that refreshes on a daily and weekly basis.
SI017 CloudZero Windsurf Pricing In 2026: Plans, Quotas, And More Old system: Monthly pool of credits. New system: Daily and weekly quotas that refresh automatically.
SI018 Tech Funding News Cognition raises $500M at nearly $10B valuation following Windsurf acquisition Cognition has secured nearly $500 million in Series C funding, valuing the company at $9.8 billion, according to reports.
SI019 ToolJunction Windsurf — Funding, Revenue & Investors (2026) Windsurf ... has raised a total of $243M in funding and is currently at the Series C (acquired by Cognition in 2025) stage.
SI020 Educative Cursor vs. Windsurf vs. GitHub Copilot Windsurf ... focuses heavily on privacy and deep project understanding.
SI021 Aloa GitHub Copilot vs Cursor vs Windsurf 2025: Complete AI Coding Assistant Comparison Windsurf’s mission focuses on democratizing AI coding assistance by providing powerful features for free.
SI022 Devin Devin Desktop Devin Desktop arrives as a standard over-the-air update, so your plan, pricing, extensions, and settings all carry over automatically.
SI023 GitLab Investor Relations GitLab Inc. - Financials & SEC Filings - Annual Reports 2025 Annual Report
SI024 TechCrunch Cognition AI defies turbulence with a $400M raise at $10.2B valuation Cognition AI ... has hit a $10.2 billion valuation after raising $400 million.
SI025 Geekflare Windsurf Made Its Pricing Plans a Lot Simpler Windsurf has revamped its pricing structure. The company has removed the flow action credit and introduced a new flat-rate system to a usage-based model.
SE001 Devin Devin Desktop Supercomplete predicts your next thought, not just your next edit.
SE002 Devin Docs Devin Desktop FAQ On June 2, 2026, Windsurf is becoming Devin Desktop.
SE003 Devin Docs Agent Command Center Manage all of your Devin Desktop agents — local and cloud — from a single Kanban-style view inside Devin Desktop.
SE004 Devin Docs Devin in Devin Desktop Delegate work to Devin, an autonomous cloud agent, directly from Devin Desktop — and review its PRs without leaving your editor.
SE005 Devin Docs Cascade Overview Cascade is Devin Desktop's agentic AI assistant with Code/Chat modes, tool calling, voice input, checkpoints, real-time awareness, and linter integration.
SE006 Devin Docs Devin Desktop Tab Supercomplete is our most powerful and recommended mode, appearing in small windows around your cursor to suggest both deletions and additions.
SE007 Devin Docs Context Awareness Overview Devin Desktop's RAG-based context engine indexes your codebase for intelligent suggestions.
SE008 Devin Docs DeepWiki DeepWiki-powered hover explains functions, variables, and classes as you read through code.
SE009 Devin Docs Agent Client Protocol ACP is an open protocol that standardizes communication between code editors and coding agents — similar to how the Language Server Protocol (LSP) standardized language server integration.
SE010 Devin Docs Integrations Overview MCP (Model Context Protocol) - Connect to hundreds of external tools and data sources
SE011 Devin Plans and Pricing $20 per month
SE012 Devin Docs Self-serve plans A daily and weekly usage quota that covers Devin sessions, Devin CLI, and Devin Desktop.
SE013 Devin Docs Usage Enterprise customers consume Agent Compute Units (ACUs) against the volume in their order form.
SE014 Devin Docs API Overview The Devin API enables you to integrate Devin into your applications, automate workflows, and build powerful tools. Use service users with role-based access control for secure, auditable API access.
SE015 Devin Docs Quick start: Enterprise Follow the principle of least privilege.
SE016 Devin Docs Plans and Usage Windsurf introduced new usage-based plans for self-serve customers in March 2026.
SE017 Devin Docs Adaptive Adaptive is Cognition's intelligent model router that automatically selects the best AI model for each task.
SE018 Devin Enterprise With Devin, your data is saved within your controlled environment and is never used for training.
