Main Func
增长和产品野心都是真实的,但公开透明度尚不足以支撑估值,价格纪律仍然重要
公开证据看,Main Func 仍有投资价值;但只有愿意重证据尽调、并在已大幅抬升的估值前守住价格纪律的投资人,才适合继续推进。
封面要素
公司概况
Main Func 是位于 Palo Alto 的公司,Genspark 出自其旗下。Genspark 正在快速扩张,是一个 AI 工作空间和智能体平台,已经从 AI 搜索演进到覆盖文档、幻灯片、电子表格、会议、邮件和工作流的更广泛自主任务执行。公开证据更支持把它看作严肃的未上市企业软件竞争者,而不是新奇 AI 应用;但收入质量和估值支撑仍需要更深的私下尽调。
- 成立时间
- 2023-01-01
- 创始人
- Eric Jing, Kay Zhu, Wen Sang
- 创立地点
- Palo Alto, California, USA
- 总部
- Palo Alto, CA
- 产品
- Genspark 销售的是一体化 AI 工作空间,核心是多模型 Super Agent,并配有 AI Slides、Docs、Sheets、GenMail、Workflows、SecondBrain、AgentBase、Meeting Notes、GenTeam 等模块。产品承诺的是完成后的工作成果,而不是泛泛聊天。
- 客户
- 知识工作者、团队,以及希望在研究、运营、财务、沟通和生产力工作流中自动化白领任务的企业买家。
- 商业模式
- 混合模式:自助购买、团队席位和企业合同并行,背后有积分制用量、高端多模态输出,并向更广泛的团队和企业部署扩张。
- 阶段
- late-growth private company
- 融资情况
- 公开融资锚点从 2025 年 11 月估值 $1.25B 的 $275M Series B,推进到 2026 年后续报道中的总融资超过 $645M、估值约 $2.6B。
执行摘要
主要优势
- 对一家 2023 年才成立的公司,公开增长信号非常强:ARR 快速放大,宣称拥有 7,000 多家企业客户,估值也多次上台阶。
- 作为早期 AI 公司,产品面铺得很宽,覆盖自主研究、创作、沟通、记忆、工作流和团队协作。
- 企业级打包和信任界面比典型 AI 初创公司成熟,已有团队与企业控制、共享上下文产品和可见治理功能。
- 公司所在市场仍愿意给 AI-native 工作流赢家溢价软件叙事。
主要风险
- 公开增长证据强于公开估值支撑;留存、利润率、客户集中度和具体融资条款仍未披露。
- 业务高度依赖前沿模型供应商和连接器,而公司并不能完全控制这些环节。
- Salesforce、Microsoft、Workday、ServiceNow、UiPath 等既有厂商已经掌握大量工作流预算和企业上下文。
- AI agents 一旦从起草走向执行动作、会议捕捉和敏感企业上下文,信任、合规与可靠性负担会迅速加重。
- 如果 $250M 级收入说法没有表面看起来那么稳,最新公开估值就会显得偏高。
未决问题
- 最新估值上调后的当前股权结构表、清算优先权和任何投资人保护条款
- 7,000 多家客户说法背后的队列留存、NRR、GRR、合同期限和客户集中度
- 按模态拆分的毛利率、模型供应商支出集中度,以及扣除支持 / 合规成本后的贡献利润率
- 误动作率、工作流完成率、事故历史、企业安全审查结果等可靠性指标
- 自助、团队、企业收入的具体拆分,以及大客户 ACV 分布
- 2026 年中估值重置后的任何定价老股交易、409A 或融资信号
目录
01公司概览
1.1 身份与产品定位
Main Func 是 Genspark 背后的公司主体。公开资料如今描述的这家公司,已经不太像搜索实验,更像面向知识工作者的智能体化工作空间供应商。公司官网称,其总部在 Palo Alto,由 Microsoft、Google、Meta 和 Pinterest 校友创立,并在 Singapore 和 Tokyo 设有办公室。法律页面把运营结构讲得更清楚:MainFunc Inc. 提供服务,Genspark Inc. 是其全资子公司,genspark.ai 是主要产品入口。产品叙事演进很快。TechCrunch 在 2024 年 6 月报道 Genspark 时,它还是围绕 Sparkpages 打造的 AI 搜索引擎;但 OpenAI、Anthropic 和公司后续发布都指向 2025 年的转向:一个提示词可以产出幻灯片、文档、电子表格、Web 应用或 AI 电话,产品重点变成自主完成任务。这个重新定位很关键,因为后续尽调应把 Main Func 看作工作流自动化和生产力平台,而不只是一个消费者 AI 搜索业务。[CO001, CO002, CO003, CO004, CO005, CO006]
Main Func 的价值链把模型编排层、成品交付、企业信任主张和赞助方资本连在一起。
这是作者对公开可见运营模式的综合,而非公司发布的架构图。
[CO001, CO006, CO010, CO011, CO027, CO039]1.2 创始人、领导层与治理信号
公司强力塑造创始人-市场匹配叙事,底层证据总体也支撑这一说法。2025 年 11 月的融资稿称,CEO Eric Jing 是 Microsoft Bing 创始成员,并曾打造一家估值达到 $5.5 billion 的公司;TechCrunch 和 Baidu Baike 则把 Jing 与 CTO Kay Zhu 都和 Microsoft、Google、Baidu、Xiaodu 的搜索和 AI 经历联系起来。COO Wen Sang 则补上企业软件能力:他拥有 MIT 博士学位,曾创办 Smarking,并在获得 Y Combinator 和 Khosla 支持后退出。治理能见度弱于创始人能见度:公开资料对创始人和投资人信息很丰富,但对正式董事会名单、独立董事或委员会结构披露很少。Emergence Capital 合伙人 Joe Floyd 在 Series B 材料中露出明显,说明主导投资方有影响力,但这不同于成熟上市公司治理体系能提供的透明度。叙事集中在三位经营者身上,既是优势,也带来关键人物依赖。[CO028, CO029, CO030, CO031, CO033, CO035]
| 人物 | 职务 | 背景 | 创始人-市场匹配 / 覆盖面 | 关键人依赖 |
|---|---|---|---|---|
| Eric Jing | 联合创始人兼 CEO | 前 Microsoft Bing 创始成员;后任 Baidu / Xiaodu 高管,并创建 Xiaoice | 锚定产品愿景、搜索 / AI 分发直觉和融资叙事 | 高:公开叙事、招聘光环和投资者信心都高度绑定 Jing |
| Kay Zhu | 联合创始人兼 CTO | 前 Google 搜索排序技术专家,以及 Baidu / Xiaodu CTO | 提供深厚的模型路由、搜索和智能体架构可信度 | 高:产品差异化高度依赖编排质量和技术速度 |
| Wen Sang | 联合创始人兼 COO | MIT 博士;Smarking 创始人,该公司是 Y Combinator / Khosla 支持的企业 SaaS 公司 | 在纯 AI 研究之外补上企业运营和商业化深度 | 中:商业化成熟度看起来与 Sang 相关,但公开细节仍有限 |
| Joe Floyd | Emergence Capital 的主投资人声音 | Emergence 普通合伙人;在 Series B 轮材料中被重点引用 | 显示赞助方相信企业工作流定位 | 中:不是运营者,但赞助方支持看起来是增长计划的关键战略资源 |
本表覆盖公开点名的领导者,以及与治理最相关的主投资人声音;它不是完整董事会或高管名册。
[CO028, CO029, CO030, CO031, CO035]1.3 资本结构、规模说法与运营成熟度
Main Func 的融资节奏异常快。公开报道和公司支持的发布材料显示:2024 年约 $60 million 种子轮,2025 年初 $100 million Series A,2025 年 11 月 20 日以 $1.25 billion 投后估值完成 $275 million Series B,2026 年 3 月扩展后 Series B 达到 $385 million、估值约 $1.6 billion,2026 年 6 月再扩展后总融资超过 $645 million、估值 $2.6 billion。运营规模披露同样快速抬升,但附带不少限制。2025 年 11 月 Series B 材料称,AI Workspace 上线五个月内业务突破 $50 million 年化收入口径;OpenAI 后来引用 Super Agent 上线 45 天后 $36 million ARR,Anthropic 引用转向后 ARR 超过 $250 million,2026 年 7 月新闻稿又加入 7,000+ 商业客户。这些数据点方向上利好,但独立验证仍有限,公开人员规模信号也互相打架:从 2024 年约 20 人团队,到 2026 年 51-100 名员工或约 143 名员工不等,Anthropic 则专门称工程师约 50 人。[CO014, CO015, CO016, CO017, CO018, CO019]
| 指标 | 数值 / 状态 | 日期 / 周期 | 可信度 | 缺口 / 备注 |
|---|---|---|---|---|
| 公司母体 | MainFunc Inc.;Genspark Inc. 为全资子公司 | 条款更新于 2026-04-02 | 中 | 法律结构清晰,但股权结构和董事会构成仍未公开 |
| 总部 | Palo Alto, California | 2025-11 至 2026-07 公开材料 | 高 | 官方和第三方来源口径一致 |
| 其他办公室 | Singapore 和 Tokyo / Japan 运营 | 官网 / Baidu 资料 / 市场帖子 | 中 | 未披露按地区划分的员工结构 |
| 成立时间 | 2023 | 公司历史 | 高 | 未公开披露确切注册日期 |
| Series B 轮锚点 | $275M,投后估值 $1.25B | 2025-11-20 | 高 | 公司背书和投资方背书的发布材料强力交叉印证 |
| Series B 轮扩展 | 总额 $385M,估值约 $1.6B | 2026-03 | 中 | 第三方报道引用 Business Wire 来源,但此处未直接获取 |
| 最新扩展 | $100M;总融资 >$645M;估值 $2.6B | 2026-06 | 中 | 扩展融资报道充分,但投资者社群和 SaaS 新闻摘要仍属于二手来源 |
| 运行收入 / ARR | 2025 年 11 月年化运行收入 $50M;2026 年中 ARR >$250M | 2025-11 至 2026-07 | 中 | 后续 ARR 数字仍主要来自公司或合作伙伴引用,而非独立审计 |
| 客户规模 | 全球数百万用户;7,000+ 商业客户 | 2026-07 | 中 | 公开来源未拆分留存、logo 质量或企业 ACV |
| 员工数可见度 | 公开信号相互冲突:2024 年约 20 人,2026 年为 51-100 人或约 143 人,Anthropic 案例研究称约 50 名工程师 | 2024-06 至 2026-06 | 低 | 当前确切员工数仍未解决 |
数值混合了公司披露和二手摘要;当规模指标彼此冲突或仍为自报时,备注保留区间,而不是强行给出虚假的点估计。
[CO001, CO002, CO003, CO004, CO016, CO019]| 利益相关方 | 角色 | 控制权 / 经济重要性 | 尽调问题 |
|---|---|---|---|
| Emergence Capital | Series B 轮主投资方 | 主导 2025 年 11 月 $275M 轮,并在 2026 年扩展融资中仍居核心位置 | 厘清治理权、董事席位数量,以及是否存在棘轮或优先权条款 |
| Lanchi Ventures | 种子轮主投资方 | 早期支持公司,并在 Series B 轮新闻材料中仍是可见支持者 | 确认后续扩展融资后的持股保留情况,以及中国关联网络是否帮助亚洲分发 |
| LG Technology Ventures | 战略投资方 | 出现在 2025 年 11 月财团名单中,也与企业技术连接性相关 | 评估 LG 除资本外,是否打开渠道或企业试点路径 |
| SBI / Mirae / Pavilion / UpHonest | 全球财务投资方 | 帮助财团横跨亚洲和跨境资本池 | 区分哪些投资者只是被动资本,哪些是主动市场准入伙伴 |
| OpenAI 和 Anthropic | 模型与开发伙伴 | 产品能力、开发者速度和客户证明都离不开它们 | 了解用量承诺、API 集中度,以及终止 / 优先访问保护 |
| Microsoft | 基础设施与分发伙伴 | Azure 加上 M365 集成,可能显著加速企业准入 | 验证集成是否带来付费分发,还是主要属于营销合作 |
这张地图同时放入股权赞助方和平台伙伴,因为 Main Func 的资本故事和产品执行故事已经交织在一起。
[CO014, CO016, CO017, CO019, CO020, CO031]最能影响决策的公开 KPI 显示商业动能极强,但运营质量透明度仍有限。
条目混合官方、合作伙伴和二级披露;在精确数字未被独立验证时保留区间。
[CO016, CO018, CO020, CO021, CO022, CO023]1.4 里程碑与一阶观察点
即便部分指标仍由公司自报,里程碑走势也相当连贯。Genspark 于 2024 年 6 月作为 AI 搜索引擎在 Product Hunt 公开上线,并获得早期社区明显关注。到 2025 年 4 月,公司已经转向智能体化执行;OpenAI 和 Anthropic 后来都把 Super Agent 架构描述为商业拐点。2025 年 11 月随后把两件尽调要事合在一起:大额 Series B,以及 AI Workspace 公开发布,后者把公司从搜索重新框定为围绕成品工作结果展开。2026 年 3 月和 6 月又延展了资本基础,并把产品面扩到 Genspark Claw、Microsoft 集成和更广的企业栈。不过,几个观察点已经出现。TechCrunch 记录过原搜索产品里未解决的安全、法律和商业模式问题。服务条款否认准确性保证,并把合规责任放回用户;隐私政策则确认,用户调用相关能力时,提示词可能流向 OpenAI 和 Anthropic。Techtimes 也点出战略依赖:Main Func 的编排层依赖模型供应商,而这些供应商也在打造自己的企业智能体产品。[CO007, CO008, CO010, CO016, CO019, CO020]
| 日期 | 事件 | 类型 | 金额 / 估值 / 状态 | 参与方 | 含义 |
|---|---|---|---|---|---|
| 2023 | Main Func 在 Palo Alto 成立 | 创立 | 公司成立 | Eric Jing、Kay Zhu、Wen Sang 和早期团队 | 确立公司是新的 AI 应用层,而不是传统软件分拆 |
| 2024-06-18 | Genspark 在 Product Hunt 上线,并排名当日第 4 | 产品 | 469 个赞、168 条评论 | Product Hunt 社区 | 显示 AI 搜索和智能体辅助研究的早期分发能力 |
| 2024-06 | 种子轮融资完成 | 融资 | 种子轮约 $60M,投后估值约 $260M | Lanchi Ventures 和天使投资人 | 在转向工作空间之前提供早期资本 |
| 2024-06 | TechCrunch 将 Genspark 评测为 AI 搜索引擎 | 负面 | 评价混合;商业模式未解决 | TechCrunch | 记录早期准确性、流量和商业模式担忧 |
| 2025-04 | 公司从搜索转向 Super Agent / 工作空间 | 产品 | Super Agent 发布阶段 | Main Func / Genspark | 将核心尽调视角从搜索改为自治工作执行 |
| 2025-11-20 | Series B 轮完成,AI Workspace 发布 | 融资 | $275M,投后估值 $1.25B | Emergence、SBI、LG、Pavilion、UpHonest 等投资方 | 形成第一个独角兽估值锚点,并正式建立企业工作空间叙事 |
| 2026-03 | Series B 轮扩展,Genspark Claw 发布 | 规模 | 总额 $385M,估值约 $1.6B | Emergence 和新增投资者 | 显示跟投信念和更宽的产品野心 |
| 2026-06-11 | Series B 轮扩展再融资 $100M | 融资 | 投后估值 $2.6B;总融资 >$645M | Sozo、UpHonest、Mirae 等 | 三个月内估值大幅上台阶,意味着极激进的前瞻预期 |
| 2026-06-26 | 媒体巡回强调 OpenAI、Anthropic 和 Microsoft 伙伴关系 | 合作 | 合作伙伴栈成型 | OpenAI、Anthropic、Microsoft | 确认平台策略,也进一步加深伙伴依赖 |
| 2026-07-21 | Workspace 6.0 推出,包含 SecondBrain、GenMail、GenTeam、AgentBase | 产品 | 宣称 $250M ARR 和 7,000+ 商业客户 | Main Func / Genspark | 推动公司从一次性生成走向带丰富记忆的工作流软件 |
这条年表是本章唯一的时间线记录;未公开的内部发布、招聘事件和董事会动作可能缺失。
[CO003, CO007, CO008, CO010, CO014, CO016]Main Func 的公开故事在短短两年多时间里,从 AI 搜索发布推进到智能体式工作空间扩张。
日期采用公开可见的最精确锚点;部分产品转型跨越数周,而非发生在某个单一日期。
[CO003, CO007, CO008, CO010, CO016, CO019]1.5 图表摘录
02市场分析
2.1 市场边界与品类契合度
测算 Main Func 机会的第一步,是先判断它到底处在哪个市场。最宽的口径——“职场 AI”——有助于理解长期预算迁移,但不适合支撑对 Main Func 这类公司的投资判断,因为它把硬件、服务、职场设备,以及远超知识工作自动化的企业 AI 支出都打包进来。更相关的是更窄的口径:智能体 AI、AI 生产力工具和 AI 知识工作自动化。这些品类都围绕同一种软件能力收敛:接收一个业务目标,理解上下文,调用工具或系统,并返回一个结果——一个工单解决、一份报告写完、一个工作流跑完、一份合同审完,或一个财务任务完成。这比泛协作软件或完整企业应用支出更贴近 Main Func。品类边界也必须排除实体机器人、模型层基础设施,以及 AI 只是小型嵌入功能的大型企业套件。Main Func 竞争的是 AI 知识自动化与面向工作流的智能体化执行之间的重叠地带,而不是整个职场 AI 宇宙。[CM001, CM002, CM009, CM010, CM011, CM012]
| 细分 / 类别 | 纳入支出 | 排除支出 | 买方 / 付款方 | 与 Main Func 的相关性 |
|---|---|---|---|---|
| 职场 AI(宽口径) | 覆盖主要垂直行业的硬件、软件、服务、职场 AI 和数字化转型支出 | 纯消费者应用基本不在范围内,但这一类别仍包含大量非 Main Func 支出 | 企业 CIO、业务职能负责人、转型预算 | 只能作为上限使用;直接承销 Main Func 时过宽 |
| AI 生产力工具 | 虚拟助理、文档管理、RPA、分析、内容创作、代码辅助、PM、协作、日程安排 | 硬件、深层模型基础设施、非生产力 AI | 部门负责人、IT、一线业务软件预算 | 更贴近 Main Func,因为它覆盖日常知识工作岗位 |
| AI 知识工作自动化 | 内容生成、企业搜索、工作流管理、客户支持、数据分析 | 物理自动化,以及缺少自动化深度的通用协作软件 | 运营、支持、研究、财务和后台负责人 | 匹配度强,因为 Main Func 销售的是完成后的工作,而不是单点回答 |
| 智能体式 AI 软件 | 自治流程自动化、多智能体编排、带工具使用的助理、智能体平台 | 传统脚本机器人、非自治助理、通用基础设施 | AI 平台负责人、自动化 COE、职能流程负责人 | 最直接地框定 Main Func 以编排为核心的产品假设 |
| Main Func 实际目标市场 | 面向研究、财务、支持、销售赋能、HR 和内部运营的跨职能知识工作自动化 | 机器人、通用 AI 基础设施、硬件和泛协作类别 | 职能负责人和平台 / IT 审批人的组合 | 最站得住脚的尽调边界,因为它匹配产品实际做的事 |
同一个供应商可能出现在多行,因为分析师分类彼此重叠;本表讨论的是市场边界逻辑,而不是互斥的公司集合。
[CM009, CM010, CM011, CM012, CM013, CM014]Main Func 可信的市场金字塔,会从宽泛的职场 AI 支出迅速收窄到智能体式知识工作中的实际软件切口。
底层只把宽泛职场 AI 伞面作为背景;中层和顶层才是更相关的投资测算视角。
[CM009, CM014, CM033, CM034]2.2 TAM、SAM 与实用测算口径
已发布的市场数字区间很宽,因为它们衡量的东西不同。The Business Research Company 对 2026 年职场 AI 的估算达到 $421.09 billion,但其定义覆盖多数主要行业中的硬件、软件和服务,因此更像一个数字工作支出伞,而不是 Main Func 可直接服务的干净软件 TAM。更窄的估算对决策更有用。Mordor Intelligence 估计 2026 年智能体 AI 市场为 $9.89 billion,2031 年前 CAGR 为 42.14%;Intel Market Research 把 2026 年 AI 知识自动化市场定为 $14.3 billion;Data Bridge 估计 AI 知识工作自动化在 2025 年为 $5.27 billion,到 2033 年达到 $14.86 billion。这些更窄定义彼此仍有差异,但指向同一结论:Main Func 今天的实用软件市场不是数千亿美元。更可能是一个增长强劲、百亿美元出头的软件机会,供应商只有拿下跨职能工作流预算,而不是像新奇助手一样停留在试用,才能真正赢。真正的建模难点不是抬高 TAM,而是判断哪个窄切口——研究、财务运营、客户支持或内部工作流自动化——最快转化为可重复的生产支出。[CM009, CM011, CM012, CM013, CM014, CM015]
| 发布方 / 视角 | 年份 | 地理范围 | 数值 | CAGR / 渗透率 | 方法论 | 可信度 | 局限 |
|---|---|---|---|---|---|---|---|
| The Business Research Company – 职场 AI | 2026 | 全球 | $421.09B | 38.9% | 宽口径职场 AI 市场,涵盖多个垂直行业的硬件、软件和服务 | 中 | Main Func 的纯软件机会用这一口径会过于宽泛 |
| Mordor Intelligence – 智能体式 AI | 2026 | 全球 | $9.89B | 至 2031 年 42.14% | 独立智能体式 AI 市场,按部署、行业和架构细分 | 中 | 依赖专有估算框架,且可能排除嵌入式智能体支出 |
| Intel Market Research – AI 知识自动化 | 2026 | 全球 | $14.3B | 至 2034 年 15.3% | 知识管理与自动化视角,覆盖 NLP、IDP、助理和分析 | 低 | 定义窄于宽口径职场 AI,但宽于纯智能体编排 |
| Data Bridge – AI 知识工作自动化 | 2025 基准 / 2033 预测 | 全球 | 基准 $5.27B;预测 $14.86B | 13.8% | 面向知识工作的工作流、内容、搜索、支持和数据分析自动化 | 低 | 预览文本没有提供清晰的 2026 年点估计 |
| Gartner – 企业应用渗透率 | 2026 | 全球 | 40% 企业应用内置任务专用智能体 | n/a | 采用渗透率视角,而不是收入 TAM | 高 | 衡量的是嵌入率,不是供应商收入池 |
| 作者约束的 Main Func SAM | 2026 | 全球 / 企业知识工作 | $10B-$16B | n/a | 在较窄的智能体式 AI 和知识自动化视角之间取重叠区间,同时排除硬件和通用职场支出 | 中 | 作者基于公开相邻视角构建,并非直接市场研究引用 |
已发布的市场数字对应不同范围。作者约束的 SAM 刻意窄于宽口径职场 AI 伞形市场,应作为建模辅助,而不是已披露的行业数据。
[CM009, CM012, CM013, CM014, CM016, CM034]公开 2026 年市场视角差异巨大,因为分析师测量的是技术栈的不同层级。
所有数值均为十亿美元;宽泛职场 AI 行有意纳入,用来说明并非所有自上而下 TAM 数字都能直接用于 Main Func。
[CM009, CM012, CM013, CM014, CM034]2.3 买家、用户、付费方与采用路径
买方结构解释了为什么这个市场增长很快,却在采购路径上高度碎片化。供应商证据显示,AI 智能体如今由记录系统和工作流系统的所有者来销售:Workday 把智能体绑定到 HR 和财务数据,ServiceNow 把智能体绑定到 IT、HR、CRM 和风险工作流,Salesforce 把它们打包进服务和销售动作,Microsoft 在 Microsoft 365 里提供通用智能体构建,SAP 把它们落在 ERP 工作流上,RPA 领导者则推动智能体、机器人和人之间的编排。因此,买家通常先是职能预算负责人,其次才是中央 AI 团队。HR 负责人关心自助服务和招聘,财务负责人关心关账流程和审计证据,客服负责人关心工单解决,IT 负责人关心事件和访问工作流,运营或销售负责人关心分析、触达和重复协调。采用通常从可信数据和权限已经存在的地方开始,再扩展到跨应用编排。对 Main Func 来说,结论很清楚:谁最靠近可信上下文,同时还能跨系统交付横向结果,谁就拥有分发优势。[CM015, CM020, CM021, CM022, CM023, CM024]
| 细分 | 买方 | 用户 | 付款方 / 预算所有者 | 工作流 | 采用触发点 |
|---|---|---|---|---|---|
| HR 自助服务与招聘 | CHRO、HRIS 负责人 | 员工、招聘人员、HR 共享服务 | HR 技术预算 | 请假、薪资帮助、招聘、合同 / 政策工作流 | 减少工单,并在可信 HR 数据内完成任务 |
| 财务运营 | CFO、会计控制负责人、采购负责人 | 分析师、AP/AR 人员、审计师 | 财务系统 / 转型预算 | 收入合同、结账、账单排期、审计证据、采购审查 | 压缩周期,并在政策密集型工作中改善控制 |
| IT 与员工支持 | CIO、ITSM 负责人、员工体验负责人 | 服务台员工、内部用户 | IT 运营预算 | 事件处理、访问权限、入职、补丁、工作流路由 | 需要在既有工作流平台内自治解决问题 |
| 客户服务 / CRM | 首席客户官、联络中心负责人、RevOps | 坐席、支持代表、客户 | 服务 / CRM 预算 | 工单解决、呼叫路由、跟进、问题升级 | 24/7 支持经济性和更快解决速度 |
| 销售、营销与外勤运营 | 销售 VP、营销运营、现场服务负责人 | SDR、销售代表、营销人员、协调员 | 收入运营预算 | 线索资格判定、提案支持、个性化触达、工单跟进 | 需要在不按比例扩编的情况下提升吞吐 |
| 研究、报告与一般知识工作 | 部门负责人或 COO | 分析师、财务人员、运营人员、经理 | 职能生产力或转型预算 | 生成报告、分析、制作演示稿、协调、检索知识 | 用结果导向的自动化替代人工归纳和跨工具杂务 |
预算归属通常先落在职能部门,其次才是企业平台;这张图呈现公开厂商材料已能锚定采用场景的地方,而不是理想化的集中式 AI 预算模型。
[CM015, CM020, CM021, CM022, CM023, CM024]可信的企业上下文已经沉淀在哪里,既有厂商就在哪里控制采用路径,也决定哪些买家最容易拿下。
[CM022, CM024, CM025, CM027, CM029, CM030]大多数组织都能想象智能体场景,但真正从助手走到受治理的跨应用自主工作流,少得多。
阶段权重仅作方向性示意,参考了分析师关于采用、治理和项目失败的评论,而不是某个单一基准数据集。
[CM003, CM005, CM016, CM018, CM036, CM037]2.4 增长驱动、采用约束与估值含义
增长逻辑确实存在。Google 称,使用生成式 AI 的组织中已有 52% 的高管把智能体投入生产;Anthropic 发现,80% 受访组织报告了可衡量的经济回报。但同一批研究也说明,采用不是线性的。Anthropic 把集成、数据质量和变革管理列为三大运营摩擦。Gartner 说得更重:到 2027 年底,超过 40% 的智能体 AI 项目会因成本上升、业务价值不清或控制不足而取消;同时,供应商版图里充满“智能体包装”。因此,市场增长依赖的不是模型新鲜感,而是更严格的东西:治理、互操作性、可解释性,以及与现有企业流程的契合。多智能体编排正在成为默认架构,买方怀疑也在成为默认反应。对 Main Func 来说,这意味着市场规模本身不能支撑估值。公司需要证明,它的编排层达到了生产级、可衡量,并且不同于 Workday、Salesforce、ServiceNow、SAP、Microsoft、UiPath 和 Automation Anywhere 已经嵌入应用中的原生智能体。[CM003, CM005, CM006, CM007, CM008, CM016]
| 驱动因素 / 约束 | 方向 | 时点 | 影响 | 尽调问题 |
|---|---|---|---|---|
| 数字化转型与平台现代化 | 驱动 | 现在至 2030 年 | 把更多工作推入智能体可操作、可检索上下文的系统 | 哪些工作流已足够系统化,Main Func 能快速自动化? |
| 意图式计算与更好的模型 / 工具编排 | 驱动 | 现在 | 把上限从简单副驾辅助抬到完整任务交付 | Main Func 能否证明演示之外的产出质量? |
| 部门 ROI 压力 | 驱动 | 当下 | 财务、HR、IT 和支持团队会先在周期缩短、人效提升可衡量的地方采用 | 试点实际回收期是多少? |
| 嵌入权威记录系统的智能体 | 混合 | 现在 | 扩大品类认知,也让既有厂商握住强分发权 | Main Func 如何与应用原生智能体共存,或取而代之? |