SE019 Devin Docs Security at Cognition Cognition obtained SOC 2 Type II certification and conducted Security Training in March 2024 for all employees at Cognition.
SE020 Devin Security Every session produces a transcript that can be exported to your SIEM.
SE021 Devin Docs Guide for Admins MCP can create infrastructure resources outside Devin Desktop's security monitoring
SE022 Windsurf The Next Chapter: Renaming to Windsurf Today, we are proud to officially rebrand our company to Windsurf.
SE023 Cognition Cognition’s acquisition of Windsurf We’re excited to share that Cognition has signed a definitive agreement to acquire Windsurf, the agentic IDE.
SE024 Cognition SWE-bench technical report Devin successfully resolves 13.86% of issues, far exceeding the previous highest unassisted baseline of 1.96%.
SE025 SWE-bench SWE-bench Leaderboards Each entry reports the % Resolved metric.
SE026 Artificial Analysis Coding Agents Comparison Windsurf — Standalone IDE.
SE027 GitHub GitHub Copilot Copilot works where you do—in GitHub, your IDE, project tools, chat apps, and custom MCP servers.
SE028 GitHub Docs Plans for GitHub Copilot GitHub Copilot Pro is designed for individuals who want more flexibility.
SE029 Cursor Cursor Pricing Cloud agents and automations with shared team context
SE030 Claude Claude Code Claude works on your local machine.
SE031 GitHub Blog Survey: The AI wave continues to grow on software development teams More than 97% of respondents reported having used AI coding tools at work at some point.
SE032 JetBrains Software Developers Statistics 2024 - State of Developer Ecosystem Report 69% of developers have tried, and 49% regularly use, ChatGPT for coding and other development-related activities.
SE033 LowCode Agency Claude Code vs Devin Devin's sandboxed environment, which is its security advantage, also means its output always needs to be imported back into your toolchain rather than emerging directly from it.
SU001 Windsurf / Devin Devin Desktop homepage
SU002 Windsurf / Devin Plans and pricing
SU003 Cognition Cognition’s acquisition of Windsurf
SU004 TechCrunch Cognition, maker of the AI coding agent Devin, acquires Windsurf
SU005 Anthropic / Claude Customer story | Windsurf
SU006 Sacra Codeium revenue, valuation & funding
SU007 Devin Customers
SU008 Devin Customers | Nubank
SU009 Cognition Funding, growth, and the next frontier of AI coding agents
SU010 Tech Funding News Cognition AI scores $400M at $10.2B valuation as demand spikes for coding agents
SU011 OfficeChai Devin-Maker Cognition Raises $1 Billion At $26 Billion Valuation
SU012 VentureBeat Cognition follows Windsurf acquisition with $400M fundraise, showing strong backing for enterprise AI coding vision
SU013 Trustpilot (archived) Windsurf reviews
SU014 G2 (archived) Windsurf reviews 2026
SU015 Business News Today Cognition AI acquires Windsurf to merge autonomous coding agents with developer-centric IDEs
SU016 CorpDev.org Cognition’s Strategic Acquisition of Windsurf Amid AI Talent Wars
SU017 Devin Enterprise
SU018 Devin Customers | Ramp
SU019 Devin Customers | Litera
SU020 Devin Customers | Gumroad
SU021 Devin Customers | Hamming
SU022 Devin Customers | AngelList
SU023 Devin Customers | Itaú
SU024 Devin Customers | Linktree
SU025 Devin Customers | Bilt
SU026 Devin Customers | Crossmint
SU027 Devin Customers | Evinova
SR001 Windsurf / Devin Security Every session produces a transcript exportable to your SIEM.
SR002 Cognition Cognition’s acquisition of Windsurf The acquisition includes Windsurf’s IP, product, trademark and brand.
SR003 Devin Plans and Pricing Increased quotas, including access to OpenAI, Claude, and Gemini frontier models.
SR004 Devin Docs Integrations Overview - Devin Docs The Model Context Protocol (MCP) allows you to connect Devin to hundreds of external tools and data sources.
SR005 Devin Docs Security at Cognition While Devin’s performance is improving daily, it can still experience hallucinations, introduce bugs into code, or suggest insecure code or procedures.
SR006 TechCrunch Cognition, maker of the AI coding agent Devin, acquires Windsurf Google’s deal occurred just hours after OpenAI’s $3 billion offer to acquire Windsurf expired, leaving much of the startup’s 250-person team behind.
SR007 Economic Times Work-life imbalance: After buying remnants of Windsurf, Cognition lays off 30, tells rest to work long hours After buying remnants of Windsurf, Cognition lays off 30, tells rest to work long hours.
SR008 WinBuzzer Cognition AI’s Culture Reset: Offers Nine-Month Buyouts to Windsurf Staff, Demands 80-Hour Weeks Cognition AI’s Culture Reset: Offers Nine-Month Buyouts to Windsurf Staff, Demands 80-Hour Weeks.
SR009 Claude Customer story | Windsurf | Claude Claude powers some of the reasoning and planning parts of Cascade.
SR010 Anthropic Claude Code by Anthropic | AI Coding Agent, Terminal, IDE Usage limits apply. Prices shown don’t include applicable tax. Price and plans are subject to change at Anthropic’s discretion.
SR011 Anthropic Consumer Terms of Service We reserve the right to modify, suspend, or discontinue the Services or your access to the Services, in whole or in part, at any time without notice to you.
SR012 OpenAI Usage policies We reserve all rights to withhold access where we reasonably believe it necessary to protect our service or users or anyone else.
SR013 GitHub Docs Plans for GitHub Copilot - GitHub Docs Starting April 20, 2026, new sign-ups for Copilot Pro, Copilot Pro+, Copilot Max, and student plans are temporarily paused.
SR014 GitHub GitHub Copilot · Your AI pair programmer Choose from leading LLMs optimized for speed, accuracy, or cost. Use your agents, your way.
SR015 Developer Tech GitHub restricts Copilot as agentic AI workflows strain infrastructure GitHub has paused new sign-ups for its Copilot Pro, Pro+, and Student individual plans.
SR016 Cursor Cursor · Pricing Hobby Free Includes: Limited Agent requests.
SR017 Cursor Cursor for Enterprise — Trusted by 64% of Fortune 500 companies 64% Fortune 500 companies using Cursor.
SR018 Cursor Cursor · Security Privacy Mode can be enabled in settings or by a team or enterprise admin.
SR019 TechCrunch Cursor's Anysphere nabs $9.9B valuation, soars past $500M ARR The company has surpassed $500 million in ARR.
SR020 Sacra Cursor at $200M ARR Cursor doubled its ARR from $100M to $200M in just three months.
SR021 Artificial Analysis Coding Agents Comparison: Cursor, Claude Code, GitHub Copilot, and more Windsurf ... Desktop coding IDE with an agent workflow for multi-file changes.
SR022 SWE-bench SWE-bench Leaderboards SWE-bench Verified is a human-filtered subset of 500 instances.
SR023 European Commission AI Act The AI Act rules on GPAI became effective in August 2025.
SR024 U.S. Copyright Office Copyright and Artificial Intelligence | U.S. Copyright Office The Office is issuing a Report in several Parts analyzing the issues.
SR025 OWASP OWASP Top 10 for Large Language Model Applications | OWASP Foundation LLM08: Excessive Agency.
SR026 CISA Artificial Intelligence | CISA Careful Adoption of Agentic AI Services.
SR027 The Verge The lawsuit that could rewrite the rules of AI copyright software piracy on an unprecedented scale
SR028 Palo Alto Networks The Rise of AI-Powered IDEs: What the Windsurf Acquisition News Mean for Security Teams - Palo Alto Networks Blog When developers use AI to generate 95% of their code, they’re effectively outsourcing many security decisions to the LLM.
SR029 Harvard Business School Windsurf and the AI Code Assistant Market - Case - Faculty & Research Windsurf faced three key strategic questions.