| 与现有系统集成 | 约束 | 当下 | 智能体拿不到干净数据,或无法跨工具可靠执行时,就会失败 | 每个用例有多少生产级连接器和安全动作路径? |
| 治理、安全与合规 | 约束 | 当下且持续 | 缺少审计轨迹、控制和政策护栏,可能卡住生产部署 | 面向受监管买家,有哪些治理证据? |
| 算力成本与模型依赖 | 约束 | 当下 | 常驻智能体可能吞掉预算,也受第三方模型经济性牵制 | 每条成功工作流的服务成本是多少? |
| 变革管理与技能 | 约束 | 持续 | 团队不能只买许可证,还要重做工作流并监督智能体 | 部署会带来多大上手和工作流重设计负担? |
方向判断反映当前企业智能体部署状态:品类拉力很强,但规模化仍取决于治理、系统集成和可衡量的 ROI。
[CM003, CM007, CM016, CM018, CM020, CM028]2.5 图表摘录
03竞争格局
3.1 竞争版图与待完成任务替代路径
Main Func 面对的不是一组清晰可比的同业。买家至少可以用五条路径满足同一需求:记录系统内的应用原生智能体、以编排为核心的自动化平台、横向工作空间 copilot、主导单一工作流的创业公司专家,以及用基础模型工具内部自建。Main Func 自己的叙事把它放在横向“成品工作”层:用户表达意图,Genspark 编排多个模型和工具,平台返回研究、幻灯片、分析、销售触达或工作流输出等完成后的交付物。因此,最直接的战略比较不在某个单一聊天机器人,而在客户是否更愿意选择一个跨职能智能体工作空间,而不是 Salesforce、Workday、ServiceNow、Microsoft、SAP、UiPath 或 Automation Anywhere 里的应用内自动化。实际替代集合也就比 AI 智能体创业公司更宽。Fortune 500 买家可以通过扩展既有可信系统、增加一个编排层,或采用 ROI 叙事更紧的专业智能体,解决同一个问题。[CP001, CP002, CP003, CP004, CP005, CP006]
| 竞争对手 | 品类 | 规模 / 融资 | 目标客群 | 差异化 | 局限 |
|---|---|---|---|---|---|
| Main Func / Genspark | 横向 AI 工作空间 / 成品交付智能体层 | 2025 年 11 月以 $1.25B 估值完成 $275M Series B;2026 年另称有后续扩展轮 | 覆盖研究、GTM、财务和一般业务工作流的知识工作者 | 70+ 个模型、成品输出、广泛跨职能覆盖 | 覆盖越宽,分发和信任负担越重 |
| UiPath | RPA + 智能体编排既有厂商 | 上市公司规模 | 自动化 COE、IT、运营、公共部门、SAP 重度企业 | Maestro 控制平面统管智能体、机器人、系统与人 | 工作流优先的基因,可能比消费级工作空间显得更重 |
| Automation Anywhere | 智能体化流程自动化既有厂商 | 成熟企业自动化厂商 | 运营、IT、医疗、制造、金融服务 | Process Reasoning Engine,加上强治理 / 合规叙事 | 在临时知识工作创作上,横向属性不够明显 |
| Salesforce Agentforce | CRM 原生智能体平台 | 大型上市平台规模 | 销售、客服、现场服务、员工服务、IT 服务 | 借 CRM 数据分发,并打包动作 / 对话计费 | Salesforce 已掌握工作流上下文时最有说服力 |
| Microsoft Copilot Studio | 横向智能体构建器与 Microsoft 365 分发 | 大型平台规模 | 通用企业生产力与部门智能体 | 进入 Teams、SharePoint 和 Microsoft 365 Copilot,并采用点数计费 | 通常在 Microsoft 语境内最强,而不是横跨所有系统 |
| Workday Sana | HR 与财务权威记录系统智能体层 | 大型上市平台规模 | CHRO、CFO、共享服务、经理、员工 | 在已有权限和审计控制下执行 HR / 财务工作流 | 重心仍在人和钱两类工作流 |
| ServiceNow AI Agents | 工作流原生企业智能体平台 | 大型上市平台规模 | IT、HR、CRM、风险、安全、应用开发 | AI Agent Orchestrator、Fabric 和 Control Tower 覆盖企业工作流 | ServiceNow 已跑着流程骨干时最契合 |
| SAP Joule Agents | ERP 原生 AI 智能体层 | 大型企业应用规模 | 财务、采购、供应链、以 ERP 为中心的买家 | 业务流程锚定和 SAP Knowledge Graph | 离开 SAP 中心化流程版图,说服力会下降 |
| IBM watsonx Orchestrate 平台 | 控制平面与受治理编排平台 | 大型企业技术规模 | CX、销售、HR、财务、采购、IT 运维 | 开放混合控制平面,提供集中可见性和策略 | 可能更偏平台赋能,而不是对终端用户工作空间友好 |
| Google 智能体平台 | 云智能体平台 / 构建器 | 超大规模云厂商规模 | 开发者、云原生团队、定制企业构建 | 透明 token 经济性与模型访问 | 买家需要自己构建和治理更多环节 |
| Sierra | 客服智能体专精厂商 | 据报 2026 年以 >$15B 估值融资 $950M | 客户体验和联络中心负责人 | 深入的对话式 CX 工具、可观测性和结果优化 | 离开客服 / CX 后,比 Main Func 更窄 |
| Cognition / Devin | 自主软件工程专精厂商 | 据报 2026 年完成巨额融资且估值很高 | 工程团队和软件组织 | 在编码、测试和交付软件上自主性最深 | 不是广义白领自动化工作空间 |
| Ema | AI 员工平台 | 风投支持的创业公司规模 | HR、IT、财务、员工服务工作流 | 100+ 模型融合、治理姿态、1,000+ 连接器主张 | 仍更偏企业流程,而不是成品式创作输出 |
| 11x | 营收团队专精智能体 | 风投支持的创业公司规模 | SDR 和销售组织 | 专用 AI SDR 和电话智能体,并有销售管线证明点 | 相对 Main Func,切入点很窄 |
| Clay | GTM 工作流与数据自动化专精厂商 | 已具规模的 SaaS 工作流厂商 | 营收运营、增长、外拓团队 | 基于动作的工作流自动化,加上数据增补和网页研究智能体 | 主要聚焦 GTM,而不是通用企业工作流自动化 |
规模字段有意混合上市公司状态与融资标记,因为私营创业公司更愿意披露融资,而可靠收入或员工数指标更难拿到。
[CP001, CP003, CP004, CP005, CP006, CP007]Main Func 夹在专精深度与既有厂商生态控制之间:横向覆盖够宽,但自有企业上下文弱于记录系统厂商。
坐标轴采用基于公开证据综合得出的 1–10 定性评分。x 值越高,跨职能工作流覆盖越宽;y 值越高,借自有企业上下文和治理锚点分发的能力越强。
[CP001, CP003, CP012, CP013, CP014, CP015]3.2 既有厂商控制点及其重要性
既有厂商拥有最好的分发位置,因为它们已经坐在可信企业上下文之上。Salesforce 围绕绑定 CRM 工作流的动作、会话和用户附加包为 Agentforce 定价。Microsoft 把智能体分发进 Microsoft 365 Copilot,并通过积分包或按量付费计费。Workday 把 Sana 直接嵌入 HR 和财务工作流,并纳入 Flex Credits,而不是另设一道许可墙。ServiceNow 把智能体延伸到 IT、HR、CRM、风险和应用开发,同时用控制塔强调治理。SAP 围绕流程落地和知识图谱包装 Joule。UiPath 和 Automation Anywhere 正在让 RPA 与智能体化编排融合,并主张真正的企业价值来自同时治理智能体、机器人、系统和人。IBM 和 Google 则从平台侧占据赋能位置:IBM 销售控制平面治理,Google 暴露透明的基于 token 的平台经济模型。合在一起,这些既有厂商说明一件事:市场正从孤立 copilot 走向工作流原生、可编排执行;而最不愿承担集成风险的买家,往往会先从已经拥有相邻预算的供应商开始。[CP012, CP013, CP014, CP015, CP016, CP017]
| 购买标准 | Main Func | 既有厂商模式 | 专精厂商模式 | 影响 |
|---|---|---|---|---|
| 跨职能成品输出 | 一个工作空间里,研究、幻灯片、文档、外联和分析都强 | 常按应用或工作流领域拆开 | 通常只在一个工作流里强 | 买家想用一个界面处理多类知识任务时,Main Func 覆盖最广 |
| 可信权威记录系统上下文 | 中等;取决于连接器和企业配置 | 已掌握 CRM、HR、ERP 或 ITSM 数据的厂商更强 | 随工作流和部署而变 | 数据引力比 UI 或模型多样性更重要时,既有厂商胜出 |
| 治理与审计控制 | 企业姿态在加强,有安全认证和零保留主张 | 平台既有厂商和自动化厂商普遍强且成熟 | 不一;企业导向专精厂商最强 | 受监管买家可能仍会默认选择既有厂商 |
| 多智能体编排 | 围绕多模型和超级智能体执行,有强营销主张 | UiPath、Automation Anywhere、ServiceNow、SAP 和 Workday 都强 | 专精厂商在自己的切口里使用编排 | 这项能力正在变成入场券,而非独特优势 |
| 无代码 / 自然语言配置 | 面向终端用户提示很强 | 各大平台都越来越强 | 常在更窄模板内较强 | 易用性是必要项,但不耐久 |
| 垂直或职能深度 | 横向覆盖中等偏广,但领域深度较轻 | 既有厂商深度掌握某个职能时更强 | 在窄岗位里很强 | Main Func 必须避免在高风险工作流里输给同类最佳 |
| 定价透明度 | 公开 Team Plan 带来中等透明度,企业交易透明度低 | 不一;Salesforce、Microsoft 和 Clay 披露多于其他厂商 | 除部分自助式厂商外通常较低 | 采购摩擦仍给透明定价的对手留下窗口 |
各单元格概括公开证据;厂商披露姿态但不披露经审计性能基准时,表述有意保持定性。
[CP002, CP013, CP014, CP015, CP016, CP020]| 厂商 | 公开套餐 / 价格信号 | 包含能力 | 折扣 / 未知项 | 影响 |
|---|---|---|---|---|
| Main Func / Genspark | Team Plan $30/user/month,每席位含 12,000 点数;企业版定制 | 工作空间访问、顶级模型、管理员控制、SSO/SAML、连接器 | 企业实际成交价未知 | 简单席位锚点有助初始采用,但企业经济性仍不透明 |
| Salesforce Agentforce | $500 / 100k Flex Credits;$2 / 对话;$125/user/month 附加项;版本从 $550 起 | 面向客户与员工的智能体,绑定 Salesforce 工作流 | 大型合同很可能混合点数与预购承诺 | 定价明确,但可能变得复杂且按用量计费 |
| Microsoft Copilot Studio | $200 / 月,含 25,000 Copilot Credits,或按量付费 | 租户级智能体创建,并部署到 Microsoft 界面 | 按组织计费可能掩盖每条工作流的真实成本 | Microsoft 版图内很强,但用量核算很关键 |
| Workday Sana | 打包进 Workday Flex Credits,核心自助智能体没有单独许可墙 | 执行 HR 和财务任务,并提供企业连接器 | 实际价值取决于现有 Workday 合同和 Flex Credit 池 | 打包销售是 Workday 的分发优势 |
| Clay | 免费层;Launch 起价 $167/mo;Growth 起价 $446/mo | 工作流自动化、数据增补、Claygent 网页研究、营销活动支持 | 动作和数据点数扩张会显著改变实际成本 | 在 GTM 专项工作上,Clay 价格低于泛企业智能体平台 |
| UiPath / Automation Anywhere / IBM / SAP / Sierra / Ema / 11x / Cognition 等竞品 | 大多按报价或定制企业定价 | 随平台、工作负载和服务强度而变 | 公开可比性仍低 | 不透明定价拖慢同口径评估,也抬高尽调负担 |
本表比较公开标价信号和包装逻辑,不比较企业实际 ASP 或利润率。
[CP002, CP017, CP020, CP021, CP024, CP027]Main Func 在完成型工作输出广度上领先;当嵌入式上下文和治理主导采购决策时,既有厂商更强。
[CP001, CP013, CP014, CP015, CP016, CP020]3.3 创业公司专家、替代品与 Main Func 的广度
创业公司专家证明,聚焦型赢家仍有空间,但大多数攻击的任务比 Main Func 更窄。Sierra 围绕客户体验和对话式支持构建;Cognition 销售自主软件工程;11x 聚焦收入团队的 AI SDR 和电话智能体;Clay 为 GTM 团队打包工作流自动化和数据补全;Ema 则用强治理叙事销售覆盖 HR、IT 和财务的 AI 员工。这些公司在单一工作流、买方或运营指标上往往比 Main Func 更锋利,也可能缩短销售周期。但它们也留下了跨职能知识工作的空白:同一个用户希望在一个工作空间里完成研究、演示生成、邮件起草、分析和多模型任务执行。Main Func 的优势是横向结果广度和成品工作叙事;劣势是它越做越宽,就越要同时面对资金更足的专家和分发更强的既有厂商。现实中,公司必须证明,一个跨职能工作空间胜过一组同类最优专业工具栈。[CP030, CP031, CP032, CP033, CP034, CP035]
公开定价和产品证明点显示,这个品类在覆盖广度、治理和打包方式上都已高度拥挤。
[CP002, CP017, CP020, CP021, CP024, CP027]3.4 切换成本、多栖使用与护城河耐久度
整个品类的竞争耐久度仍大多未被证明。今天最强的锁定来自数据位置、工作流嵌入、管理员控制和采购便利,而不是纯模型质量。Workday、ServiceNow、Salesforce、SAP 和 Microsoft 都受益于既有身份、数据和工作流锚点。UiPath 和 Automation Anywhere 受益于流程图谱、机器人资产群和治理层,这些东西替换成本高。Sierra 和 Cognition 这类专家如果占据关键任务工作流并收集丰富运营反馈,也能形成更深的切换成本。Main Func 的护城河假设建立在编排广度、输出质量,以及围绕单一界面完成多类任务的用户习惯之上。如果团队把研究、报告和工作流执行集中到 Genspark 内部,这可能变得粘性很强。但如果买家在多个模型工作空间之间多栖使用,把敏感工作流留在记录系统内,或认定专业智能体比宽泛横向层更安全,它就会很脆弱。竞争问题不是 Main Func 有没有产品,而是它的广度能否在既有厂商智能体和更便宜的模型型替代方案吸收同类用例之前,复利成持久的工作流重力。[CP039, CP040, CP041, CP042]