SR030 OpenAI GitHub - openai/codex: Lightweight coding agent that runs in your terminal If you want Codex in your code editor (VS Code, Cursor, Windsurf), install in your IDE.
SR031 MintMCP Windsurf security: how to use AI coding safely | MintMCP Blog MCP tool usage expands the blast radius of agent workflows—enforce admin-approved MCP allowlists, and keep human approval required for side-effecting commands and sensitive tool actions.
SR032 Forbes 20 AI-Assisted Coding Risks And How To Defend Against Them AI-assisted coding tools are prone to unilaterally focusing on getting the coding job completed.
SR033 DEV Community Benchmarking AI-Generated Code: Cursor vs Windsurf vs Secure Coding Standards their outputs aren’t inherently secure.
SV001 Sacra Codeium revenue, valuation & funding Windsurf (formerly Codeium) raised approximately $243M in total funding ... with its last disclosed round being a $150M Series C led by General Catalyst in August 2024 at a $1.25B valuation.
SV002 ToolJunction Windsurf — Funding, Revenue & Investors (2026)
SV003 TechCrunch Windsurf's CEO goes to Google; OpenAI's acquisition falls apart OpenAI’s deal to acquire the viral AI coding startup Windsurf for $3 billion fell apart on Friday.
SV004 Computerworld Google snatches Windsurf execs in a $2.4B deal, derailing OpenAI’s biggest acquisition yet
SV005 CNBC Cognition to buy AI startup Windsurf days after Google poached CEO in $2.4 billion licensing deal
SV006 Cognition Cognition’s acquisition of Windsurf $82M of ARR and a fast-growing business, with enterprise ARR doubling quarter-over-quarter
SV007 TechCrunch Cognition, maker of the AI coding agent Devin, acquires Windsurf
SV008 CNBC Cognition valued at $10.2 billion two months after Windsurf purchase
SV009 Tech Funding News Cognition raises $500M at nearly $10B valuation following Windsurf acquisition
SV010 VentureBeat Cognition follows Windsurf acquisition with $400M fundraise, showing strong backing for enterprise AI coding vision
SV011 TechCrunch Cognition AI defies turbulence with a $400M raise at $10.2B valuation
SV012 Sacra Cognition revenue, valuation & funding
SV013 Microsoft News Microsoft to acquire GitHub for $7.5 billion
SV014 GitLab Inc. (SEC EDGAR) GitLab Inc. Form 10-K — Fiscal Year 2025
SV015 CompaniesMarketCap GitLab (GTLB) - Market capitalization
SV016 JFrog Ltd. (SEC EDGAR) JFrog Ltd. Form 10-K — Fiscal Year 2024
SV017 CompaniesMarketCap JFrog (FROG) - Market capitalization
SV018 TechCrunch Cursor's Anysphere nabs $9.9B valuation, soars past $500M ARR
SV019 Sacra Cursor at $200M ARR
SV020 GitHub GitHub Copilot · Your AI pair programmer
SV021 GitHub Docs Plans for GitHub Copilot
SV022 GitHub Blog Research: quantifying GitHub Copilot’s impact on developer productivity and happiness
SV023 GitHub Blog Octoverse: AI leads Python to top language as the number of global developers surges
SV024 Artificial Analysis Coding Agents Comparison: Cursor, Claude Code, GitHub Copilot, and more
SV025 SWE-bench SWE-bench Leaderboards
SV026 WinBuzzer Cognition AI’s Culture Reset: Offers Nine-Month Buyouts to Windsurf Staff, Demands 80-Hour Weeks
SV027 Economic Times Work-life imbalance: After buying remnants of Windsurf, Cognition lays off 30, tells rest to work long hours
SV028 Microsoft Investor Relations FY25 Q3 - Press Releases - Investor Relations - Microsoft
SV029 DeepLearning.AI Google, Cognition Carve Up Windsurf After OpenAI’s Failed $3B Acquisition Bid
SV030 TechSpot Google hires Windsurf founders, derailing OpenAI's $3 billion acquisition