| 护城河主张 | 威胁 | 严重性 | 缓解 / 尽调问题 |
|---|---|---|---|
| 广泛的成品交付工作空间 | 既有厂商增加更多跨应用动作和文档生成 | 高 | 衡量客户是否把任务整合进 Genspark,而不只是试用功能 |
| 多模型编排优势 | 模型路由在竞争对手之间商品化 | 高 | 要求证明产出质量或完成成本在结构上优于竞争对手 |
| 用户习惯与界面简洁 | Microsoft、Salesforce 和 Workday 已掌握日常工作流界面 | 高 | 按职能评估日活使用和粘性,而不只看客户 logo 数 |
| 安全与合规姿态 | 既有厂商仍有更长治理记录和原生审计锚点 | 中高 | 审查企业安全评估、事故历史和续约阻碍 |
| 横向品类宽度 | 专精厂商在 CX、编码、GTM 或 HR / 财务工作流深度上击败 Main Func | 高 | 量化横向宽度真正胜出的场景,以及专精厂商胜出的场景 |
| 价格门槛 | 窄任务上,用量计费对手可能看起来更便宜;现有套件里,打包方案可能看起来更便宜 | 中 | 把常见工作流全年总成本与 3-5 个替代方案对比 |
| 连接器广度与生态触达 | 权威记录系统厂商可能限制开放性,或偏向自家智能体 | 中高 | 审计生产环境中的连接器质量、权限和工作流完成率 |
| 产出质量与信任 | 单一泛平台可能在高风险领域准确性失守 | 高 | 在财务、HR、法务和客服用例上跑竞争性对比测试 |
风险登记表把分发权和工作流嵌入视为企业智能体软件最难跨越的护城河。
[CP038, CP039, CP040, CP041, CP042]3.5 图表摘录
04财务
4.1 收入来源与变现逻辑
Main Func 的变现看起来是自助订阅、团队计划和定制企业协议的混合,而不是单一经典 SaaS 席位模式。最清楚的公开锚点是 Genspark for Business:其 Team Plan 标价为每用户每月 $30,每席位含 12,000 credits,并提供集中计费、SSO/SAML、发票、连接器管理,以及对高端模型的广泛访问。帮助中心的工作流文档展示了产品如何扩展到定时和邮件触发的自动化,这说明变现可以从简单席位访问延伸到更高价值的工作流用量。这个点很重要,因为平台卖的不只是聊天,而是覆盖幻灯片、图像、视频、电话、研究和集成的高成本输出。公开叙事也反复同时强调个人用户和企业客户,说明收入基础可能是混合的:上层靠病毒式自助获客,下层承接谈判型企业合同。实际看,Main Func 的商业模式更像一个成本随消耗变化的工作空间平台,而不是一个用量模式统一、按席位收费的干净 SaaS 产品。[CI001, CI002, CI003, CI004, CI005, CI006]
| 收入流 | 机制 | 单位 | 当前价值 / 状态 | 质量 | 尽调问题 |
|---|---|---|---|---|---|
| 个人订阅 | 消费者 / 专业用户访问 AI 工作空间 | 按用户 / 订阅 | 产品定位可见,但公开个人套餐细节少于商业套餐细节 | 中 | ARR 中自助套餐与企业版分别占多少? |
| Team Plan | 多席位订阅,含月度点数 | 按用户每月 | 公开标价为 $30/user/month,每席位含 12,000 点数 | 高 | 折扣和年度条款后,有效净席位价格是多少? |
| 企业版计划 | 定制合同,包含管理员控制、连接器、安全,以及可能的支持 / 服务 | 年度合同 / 企业 ACV | 公开信息只有描述,没有定价 | 中 | 中位 ACV、实施费和续约条款是多少? |
| 工作流自动化用量 | 跨邮件和已连接工具的更高频定时或触发式工作 | 运行次数 / 任务量 / 点数 | 公开工作流和连接器可见,但具体变现方式未披露 | 中低 | 收入中有多少与用量挂钩,又有多少打包进订阅层级? |
| 创意 / 多模态输出工作负载 | 一个套餐内生成幻灯片、图像、视频、语音和研究内容 | 积分 / 计算负载 | 明确属于产品方案,但未按模态公开工作负载定价 | 低 | 各模态、各企业客户队列的贡献利润率是多少? |
收入组合几乎肯定把按席位访问和对用量敏感的成本结构混在一起;缺口在于,多少使用量被明确计费,多少被订阅包吸收。
[CI001, CI002, CI003, CI004, CI005, CI006]| 价格 / 合同模式 | 标价与实际价格 | 折扣 / 未知项 | 来源 | 含义 |
|---|---|---|---|---|
| Genspark Team 计划:$30/user/month,12,000 credits/seat | 仅标价 | 企业折扣和超额使用规则未知 | Genspark for Business 企业方案 | 进入门槛低,但不足以推断实际 ASP |
| Enterprise 计划:定制 | 无公开标价 | 合同期限、实施费用、最低消费和超额用量未知 | Genspark for Business 企业方案 | 企业变现质量仍是核心尽调问题 |
| Salesforce:每 100k Flex Credits 收 $500;每次对话 $2;用户加购项 | 公开标价 | 大企业预购折扣很可能存在 | Salesforce 定价页 | 竞争定价压力越来越转向按用量计费 |
| Microsoft:$200 / 25,000 Copilot Credits,或按量付费 | 公开标价 | 真实成本取决于智能体工作负载组合 | Microsoft Copilot Studio | 这类定价让客户按用量核算智能体成本 |
| Clay:免费层;Launch 起价 $167/mo;Growth 起价 $446/mo | 公开标价 | 靠动作和数据积分扩张 | Clay 定价 | 单一功能场景里,垂直专家看起来更便宜 |
Main Func 的标价比许多企业 AI 厂商更透明,但买方仍需要按工作流和使用强度拆开的实际成本视图。
[CI001, CI005, CI026, CI027]Main Func 的商业化桥梁从自助意图捕获出发,延伸到团队订阅、企业工作流采用,以及价值更高的经常性自动化。
[CI001, CI002, CI003, CI004, CI005, CI006]4.2 公开牵引力与收入质量信号
最醒目的公开证据,是公司报告收入里程碑的速度。OpenAI 称 Super Agent 上线后仅 45 天就达到 $36 million ARR;LG 和 PR Newswire 称 Genspark 在五个月内突破 $50 million 年化收入口径;Business Wire 称公司九个月内超过 $100 million ARR;Anthropic 称其转向 Super Agent 后超过 $250 million ARR;2026 年 7 月发布稿报道又称其服务 7,000 多个商业客户,同时 ARR 仍约为 $250 million。即便考虑公司主导叙事,以及 ARR 和年化收入口径之间的差异,这一时间序列在方向上是一致的:Main Func 正在快速变现。收入质量则没那么清楚。业务似乎混合了企业合同和病毒式产品驱动采用,但公开资料没有披露流失、同期群表现、客户集中度,或自助、团队与真正企业收入之间的拆分。对投资人来说,这意味着收入增长故事令人兴奋,但底层耐久性故事仍大多不透明。[CI007, CI008, CI009, CI010, CI011, CI012]
| 指标 | 数值 / null | 可信度 | 为什么重要 | 尽调要求 |
|---|---|---|---|---|
| ARR / 年化收入运行率 | 不同日期和来源的公开里程碑从 $36M 到 >$250M 不等 | 中 | 营收动能是估值的核心驱动 | 提供按月 ARR 桥表,注明定义和权威日期 |
| 企业客户数 | 2026 年 1 月称 1,000+ 家组织;2026 年 7 月称 7,000+ 家商业客户 | 中 | 客户 Logo 增长能反映销售动作质量和收入集中度 | 按细分市场和合同类型披露活跃付费客户 |
| 毛利率 | null | 低 | 毛利质量决定增长是在复利滚动,还是被计算成本拖着跑 | 按季度和产品类型提供 GAAP / 管理口径毛利率 |
| CAC 回本周期 | null | 低 | 评估 PLG 与企业销售效率必须看这个指标 | 按细分市场披露付费获客、销售费用和回本周期 |
| 净留存 / 扩张 | null | 低 | 判断企业粘性和先落地再扩张表现不可缺 | 按队列提供 NRR / GRR 和扩张情况 |
| 按工作流拆分的贡献利润率 | null | 低 | 高成本输出可能被席位包裹住,掩盖弱经济性 | 展示研究、幻灯片、语音、视频和自动化工作负载的利润率 |
公司有公开的收入速度,但几乎没有传统 SaaS 投资判断所需的内部运营指标。
[CI007, CI008, CI009, CI010, CI011, CI012]公开收入和融资里程碑指向异常增长,但它们来自不同时间点和叙事来源;应按区间看,而不是当成一条干净的时间序列。
这些数值只是公开的方向性检查点,不是 GAAP 收入桥。ARR 和年化 run rate 保留引用来源披露的表述。
[CI007, CI008, CI009, CI010, CI011, CI012]4.3 成本结构、利润率与运营杠杆代理指标
Main Func 仍然过于私密,无法直接测算毛利率;最好的公开方法,是把产品架构线索和规模化可比公司合在一起看。架构很可能承载不低的推理和服务成本,因为 Genspark 在多个前沿模型、语音、图像和视频输出之间路由工作,并且越来越多地运行定时或长周期自动化。零保留、SSO、连接器和合规等企业安全承诺,也会带来平台和支持开销。这些成本意味着,Main Func 近期毛利率几乎肯定低于成熟工作流软件同行。但公开可比公司显示,终局仍可能有吸引力:UiPath FY2026 GAAP 毛利率为 83%,Workday 和 Salesforce 也展现了强订阅收入规模、运营利润率和现金生成能力。财务解读是,如果 Main Func 改善路由效率、与模型供应商的采购经济性以及企业支持杠杆,利润率路径最终可能接近高质量软件。在那之前,主要风险是多模态工作负载和慷慨的内含 credits 让收入看起来强于真实单位经济。[CI014, CI015, CI016, CI017, CI018, CI019]
| 手头现金 / 资本来源 | 月度烧钱 / 现金跑道(月) | 计划资金用途 | 下一轮触发因素 | 债务 / 项目融资义务 |
|---|---|---|---|---|
| 截至 2026 年 6 月公开报道累计股权融资 >$645M | null / 未披露 | 产品研发、模型采购、国际扩张、企业 GTM、支持和基础设施可能是主要用途 | 下一轮很可能取决于能否在维持超高速增长的同时,证明单位经济性能长期成立 | 未公开披露债务、项目融资或结构化融资工具 |
| 2025 年 11 月以 $1.25B 投后估值完成 $275M Series B 轮 | null / 未披露 | 推出 AI Workspace,并推进企业 GTM | 为早期验证后的规模化提供资金 | 未公开披露债务 |
| 据报道,2026 年 4 月完成 $385M Series B 延伸轮,估值约 $1.6B | null / 未披露 | 更大的产品野心和规模化 | 显示投资人在盈利能力证明前就出钱押注加速 | 未公开披露债务 |
| 据报道,2026 年 6 月以 $2.6B 估值完成 $100M 延伸轮 | null / 未披露 | 报道未披露具体资金用途 | 大幅抬高增长预期 | 未公开披露债务 |
资本故事从累计融资额看很强,但没有公开证据能钉住现金余额、烧钱速度或现金跑道。
[CI029, CI030, CI031, CI032, CI033, CI034]公开证据显示,软件式收入经济性要穿过计算和支持成本较重的成本栈;规模扩大后应会改善,但今天仍未被证明。
[CI014, CI015, CI016, CI017, CI018, CI019]公司资本强度更多由计算、模型采购和企业支持驱动,而不是硬资产;但未披露烧钱速度,仍是投资判断最大的卡点。
[CI029, CI030, CI031, CI032, CI033, CI034]4.4 资本充足性、融资依赖与尽调障碍
从资本充足性看,公司短期很强,中期不透明。按公开报道,Main Func 于 2025 年 11 月以 $1.25 billion 投后估值完成 $275 million Series B,2026 年 4 月前后把该轮扩展到约 $1.6 billion 估值、$385 million,随后在 2026 年 6 月又融资 $100 million,使总融资超过 $645 million、估值达到 $2.6 billion。对一家 2023 年成立的软件公司来说,这笔股权资本很可观,应该足以支持公司投入基础设施、模型采购、国际 GTM 和企业支持。未解决的问题是烧钱速度。没有公开来源披露现金余额、月度现金消耗、资金可支撑时间、销售效率,或公司预计何时转正自由现金流。此外,2026 年融资叙事内部有不一致:一份发布称该轮补到 $300 million;另一份称 $385 million;后续报道又称总融资超过 $645 million。这些不是致命矛盾,但确实强化了一个需求:在承销下一轮或盈利路径前,必须先拿到干净的融资台账。[CI007, CI010, CI011, CI029, CI030, CI031]
| 缺失的内部指标 | 影响 | 精确尽调路径 |
|---|---|---|
| 现金余额和月度烧钱 | 无法准确评估现金跑道或融资依赖 | 索取董事会报告包,包含月度现金桥表和预测烧钱 |
| 按产品 / 模态拆分的毛利率 | 无法判断计算密集型输出是否赚钱 | 审阅季度毛利率,以及按模态和企业层级拆分的 COGS |
| 客户集中度和收入结构 | 无法判断 ARR 是否分散,还是依赖少数大客户 | 索取前 20 大客户贡献,以及自助 / 团队 / 企业收入结构 |
| 留存和扩张指标 | 无法为当前 ARR 声明的耐久性背书 | 提供按队列的留存、NRR、GRR 和按细分市场的增购 |
| 销售效率和管道转化 | 无法评估当前增长能否在不过度花钱的情况下复制 | 审阅各渠道 CAC、回本周期、管道覆盖和成交率 |
| 模型供应商承诺和云经济性 | 无法评估成本杠杆或供应商集中风险 | 披露主要模型 / 云合同、支出集中度和承诺最低消费 |
这些不是锦上添花的指标;它们是判断 Main Func 究竟是高质量软件,还是只是高增长软件时缺失的拼图。
[CI019, CI020, CI028, CI036, CI037, CI038]4.5 图表摘录
05产品与技术
5.1 产品定义与模块地图
Main Func 把 Genspark 定位为面向知识工作者的一体化 AI 工作空间,而不是单点工具。产品面已经从早期搜索和深度研究工作流,快速扩展成一个模块家族,包含 AI Slides、AI Docs、AI Sheets、GenMail、Workflows、AgentBase、SecondBrain、Call For Me、Skills,以及作为总入口的 Super Agent 体验。每个模块都对应真实客户任务:制作演示、起草文档、分析电子表格、分流邮件、构建周期性自动化、生成轻量业务系统,或从连接数据中取回个性化上下文。这种广度是产品策略的核心。Main Func 不是让用户把多个 SaaS 产品串起来,而是试图成为“一个提示词变成成品工作”的地方。最重要的尽调含义是,公司交付的不是一个功能,而是一个运行层加一套用例界面;做好了会强复利,做不好会在运营上变得很乱。[CE001, CE002, CE003, CE004, CE005, CE006]
| 模块 / 资产 | 用户 | 状态 / 成熟度 | 差异化 | 尽调缺口 |
|---|---|---|---|---|
| Super Agent | 通用知识工作者 | 核心平台,持续扩展 | 将一个提示转成跨多种输出的多步成品 | 需要客观成功率和完成度基准 |
| AI Slides | 运营、分析师、GTM、管理层 | 已有文档的生产功能界面 | 对话式生成演示文稿,配有代码支撑的图表和风格技能 | 需要按演示文稿类型拆分的错误率 / 事实核查表现 |
| AI Docs | 知识工作者和报告作者 | 已有文档的生产功能界面 | 生成富文本和 markdown 文档,并提供导出路径 | 需要证明企业采用不止停留在演示 |
| AI Sheets | 分析师、财务、运营 | 已有文档的生产功能界面 | 电子表格智能体,能采集、清洗、分析和可视化数据 | 需要可审计性和可复现性指标 |
| GenMail | 邮件负荷高的专业人士 | 已有文档的生产功能界面 | 统一 Gmail/Outlook 客户端,学习写作语气并分流邮件 | 需要自动草稿的数据治理和准确性指标 |
| Workflows | 运营、GTM、支持、财务 | 已有文档的生产功能界面 | 无代码周期性自动化,带触发器和测试运行 | 需要生产规模完成率和失败率数据 |
| SecondBrain | 个人专业人士和团队 | 已有文档的生产功能界面 | 跨邮件、文件、会议和应用的个人记忆 / 上下文层 | 需要明确同步延迟、索引准确性和访问日志 |
| AgentBase | 搭建轻量系统的运营人员 | 较新的可见模块 | 把提示转成仪表盘、CRM、追踪器和业务系统 | 需要证明耐久性不止模板生成 |
| Call For Me | 需要现实世界电话操作的用户 | 已有文档的生产功能界面 | 实时外呼智能体,带转录和定时任务 | 需要按地区验证可靠性、合规和同意控制 |
| Skills | 高阶用户和团队 | 已有文档的生产功能界面 | 可复用专家工作流,用于保持一致性和风格 | 需要证明复用技能能显著提高输出质量 |
模块集合足够宽,已经像套件一样运转,但公开证据多数仍是文档多、基准少。
[CE001, CE002, CE003, CE004, CE005, CE006]| 用户任务 | 当前工作流 | 公司方案 | 可衡量收益 | 限制 |
|---|---|---|---|---|
| 生成可直接上会的董事会演示文稿 | 手工研究 + 起草 + 设计交接 | AI Slides 研究、搭结构、设计并导出演示文稿 | 更快产出演示文稿,且结果可编辑 | 事实质量和叙事判断仍需验证 |
| 起草一份成熟文档 | 从聊天复制 / 粘贴到文档 | AI Docs 直接生成并编辑富格式文档 | 初稿更干净,迭代更容易 | 需要来源锚定和版本控制证据 |
| 分析电子表格式数据 | 手工整理数据或使用分析师工具 | AI Sheets 汇总、清洗、查询并可视化数据 | 缩短到达分析结论的时间 | 敏感用例需要可复现性 / 审计轨迹 |
| 分流收件箱并安排回复 | 手工查看收件箱和起草 | GenMail 总结、分类、起草并定时发送 | 节省重复沟通工作的时间 | 语气建模和自主性可能引发信任问题 |
| 运行周期性业务自动化 | 类 Zapier 工具或手工周期性任务 | Workflows 用自然语言构建触发器和动作 | 把杂活转成可重复自动化 | 连接器深度和错误处理需要测试 |
| 跨个人工作上下文搜索 | 分别搜索多个应用和文件 | SecondBrain 统一已连接来源的记忆 | 个性化和检索更好 | 敏感数据集中让隐私风险更重 |
可衡量收益大多来自公司声称或用户示例;独立基准仍然稀缺。
[CE003, CE005, CE006, CE007, CE008, CE009]Genspark 可见架构把意图捕获、编排、模型与工具、上下文记忆,以及完成型工作输出界面叠在一起。
[CE012, CE013, CE014, CE015, CE016, CE017]5.2 架构与运营模型
公开技术证据指向以编排为核心的架构。Mainfunc.ai 称 Genspark 使用 Super Agent 处理信息,借助 mixture-of-agents 系统编排 30 多个 AI 模型。OpenAI 的案例研究描述了一个无代码自主助手,可协调九个专业大语言模型和 80 多个工具;Anthropic 则称后来的 Super Agent 编排超过 150 个工具,并在工具选择循环和代码生成中都使用 Claude。因此,公司看起来在运营一个分层运行时架构:顶部是用户意图,中间是规划器 / 路由器,下方是异构前沿模型和自研工具,生成的工件和动作之上再有一组输出界面。SecondBrain 随后充当记忆层,给 Super Agent 更个性化的上下文;Skills 和 AgentBase 则打包可重复的执行模式。这个架构不同于单模型聊天,但也直接依赖模型供应商、连接器和持续评估纪律。[CE012, CE013, CE014, CE015, CE016, CE017]
| 层级 / 组件 | 角色 | 依赖 | 风险 |
|---|---|---|---|
| 用户意图 / 提示界面 | 捕捉目标和期望输出 | 前端 UX 和产品界面 | 模糊提示会抬高执行方差 |
| Super Agent 规划循环 | 选择工具、模型和下一步 | 模型推理质量和编排逻辑 | 循环、误路由或过早停止 |
| 模型路由 / 多智能体混合 | 将子任务分配给最合适的模型 | 第三方模型访问和经济性 | 供应商集中和价格变化 |
| 专用工具 / 自研工具包 | 执行具体任务和转换 | 内部工具质量和连接器健康度 | 工具失效会打断工作流 |
| SecondBrain / 记忆层 | 提供个性化上下文和检索 | 来源同步、索引、权限 | 隐私风险和过期上下文 |
| 输出界面 | 返回幻灯片、文档、表格、邮件草稿、电话、仪表盘 | 渲染器、代码生成、导出路径 | 输出打磨度会掩盖推理错误 |
| 评估 / 安全 / 治理层 | 约束模型使用和企业行为 | 政策、日志、限制、人工审核 | 弱护栏会削弱企业信任 |
这套栈是根据产品文档和合作伙伴案例研究重构出来的,不是官方架构图。
[CE012, CE013, CE014, CE015, CE016, CE017]Main Func 的产品流程从用户意图开始,加入上下文和文件,经 Super Agent 路由,返回可编辑的完成型工作或经常性自动化。
[CE003, CE004, CE005, CE006, CE007, CE008]产品依赖前沿模型供应商、连接器生态、企业身份与治理,以及自身评估纪律。
[CE013, CE014, CE015, CE017, CE021, CE023]5.3 部署、集成、可靠性与控制
部署叙事比公司年龄暗示的更成熟。Team 和 Enterprise 计划文档为企业买家描述了集中计费、SAML SSO、连接器控制、用户分析、AI 模型限制、会话日志、数据驻留选项、定制 DPA、专用 VPC,以及 99.9% uptime SLA 承诺。工作流文档暴露测试运行、待确认项、运行历史和一次一个触发器逻辑,这些都说明系统为受监督执行而设计,而不是盲目自治。AI Slides 暴露模型质量层级、供应商限制、引导式和直接生成模式、文件导入,以及代码驱动图表。GenMail 显示可直接连接邮件供应商和学习型 “Email Brain”;SecondBrain 要求明确授权,并允许用户随时断开来源。因此,产品把高自主性和可见控制面结合起来——这是赢得企业信任所必需的设计选择。尚未解决的问题是,这些控制只是写在文档里的功能,还是已经在大规模生产环境中被深度证明的运营纪律。[CE023, CE024, CE025, CE026, CE027, CE028]
| 控制 / 认证 / 质量指标 | 状态 | 范围 | 缺口 |
|---|---|---|---|
| SOC 2 Type II | 已认证 | 商业产品和企业信任基础 | 未公开披露审计细节或例外事项 |
| ISO 27001 | 已认证 | 信息安全管理 | 未独立描述运营控制 |
| ISO 42001 | 进行中 | AI 治理态势 | 尚未完成 / 范围不清 |
| GDPR | 进行中 / 声称合规 | EU 数据处理合规态势 | 公开证据不如正式认证详尽 |
| 零训练 / 零数据留存 | 面向企业客户的承诺 | 敏感企业使用和模型处理 | 需要合同语言和技术执行证明 |
| SAML SSO、会话日志、模型限制 | 已有文档的企业控制 | 身份、分析、治理 | 需要来自规模化客户部署的证据 |
| 99.9% 可用性 SLA 和 4 小时关键支持响应 | 已有文档的企业承诺 | 企业支持和可靠性 | 需要公开可用性历史或事故数据 |
公司在这个阶段已有异常具体的书面控制,但第三方验证只覆盖列出的认证,未覆盖运营结果。
[CE023, CE024, CE025, CE026, CE027, CE028]| 日期 / 阶段 | 功能 / 里程碑 | 状态 | 含义 | 来源 |
|---|---|---|---|---|
| 2024 年发布期 | 搜索和深度研究产品 | 历史阶段 | 说明公司先从信息综合起步,再走向执行 | OpenAI / Anthropic / TechCrunch |
| 2025 年初转向 | Super Agent 架构重写 | 已完成 | 标志着从僵硬工作流转向自适应智能体循环 | OpenAI / Anthropic |
| Late 2025 | AI Workspace 正式发布 | 已完成 | 提出面向企业的成品交付定位 | LG / PR Newswire |
| Jan 2026 | AI Workspace 2.0,包含 Speakly、AI Inbox 2.0、媒体智能体 | 已完成 | 显示发布节奏快,模态支持更宽 | Business Wire |
| Jun-Jul 2026 | 强调与 OpenAI、Anthropic、Microsoft 的合作 | 已完成 | 确认平台依赖,也显示合作伙伴杠杆 | TechTimes / TMCnet |
| Jul 2026 | Workspace 6.0,包含 SecondBrain、GenMail、GenTeam、AgentBase | 已完成 | 把产品推向记忆更深的工作流软件 | TMCnet / FinancialContent |
路线图看起来推进很快,但高频发布也抬高了质量保证压力。
[CE013, CE014, CE015, CE018, CE029, CE035]核心模块看起来已经上线且文档充分;企业成熟度取决于文档中的控制在生产环境里能否撑住。
[CE001, CE003, CE004, CE005, CE006, CE007]5.4 差异化、依赖与技术判断
从技术上看,Main Func 的差异化来自三件事合在一起:输出成品工作而不是聊天;跨多个模型和工具编排,而不是押注单一模型家族;以及个人或企业上下文层,让智能体随时间变得更有用。这些都是有意义的强项。但它们单独并不是护城河。OpenAI、Anthropic、Microsoft 和其他前沿模型供应商越来越多地开放让这些工作流成为可能的基础构件;竞争平台如今也公开谈编排;产品仍依赖第三方模型质量、API 经济性和连接器可靠性。就连创始人在 Anthropic 案例研究中也承认,“没人真的还有护城河了”,执行速度最重要。因此,技术判断是正面但有条件的:架构可信,产品面丰富,企业控制叙事强于典型早期 AI 创业公司——但公司仍需证明,它的编排层交付成品结果的可靠性,强于买家用同一套底层模型生态在别处自行拼出的方案。[CE035, CE036, CE037, CE038, CE039, CE040]
5.5 图表摘录
06客户
6.1 客户分层与买方结构
Main Func 并不是只卖给一个整齐的企业细分市场。公开材料显示,它的客户基础分层明显:个人知识工作者、小团队自助购买席位,以及转向受治理企业合同的大型组织。买家、用户和付费方之间的关系会随着层级变化。个人用户可以先独自使用免费或付费计划;团队管理员随后为 2 到 150 人的群组集中计费和管理席位;当组织超过约 151 名用户,或需要定制治理、数据驻留、DPA 或支持时,企业买家再进入谈判条款。这种包装很重要,因为它说明 Main Func 可以通过自下而上的试验进入,同时仍提供正式推广所需的采购控制。2026 年发布的协作模块——Hub、Teams、GenTeam、Meeting Notes 和共享项目——进一步表明,公司正试图把单用户新鲜感转化为多用户系统采用,嵌入全球组织范围内反复发生的日常工作。[CU001, CU002, CU003, CU004, CU005, CU006]
| 客户分层 | 买方 / 用户 / 付款方 | 主要用例 | 规模信号 | 收入 / 战略价值 | 缺口 |
|---|---|---|---|---|---|
| 个人自助用户 | 买方与用户通常是同一人;付款方是个人持卡人,也可能是免费用户 | 研究、幻灯片、文档、通话、轻量自动化 | Free / Plus / Pro 套餐与自助式产品入口 | 推动病毒式发现和种子用户采用 | 未公开转化率、留存或 ARPU |
| 小型组织化团队 | 经理或管理员购买;知识工作者使用;公司付款 | 共享生产力、项目协作、团队 AI 使用 | 面向 2-150 个席位的 Team 计划,按自助价格销售 | 商业使用的第一层结构化变现 | 未公开平均团队规模或扩张率数据 |
| 大型企业账户 | IT、运营或业务负责人购买;多名员工使用;公司按合同付款 | 受治理的部署、SSO、数据驻留、连接器、支持 | 面向 151+ 用户的 Enterprise 计划和定制条款 | ACV 更高,扩张潜力更深 | 未公开具名大型企业客户簿或合同规模 |
| 特定职能专业用户 | 部门领导推动;分析师、营销人员、运营人员、高管使用 | 演示文稿制作、数据分析、会议记录、工作流自动化 | 公开案例覆盖咨询、广告、房地产和高管用例 | 将 TAM 扩展到多个白领职能 | 除个案外,垂直行业组合仍不清晰 |
从套餐和产品界面能看出客户分层,但各分层收入结构未披露。
[CU001, CU003, CU004, CU005, CU006]| 指标 | 数值 | 日期 | 来源 | 置信度 | 含义 | 缺失的分母 |
|---|---|---|---|---|---|---|
| 使用 Team / Enterprise 计划的商业组织 | 1,000+ | 2026-01-28 | Business Wire | 中 | 显示工作空间发布后,企业早期采用很快 | 各组织付费席位数未知 |
| 服务的商业客户 | 7,000+ | 2026-07-21 | TMCnet / FinancialContent 镜像 | 中 | 暗示到 2026 年中已触达大量商业客户 | 「客户」定义未知,活跃与非活跃占比未知 |
| Super Agent 发布后的 ARR | 45 天内 $36M | 2025-04 | OpenAI 案例研究 | 中 | 表明早期用户带来强变现拉力 | 企业与消费者使用各自贡献占比未知 |
| 商业化推进后的 ARR | $100M ARR | 2026-01-28 | Business Wire | 中 | 表明商业套餐很快转化为可观收入 | ARR 背后的留存和合同质量未知 |
| ARR / 年化运行收入规模 | $250M 年化运行收入 | 2026-07-21 | TMCnet / FinancialContent / Anthropic | 中 | 暗示到 2026 年 7 月,已落地的客户工作负载规模较大 | 各来源对 ARR 和年化运行收入的口径并不完全一致 |
增长信号很强,但所有客户数量指标都缺少活跃席位、单账户支出和续约分母。
[CU010, CU011, CU012, CU013, CU014]| 界面 | 客户能做什么 | 对买方的意义 | 扩张含义 |
|---|---|---|---|
| Hub | 在持久工作空间内共享文件、对话历史、指令和项目 | 支持团队知识复用和受治理的上下文共享 | 持续项目积累后,切换成本上升 |
| Teams | 私信、群聊、项目共享和跨组织联系人请求 | 让协作内生于产品,而不是依赖外部工具 | 帮助把个人使用转化为组织使用 |
| GenTeam | 把人和 AI 智能体放进带持久记忆的频道、线程、任务和私信 | 面向协作团队,把工作流嵌得更深 | 为每个账户创造更大的席位和工作负载机会 |
| AI Meeting Notes | 记录会议、自动加入线上会议,并与参会者共享笔记 | 把产品推入反复发生的运营场景 | 可带来重复使用,并触达更多利益相关方 |
| Custom Agent / Design / AgentBase 生态 | 构建可复用的智能体和输出,可在不同工作中调用或共享 | 鼓励部门专业化和内部复用 | 提高初次落地后横向扩张的概率 |
这些界面不能直接证明留存,但能看出 Main Func 在按重复、多用户部署来设计产品,而不是只服务一次性提示词使用。
[CU007, CU008, CU009, CU023, CU024]Main Func 的理想转化路径从个人试验走向团队协作,再进入受治理的企业部署,最后扩展到更广的 AI 工作层使用。
[CU001, CU003, CU007, CU008, CU023]可见产品和定价结构鼓励自下而上的漏斗,但要转化成大型、持久账户,还需要自上而下的控制。
[CU003, CU004, CU010, CU021, CU024]6.2 采用轨迹与具名客户证明
最可见的采用指标偏营销口径,但方向上仍有意义。Business Wire 在 2026 年 1 月称,咨询、广告等行业已有 1,000 多家组织开始使用 Team 和 Enterprise 计划。到 2026 年 7 月,TMCnet 和一篇镜像的 FinancialContent 文章称 Genspark 服务超过 7,000 个商业客户。支撑这些头部数字的具名客户证明要窄得多:公开引用指向 Spyglaz AI、GEOPARK 和 ADK Marketing Solutions;合作伙伴案例研究只增加了匿名例子,比如一位 New York 房地产分析师或日本海鲜行业 CEO。这种不平衡意味着 Main Func 确实有采用故事,但公开证明仍偏向公司挑选的轶事,而不是一份经过深度独立验证的参考客户名单。[CU010, CU011, CU012, CU013, CU014, CU015]
| 客户 | 分层 | 部署 / 用例 | 生产使用 / 试点 | 结果 | 局限 |
|---|---|---|---|---|---|
| Spyglaz AI | 小企业 / 初创公司 | 用 Genspark 在 25 分钟内生成 50 页幻灯片;创始人称产品上市和项目交付都提速 | 看起来是真实使用,但证据是客户证言式引用,不是独立验证的生产使用证明 | 创始人正面评价输出质量和速度 | 只有单一官方证言;无合同规模或持续使用数据 |
| GEOPARK | 企业 / 大型组织 | CIO 称,用户先是想做出更好的演示文稿,随后在更广泛测试多智能体后申请企业访问权限 | 显示使用从试用推进到企业协议 | 表明内部用户拉动,并从工具测试升级为企业采购 | 未披露部署规模、续约或 ROI 细节 |
| ADK Marketing Solutions | 大型代理商 / 日本扩张账户 | 数据分析和文档生成工作流 | 公告称过去几个月一直在活跃使用 | 据公司公告,相关工作量约减少 80% | 证据来自发布公告,而非独立案例研究 |
| 匿名房地产分析师 / 日本海鲜企业 CEO | 专业服务和高管用户 | 商业演示稿制作和需求 / 销售线索分析 | 合作伙伴案例研究中有真实用户轶事,但未披露具名 logo | 展示跨行业适用性和提速收益 | 证据是轶事且匿名,引用质量较低 |
这是部分公开样本,不代表完整客户名单。公开证据明显弱于客户数量标题式主张。
[CU015, CU016, CU017, CU018, CU019]| 指标 | 数值 / null | 分层 | 置信度 | 尽调要求 |
|---|---|---|---|---|
| NRR | 团队 / 企业 | 低 | 按分层和同期群提供季度 NRR | |
| GRR / logo 流失 | 团队 / 企业 | 低 | 按同期群和合同规模提供流失与续约数据 | |
| 平均合同期限 | 企业 | 低 | 提供标准订单表期限长度和续约机制 | |
| 席位扩张率 | 团队 / 企业 | 低 | 展示现有账户内净新增席位的时间序列 | |
| 产品满意度信号 | 正面但偏轶事 | 混合 | 低-中 | 提供 NPS / CSAT / 支持工单量趋势及方法说明 |
公开证据没有披露耐久性指标,客户质量尽调仍不完整。
[CU021, CU022, CU030]公开客户证据最强的是用例广度,最弱的是留存和部署深度的独立证明。
[CU015, CU016, CU017, CU018, CU019, CU030]6.3 留存、扩张与集中度能见度
公开资料讲获客和协作远多于讲耐久性。公司没有披露 NRR、GRR、流失、同期群留存、logo 留存、平均合同期限或头部客户集中度。不过,如果平台成功落地,产品面确实暗示了扩张路径。Shared Hubs 在项目之间保留上下文;Teams 支持组织级消息和项目共享;GenTeam 引入持久的人加智能体频道;Meeting Notes 自动加入会议并跨会议平台共享输出;管理员工具暴露席位管理、用量分析、连接器控制,以及新增席位或额外积分包等增购杠杆。这些功能构成了一个可信的先落地再扩张动作,但还不能证明经济性耐久。因此,尽调结论是混合的:扩张机制可见,留存质量和集中度风险仍大多不透明。实际看,买家显然能看到一个团队采用后产品如何扩散,但外部投资人仍无法判断这种扩散来自健康续约、广泛席位增长,还是由新 AI 功能驱动的短促试验。缺失的耐久层证据,是公开资料中最核心的客户风险。[CU021, CU022, CU023, CU024, CU025, CU026]
| 扩张驱动因素 | 集中度风险 | 影响 | 尽调路径 |
|---|---|---|---|
| 从自助 Team 计划扩展到更大范围部署的席位增长 | 如果企业扩张贡献大部分收入,少数大客户可能主导 ARR | 中高 | 索取前 20 大客户集中度和席位增长瀑布图 |
| 重度使用团队内的积分包和工作负载增长 | 收入可能依赖少数高频用户或计算密集型账户 | 中 | 按客户、用户和模态审查使用集中度 |
| Hub、Teams、GenTeam 等协作界面 | 共享上下文采用可能形成锁定,也可能停留在浅层或试验阶段 | 中 | 要求提供周活跃团队、共享项目数和多用户留存 |
| SSO、连接器、日志、支持等企业治理功能 | 企业销售周期长,可能拖慢转化并利好既有套件厂商 | 高 | 审查销售管线转化率和安全审查通过率 |
| 向日本、欧洲和亚洲扩张 | 本地支持投入未必转化为持久的本地收入密度 | 中 | 索取按地区划分的 ARR 和流失率,以及本地支持使用率 |
可见的扩张回路说得通,但集中度和续约风险在公开信息中大多没有量化。
[CU023, CU024, CU025, CU026, CU027, CU028]6.4 客户质量判断
Main Func 看起来比多数同类公司更快地从消费者式 AI 好奇心,跨进了真实商业使用;产品也越来越体现组织型客户需要的东西:集中计费、协作、共享记忆、任务管理和跨渠道智能体部署。这是正面逻辑。负面逻辑是,公开证据仍没有显示这些客户是否深度部署、以高比例续约、盈利性扩张,或集中在少数大账户中。截至 2026-07-29,客户章节支持的结论是:广度真实,但深度验证不完整。客户动能看起来可信;客户质量仍需要直接尽调材料。最好的解读是,Main Func 已经赢得被认真看作企业导向平台的资格,但它还没有提供足够公开证据,让外部读者区分持久企业采用、快速移动的 AI 热情,或由发布驱动的临时好奇心。[CU030, CU031, CU032, CU033, CU034, CU035]
6.5 图表摘录
07风险
7.1 监管、隐私与法律暴露
Main Func 所处的品类里,法律预期上升的速度快过公司年龄增长。EU AI Act 的透明度规则将于 2026 年 8 月生效,欧盟委员会概览明确指出,AI 系统部署方需要知情的人机交互、监督、监控和事件处理;部分高风险系统随后还会面临严格得多的义务。California 隐私法已经赋予用户知情、删除、更正和选择退出数据共享的权利;Genspark 自己的隐私政策称,当用户在请求中包含个人数据并调用相关能力时,用户提供的提示内容可能被发送给 OpenAI 和 Anthropic。Main Func 的条款还否认对内容准确性、质量和用户依赖负责,并把 IP 和合规责任分配回用户。这在 AI 里并不罕见,但会制造真实的企业摩擦:产品越多处理会议转录、邮件上下文、CRM 数据和对外动作,法律审查就越会从通用 SaaS 采购转向 AI 治理审查。[CR001, CR002, CR003, CR004, CR005, CR006]
| 规则 / 案件 | 管辖区 | 状态 | 可能性 | 严重性 | 缓释措施 | 剩余风险敞口 | 尽调路径 |
|---|---|---|---|---|---|---|---|
| EU AI Act 透明度和 GPAI 义务 | EU | 透明度规则于 2026 年 8 月开始适用;GPAI 治理已生效;部分高风险义务稍后分阶段落地 | 高 | 高 | 产品控制、披露、日志和人工监督设计 | 中高,因为智能体工作流可能滑向敏感用例 | 将产品功能和企业部署映射到第 50 条和 GPAI 义务 |
| 加州隐私 / CCPA 合规 | 加州 / 美国 | 持续适用的隐私权制度,覆盖访问、删除、更正和选择退出义务 | 高 | 高 | 隐私政策、同意流程、内部请求处理、处理方管理 | 中高,因为产品摄取邮件、会议和提示词内容 | 审查删除工作流、请求 SLA 以及处理方 / 分处理方清单 |
| FTC 对误导性 AI 宣传和输出完整性的执法 | 美国 | 围绕误导性 AI 宣传和虚假输出,执法与指南都很活跃 | 中高 | 高 | 准确性免责声明、营销纪律、人工审核、安全控制 | 中高,因为 Main Func 主打高自主度的输出交付 | 审查营销审核流程、客户准确性投诉和宣传主张的证据支撑 |
| IP 和输出权利纠纷 | 多司法辖区 | 面向用户的条款把大量责任放在客户身上,并限制公司保证 | 中 | 中高 | 条款、赔偿责任定位、内容过滤、用户教育 | 中,因为生成输出和客户输入可能触及第三方权利 | 审查 IP 投诉、下架历史和企业赔偿补充协议 |
这些风险按其对企业采用和尽调摩擦的近期可能影响排序,而不是按理论最高法定罚款排序。
[CR001, CR002, CR003, CR004, CR005, CR006]Main Func 最高的剩余风险集中在自主性、敏感数据、供应商依赖和既有厂商分发交叠的位置。
[CR004, CR012, CR022, CR028, CR031, CR039]7.2 运营、质量与安全风险
从运营看,Main Func 正在把自主性卖进一些小错误也会造成放大后果的工作流。产品可以起草外部沟通、拨打实时电话、总结会议、从邮件和日历历史中拉取信息,并跨许多工具和模型路由工作。风险不只是幻觉,还包括执行失败、上下文过期、连接器损坏、token 过期、权限静默漂移,以及越界进入用户可能没有充分监督的动作。公开文档显示团队意识到了这些问题。Team 和 Enterprise 计划默认不用于模型训练;管理员不能查看成员项目内容;GenTeam 将类似创建者权限的外部动作限制在创建者本人,并要求对难以逆转的任务明确批准;企业文档也暴露日志、SSO 和连接器控制。但同一批文档也承认运营失败模式:必需 SSO 推广配置错误可能把整个组织锁在外面;连接器设置按组织全局生效,而不是按用户生效;Meeting Notes 集成会在 token 过期时中断。产品看起来有周到的监测和控制设计,但公开证据仍没有展示事件率、误动作率或经客户审计的可靠性。[CR011, CR012, CR013, CR014, CR015, CR016]
| 失效模式 | 可能性 | 严重性 | 缓释成熟度 | 剩余风险敞口 | 未解决缺口 |
|---|---|---|---|---|---|
| 高风险工作流中出现幻觉或低可信输出 | 中高 | 高 | 中等 | 高 | 未公开任务完成准确率或错误行动率基准 |
| 电话、邮件或工作流步骤等自主行动执行错误 | 中 | 高 | 中等 | 中高 | 需要行动日志、批准率和回滚证据 |
| SecondBrain、会议笔记、收件箱、CRM 和共享项目中的敏感数据集中 | 高 | 高 | 中等 | 高 | 未公开事故、访问审计或渗透测试摘要 |
| 身份 / 管理员配置错误,包括 SSO 锁定 | 中 | 中高 | 中等 | 中 | 需要管理员变更保护和支持运行手册 |
| 连接器或令牌故障导致上下文过期或自动化中断 | 高 | 中高 | 中等 | 中高 | 需要同步延迟、连接器健康度和故障恢复指标 |
| 音频 / 会议采集失败,或录制流程中同意处理出错 | 中 | 中高 | 中等 | 中 | 需要地区同意政策、失败率和投诉数据 |
产品展示了有意义的控制项,但剩下的问题是这些控制是否真正深入运营,而不只是文档写得好。
[CR011, CR012, CR013, CR014, CR015, CR016]准确性、权限或可靠性一旦失守,可能很快传导到客户信任、收入质量、法律风险和融资假设。
[CR011, CR014, CR015, CR019, CR030, CR040]7.3 供应商、分发与财务模型风险
公司的商业上行空间和依赖风险绑在一起。公开材料称,Genspark 编排 30 多到 70 多个模型,核心推理和语音能力依赖 OpenAI 与 Anthropic,并接入 Slack、Notion、Salesforce、HubSpot、Google Workspace、Microsoft 365 等企业系统。Anthropic 的商业条款明确说明,若法律、攻击、成本或上游供应商问题造成干扰,服务可以被暂停;OpenAI 的服务条款也把 beta 功能和部分输出赔偿责任划在保障之外。因此,Main Func 承受的不只是供应商涨价,还包括政策变化、宕机、模型质量回退,以及合同边界收紧。财务上,积分体系叠加宽广的多模态产品线,意味着使用量增长会同时推高收入和交付成本。在毛利率透明的成熟 SaaS 平台里,这还可控;但在这里,客户集中度、流失率和利润率仍未公开,风险真实存在。与此同时,Salesforce、Microsoft、Workday、ServiceNow 和 UiPath 都在既有企业系统内部推进智能体式自动化,Main Func 必须一边跑赢供应商依赖,一边跑赢掌握分发入口的既有巨头。[CR022, CR023, CR024, CR025, CR026, CR027]
| 依赖 | 交易对手 | 角色 | 集中度 | 失效场景 | 严重性 | 缓释措施 | 剩余风险敞口 |
|---|---|---|---|---|---|---|---|
| 前沿推理和语音模型 | OpenAI / Anthropic | 核心模型能力、实时语音、编码、工具选择 | 高 | 定价、质量、政策或可用性变化会拉低产品质量或利润率 | 高 | 多模型编排与路由 | 高:核心客户价值仍取决于外部模型供应商 |
| 业务系统连接器 | Google、Microsoft、Slack、Notion、Salesforce、HubSpot 等 | 数据访问、记忆、工作流上下文和执行 | 高 | API 变化、宕机、权限被撤销或企业安全拦截会削弱效用 | 高 | 连接器覆盖面广,管理员控制齐全 | 中高:连接器健康度是差异化上下文的基础 |
| 云 / 服务合同条款 | OpenAI / Anthropic 及其依赖的任何上游供应商 | 约束服务暂停、赔偿范围和支持义务 | 中高 | 服务暂停、赔偿缩窄或 beta 除外条款会造成客户或平台中断 | 高 | 合同谈判与备用路由 | 中高:关键保护写在 Main Func 无法控制的第三方合同里 |
| 企业分销渠道 | Salesforce、Microsoft、Workday、ServiceNow、UiPath 等既有厂商 | 争夺工作流预算,且已掌握系统记录上下文 | 高 | 既有厂商把同类 agent 打包进既有支出和信任锚点 | 高 | 更快的 UX 创新和更广的输出范围 | 高:分销重力可能压过功能质量 |
| 高增长 AI 的资本市场支持 | 成长型投资者 / 私募市场胃口 | 如果利润率落后于增长,资本继续支撑扩张 | 中 | Main Func 尚未证明经济性可持续前,AI 市场情绪先降温 | 中高 | 大额现金融资和强劲的公开 ARR 轨迹 | 中:当前资本通道强,但估值敏感性仍高 |
供应商集中度在这里不只关系到可用性,也关系到谈判筹码和毛利率韧性。
[CR022, CR023, CR024, CR025, CR026, CR027]Main Func 同时依赖模型供应商、连接器生态、合规预期和企业上下文所有者。
[CR022, CR023, CR024, CR025, CR026, CR028]7.4 团队、执行,以及推翻投资假设的触发点
剩下的风险是极高速增长下的执行质量。OpenAI 曾描述,一个 20 人团队推动了早期超级智能体增长;Anthropic 后来称团队约有 50 名工程师,内部还相信,除了执行速度之外,“nobody really has a moat anymore”。速度可以是优势,也可能遮住支持体系不足、控制薄弱,或创始人和小核心工程团队的关键人集中。客户口径也比客户透明度扩张得更快:2026 年 1 月称有 1,000 多家组织,2026 年 7 月称有 7,000 多个商业客户,但没有公开 NRR、GRR、事故记录、集中度表或续约指标。投资含义很直接:如果 Main Func 能证明企业部署可靠、毛利率可控、客户足够分散,风险画像会很快收敛。否则,增长可能比头部动能暗示的更浅。正确的尽调姿态不是因为公司有创业风险就否定它,而是在按激进估值倍数承销之前,围绕可靠性、客户质量和供应商依赖设定明确的否决标准。[CR033, CR034, CR035, CR036, CR037, CR038]
| 角色 / 职能 | 依赖或缺口 | 发生概率 | 严重性 | 缓释措施 | 尽调路径 |
|---|---|---|---|---|---|
| 创始人 / 核心产品负责人 | 品类战略、推进节奏和产品判断看起来都与创始人高度绑定 | 中 | 高 | 加深梯队储备并建立正式产品治理 | 审查继任梯队深度、决策权和关键人风险 |
| 工程组织 | 产品范围很宽,团队相对野心仍小,必须持续顶住 | 高 | 高 | AI 辅助编码杠杆和招聘增长 | 审查组织架构、事故责任归属和值班容量 |
| 客户成功 / 支持 | 账户快速增长可能跑在企业导入和支持质量前面 | 中高 | 中高 | 日本支持团队建设和企业 CSM 模型 | 审查支持人力、积压工单、升级处理和续约阻碍 |
| 安全 / 合规运营 | 公开控制承诺可能高于当前内部运营成熟度 | 中 | 高 | 认证和管理员工具 | 审查审计、渗透测试和例外处理流程 |
问题不在团队是否有才华,而在治理是否跟上了产品广度和增长速度。
[CR033, CR034, CR035, CR036]| 风险 | 可监控触发因素 | 阈值 / 事件 | 行动含义 |
|---|---|---|---|
| 可靠性风险 | 企业账户的工作流失败率显著上升 | 反复发生关键事故,或在缺少站得住脚的控制措施时出现错误动作 | 暂停估值测算,直到审计数据和补救措施证明改善 |
| 客户质量风险 | 留存偏弱或 ARR 集中 | NRR 低于优秀企业软件常模,或头部客户集中度过高 | 将增长重估为低质量,并下调倍数假设 |
| 供应商依赖风险 | 模型供应商成本或政策冲击 | 供应商变化带来利润率挤压或产品退化 | 压测下行情景现金消耗和护城河韧性 |
| 监管 / 隐私风险 | 执法问询、高严重性隐私事件或不合规 AI 披露流程 | 正式调查或企业安全审查反复失败 | 推迟投资,或要求补救作为交割前提条件 |
| 分销风险 | 企业订单明显转向既有套件 | 竞争失利源于系统记录打包,而不是功能差距 | 收窄目标市场,或修正终局份额假设 |
| 执行风险 | 支持、导入或安全运营在增长中退化 | 积压升级、宕机增多或实施摩擦上升 | 将超高速增长视为运营成本不可持续 |
叫停标准聚焦能打破“快速增长转化为可持续企业软件”这一论点的证据。
[CR037, CR038, CR039, CR040]7.5 图表
08估值
8.1 投资假设与反面假设
核心投资假设是,Main Func 可能在搭建最早一批宽口径 AI 工作层,而且能按企业级速度变现。按公开说法,公司在 2024 年还处在搜索产品阶段,2025 年推出 Super Agent,2026 年推进到 AI Workspace 6.0,同时宣称 ARR、客户数和产品覆盖面都快速扩张。只要其中相当一部分增长具备持续性,公司就不该和普通 SaaS 均值相比,而应进入高溢价的 AI 原生软件叙事。反面假设是,公开记录看起来仍更像动能证明,而不是质量证明。流失率、NRR、毛利率、现金余额、优先权结构和客户集中度都没有披露。产品宽度令人印象深刻,也同步抬高了可靠性、供应商和治理风险。因此,投委会更合适的框架是有条件乐观:公司强到不能轻易放弃,信息不透明到不能随意承销,发展速度又快到不能只靠过时的 2025 年估值锚来判断。这个细节很重要,因为正确的投资者行为不是二元化信念,而是分阶段信念:兴趣足够强,值得继续推进;怀疑也足够强,不能接受只靠叙事定价;同时还要承认,隐藏的客户分群或条款数据仍可能让价值向任一方向大幅移动。正因为存在这种不对称,纪律严明的长期投资者可能靠耐心创造比速度更多的价值。[CV001, CV002, CV003, CV004, CV005, CV006]
| 建议 | 置信度 | 风险评级 | 估值立场 | 决策含义 |
|---|---|---|---|---|
| 继续研究 / 选择性推进 | 中 | 高 | 对价格敏感;最新公开估值上跳并非显然错误,但仅凭公开证据还撑不起稳妥的投资测算 | 仅在私有尽调确认留存、利润率和条款干净后继续 |
建议刻意设置为条件式,因为证据质量落后于增长质量。
[CV022, CV023, CV024, CV025]| 论点 | 改变判断的证据 |
|---|---|
| 拥有真实客户采用的超高速增长 AI 工作空间,可以支撑溢价软件定价 | 证明 $250M 规模 run-rate 可持续、足够多元且续约良好 |
| 产品广度和成品交付导向,能撑起可信的平台叙事 | 证明可靠性和治理在企业生产环境中顶得住 |
| 多模型编排和上下文层可以支撑差异化工作流 | 证明供应商或既有厂商能以更低摩擦复制该价值 |
| 公开证据仍缺少留存、利润率和股权结构透明度 | 干净的 cohort 数据、利润率桥和优先权披露会显著提高投资测算可信度 |
| 最新估值可能已折现近期成功的大部分 | 更低进入价格或异常投资者友好的条款,会迅速改善回报测算 |
这是一张动态论点表,不是静态结论;每行都点出能改变价格纪律的证据。
[CV001, CV004, CV007, CV018, CV024]投资建议从快速增长证据出发,再经过估值支撑和未解的质量缺口,落到对价格敏感的「继续研究」立场。
[CV001, CV011, CV015, CV022, CV024, CV025]投委会式记分卡,判断在未知私募价格下是否具备可投性。
[CV001, CV004, CV008, CV018, CV022, CV023]8.2 融资背景、公开可比公司的护栏与入场纪律
公开定价背景在短时间内大幅上移。2025 年 11 月 Series B 轮的锚点是 $275M、估值 $1.25B。2026 年 1 月的公开材料把 $100M ARR 和增补后的 $300M Series B 轮放在一起。到 2026 年中,公开报道指向超过 $645M 的总融资、$2.6B 估值,以及约 $250M ARR 或年化运行收入。如果这些收入口径扎实,最新估值对应低双位数收入倍数——相对宽口径横向 SaaS 均值偏贵,但对高速增长的 AI 原生平台并不荒唐。如果这些口径没那么可持续,价格很快就难以自圆其说。公开可比公司护栏的作用就在这里。Salesforce、Workday、UiPath 等成熟公开平台披露收入质量、现金流和经营杠杆,Main Func 还没有;分析师市场数据也显示,2026 年软件市场只给少数赢家溢价倍数。实际结论是,入场纪律比叙事兴奋更重要。尤其是,公司最新一次公开估值上台阶发生时,市场仍看不到续约质量、贡献毛利,或自助式使用与受治理企业合同之间的精确经济结构。[CV011, CV012, CV013, CV014, CV015, CV016]
| 情景 | 假设 | 估值 / 回报逻辑 | 关键风险 | 概率信号 |
|---|---|---|---|---|
| 乐观 | $300M+ 可持续 ARR / run-rate,企业留存强,尽管有模型成本仍保持高毛利率,优先权包袱有限,品类领导地位守住 | 若进入价接近或低于最新公开标记,支撑 $3.5B-$5.0B 股权价值和有吸引力的上行空间 | 竞争打包、利润率漏损或治理事故会封顶上行空间 | 私有指标必须明显好于公开证据缺口所暗示的水平 |
| 基准 | ~$250M 可持续 ARR / run-rate,留存不错但不顶尖,毛利率仍低于成熟 SaaS,增长仍强但放缓,条款标准 | 支撑约 $2.2B-$3.0B 股权价值;仍有上行,但对倍数更敏感 | 客户质量不透明和供应商依赖让折现率维持高位 | 最符合当前公开证据 |
| 悲观 | 收入质量弱于声称,利润率负担被证实很重,企业端耐久性未证实,或下一轮定价把完美预期计入过多 | 估值压缩至约 $1.4B-$2.0B;若以 $2.6B+ 进入,上行空间有限 | 留存、竞争或执行失望触发倍数重置 | 鉴于私有数据缺失,这一情景仍有可能 |
区间是作者情景,锚定公开融资背景、软件可比公司护栏和公司披露的增长声明。
[CV011, CV012, CV013, CV014, CV015, CV026]| 可比对象 | 指标 | 倍数 / 估值 / 状态 | 参考意义 | 局限 |
|---|---|---|---|---|
| Main Func 公开标记 | 到 2026 年 7 月约 $250M ARR 或年化 run rate | 隐含估值约 $2.6B;按 $250M 约 ~10x,按 $100M ARR 约 ~26x | 当前直接定价锚 | 收入质量和具体条款仍未公开 |
| UiPath | FY2026 ARR $1.853B;GAAP 毛利率 85% | 成熟公开市场自动化平台,增长较慢但披露好得多 | 治理和企业执行预期的最佳公开自动化可比公司 | 不像 Main Func 那样 AI 原生,且有公开市场成熟度折价 |
| Workday | FY2026 收入 $9.552B;non-GAAP 运营利润率 29.6%;客户 11,500+ | 可信企业平台和系统记录可比对象 | 可作为信任、披露和企业耐久性预期的上限 | 规模更大、嵌入更深,超过 Main Func |
| Salesforce | FY2026 Q4 订阅与支持收入 $10.7B;当前 RPO $35.1B;FY26 FCF $14.4B | 说明公开市场如何奖励具备现金流和待履约订单可见性的耐久企业软件 | 良好的信任与分销可比对象 | 过于成熟且多元,不能直接做倍数可比 |
| 公开横向 SaaS 市场数据 | Multiples.vc 2026 年 7 月横向 SaaS 平均值 / 参考点 | 约 6.1x EV / NTM revenue,离散度由增长和 AI 定位驱动 | 防止为普通软件增长支付过高价格的有用护栏 | 市场数据篮子,不是公司特定可比对象 |
| BVP Nasdaq Emerging Cloud Index 指数 | 新兴公开云软件指数 | 跟踪公开云软件 cohort;可作为情绪和相对倍数背景 | 成长软件的公开市场情绪指标 | 指数层面数据,非针对 Main Func 具体商业模式 |
这张表是估值框架,不是声称任何单一可比对象都完全匹配 Main Func。
[CV011, CV012, CV013, CV014, CV015, CV016]Main Func 的隐含倍数取决于把哪个公开收入锚点视为可持续,差异会很大。
[CV011, CV012, CV016, CV017]区间图展示公开定价背景,以及作者设定的悲观 / 基准 / 乐观估值带。
[CV011, CV012, CV013, CV026, CV027, CV028]8.3 建议、区间与尽调关口
公开证据支持的建议是继续研究 / 只在价格和条款有纪律时推进。若估值接近最新公开的 $2.6B 标记,只要私下尽调能证明企业留存强、毛利率可控、优先权包袱有限,公司仍可能值得投资。若没有这些证据却大幅抬价,风险调整后回报逻辑会迅速恶化。因此,基础情形不是“买故事”,而是“只有隐藏指标配得上故事时才买”。牛市情形需要真实的企业级耐久性和持续的 AI 品类领导力。熊市情形不是破产,而是在增长质量、利润率现实或竞争压力更清晰后,估值倍数向更宽的软件可比公司压缩。对投资者来说,最重要的问题不是 Main Func 是否令人印象深刻,而是下一轮相对仍未知的事实是否给出足够安全边际。换句话说,本章并不认定公司在绝对意义上估值过高;结论是,估值精度远弱于公司带来的兴奋感,因此,价格、优先权和下行保护上的纪律,本身就是投资假设的一部分。[CV022, CV023, CV024, CV025, CV026, CV027]
| 触发因素 | 阈值 | 对论点的传导 | 行动含义 |
|---|---|---|---|
| 可持续 ARR / 客户声明未通过尽调 | cohort、使用量或收入确认无法支撑标题增长声明 | 打破核心溢价增长叙事 | 按悲观情景重切估值,或放弃 |
| 毛利率显著低于预期 | 模型、媒体或支持成本指向结构性偏弱的经济性 | 限制软件式扩展性,并压缩应得倍数 | 要求更低价格,或直接放弃 |
| 客户质量集中或续约偏弱 | NRR / GRR 不达溢价软件预期,或头部客户集中度不符合预期 | 把广泛采用故事变成脆弱收入故事 | 立即切到悲观情景 |
| 重大信任 / 治理事故 | 安全、隐私或错误动作事件击穿企业信任 | 推高折现率,并放慢企业端落地 | 推迟或拒绝投资 |
| 下一轮价格在没有更强证据时超过公开逻辑 | 条款意味着在缺少新尽调支撑时显著高于最新标记价格买入 | 消除安全边际 | 除非私有数据显著强于公开证据,否则放弃 |
叫停触发因素设计成能在尽调中验证,而不是只依赖叙事解读。
[CV024, CV026, CV027, CV029]| 主题 | 缺失证据 | 重要性 | 负责人 / 尽调路径 |
|---|---|---|---|
| 股权结构和优先权 | 完全稀释后所有权、优先级结构、清算优先权和反稀释条款 | 决定表面上行空间是否真正可投资 | 要求提供股权结构表、历史 term sheet 和 waterfall 模型 |
| 收入质量 | ARR 定义、cohort 留存、流失、扩张,以及自助服务与企业客户组合 | 区分可持续软件价值和短期使用峰值 | 要求提供 cohort 表和财务包 |
| 利润率桥 | 按模态拆分的毛利率、模型供应商支出、支持负担和单位经济性 | 决定规模是否转化为软件式盈利能力 | 要求提供交付成本和供应商支出分析 |
| 客户集中度 | Top-20 客户敞口、ACV 分布、合同期限和续约状态 | 衡量快速客户增长声明背后的脆弱性 | 要求提供销售运营和财务导出数据 |
| 可靠性 / 信任指标 | 事故历史、错误动作率、工作流完成率、企业安全审查结果 | 广泛的自主工作平台成败取决于信任 | 要求提供事故日志、基准测试结果和审计摘要 |
| 轮次价格与证据 | 2026 年 6 月后的任何私有标记、二级交易信号、409A 或条款演变 | 锚定当前进入价格是否仍能提供目标回报 | 要求提供融资时间线和任何近期交易材料 |
这些要求刻意聚焦高杠杆点;每一项都可能实质性影响价格或建议。
[CV020, CV021, CV024, CV025, CV028, CV030]8.4 图表
免责声明
本报告基于截至 2026-07-29 的公开信息生成,仅用于尽调研究,不构成投资建议。任何关于融资、估值、客户质量或合同的结论,都应以一手尽调材料核验。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | Main Func is the Palo Alto-based corporate parent behind the Genspark product suite. | 高 | SO001, SO015 |
| CO002 | MainFunc Inc. provides the services while Genspark Inc. is described as a wholly owned subsidiary in the terms of service. | 中 | SO003 |
| CO003 | Public company and third-party profiles consistently place Main Func’s founding year in 2023. | 高 | SO001, SO015, SO025 |
| CO004 | Main Func publicly says it is based in Palo Alto with offices in Singapore and Tokyo, while third-party profiles also describe broader Japan operations. | 中 | SO001, SO015 |
| CO005 | Main Func’s privacy and terms pages define the services as the genspark.ai search and content-generation service plus the company websites at mainfunc.ai. | 高 | SO002, SO003 |
| CO006 | The legal pages describe the core service as productivity-oriented search and content generation rather than a single narrow workflow tool. | 高 | SO002, SO003 |
| CO007 | Genspark launched publicly in June 2024 as an AI-powered search engine built around Sparkpages. | 高 | SO010, SO018, SO015 |
| CO008 | Genspark’s Product Hunt launch on June 18, 2024 placed fourth for the day with 469 upvotes and 168 comments. | 中 | SO018 |
| CO009 | OpenAI’s case study says user demand evolved from search summaries toward finished outcomes such as decks, scripts, and follow-up emails by late 2024. | 中 | SO006 |
| CO010 | By April 2025 Main Func had pivoted from AI search into an agentic workspace organized around the Super Agent. | 高 | SO006, SO007, SO022 |
| CO011 | The current product narrative is autonomous task completion for knowledge workers, with outputs spanning slides, sheets, docs, web apps, and AI phone calls. | 高 | SO001, SO005, SO006 |
| CO012 | The November 2025 workspace launch materials said Genspark orchestrated 30+ AI models, 150+ in-house tools, and 20+ premium datasets. | 高 | SO008, SO009, SO017 |
| CO013 | Later 2026 company and partner materials describe the platform as orchestrating more than 70 AI models. | 中 | SO012, SO019, SO020, SO022 |
| CO014 | Public reporting supports an approximately $60 million seed round in 2024 led by Lanchi Ventures. | 高 | SO009, SO010, SO015 |
| CO015 | Public profiles also support a $100 million Series A in early 2025, although exact month references vary across secondary sources. | 中 | SO011, SO015 |
| CO016 | On November 20, 2025 Genspark announced a $275 million Series B financing round at a $1.25 billion post-money valuation. | 高 | SO008, SO009, SO017 |
| CO017 | The named November 2025 Series B investors included Emergence Capital, SBI Investment, LG Technology Ventures, Pavilion Capital, and UpHonest Capital. | 高 | SO008, SO009, SO016 |
| CO018 | The November 2025 Series B materials said Genspark exceeded $50 million in annualized run rate within five months of the workspace launch. | 高 | SO008, SO009 |
| CO019 | By March 2026 Genspark had extended Series B to $385 million and was being valued at roughly $1.6 billion. | 中 | SO016, SO012 |
| CO020 | A June 2026 Series B extension added $100 million, lifted total funding above $645 million, and reset valuation to $2.6 billion. | 中 | SO021, SO019, SO025 |
| CO021 | July 2026 launch materials say Genspark serves more than 7,000 business clients. | 中 | SO019, SO020 |
| CO022 | July 2026 launch materials say Genspark reached $250 million in annualized run rate within 12 months. | 中 | SO019, SO020 |
| CO023 | Anthropic’s customer case study says Genspark surpassed $250 million ARR after pivoting to the Super Agent in early 2025. | 中 | SO007 |
| CO024 | OpenAI’s customer case study says the Super Agent reached $36 million ARR in 45 days after launch. | 中 | SO006 |
| CO025 | Main Func’s homepage says Genspark is trusted by millions of users worldwide. | 中 | SO001 |
| CO026 | The privacy policy says prompts and outputs may be processed by third-party AI providers including OpenAI and Anthropic when users invoke those features. | 中 | SO002 |
| CO027 | The business page says Genspark is SOC 2 Type II and ISO 27001 certified, with ISO 42001 and GDPR work still in progress. | 中 | SO005 |
| CO028 | Public founder biographies consistently describe CEO Eric Jing as a former Microsoft Bing leader who later ran major AI and search businesses at Baidu and Xiaodu. | 高 | SO008, SO010, SO015 |
| CO029 | Public founder biographies consistently describe CTO Kay Zhu as a former Google search-ranking technologist who later worked at Baidu and Xiaodu. | 高 | SO008, SO010, SO015 |
| CO030 | The November 2025 financing releases present COO Wen Sang as an MIT PhD and prior founder of Smarking, a YC- and Khosla-backed enterprise software company. | 高 | SO008, SO009 |
| CO031 | Emergence Capital publicly framed Genspark as an enterprise AI workspace delivering autonomous execution rather than mere assistance. | 高 | SO008, SO009 |
| CO032 | TechCrunch’s June 2024 review found the original search product still faced accuracy, ethics, publisher-traffic, and unresolved business-model questions. | 中 | SO010 |
| CO033 | The terms of service say Main Func does not guarantee the completeness, accuracy, or currency of Genspark content and places data-compliance responsibility on team and enterprise customers. | 中 | SO003 |
| CO034 | Tech Times argues Main Func’s orchestration thesis depends on continued access to model providers that are also building competing agent products. | 中 | SO022 |
| CO035 | Anthropic’s case study says Genspark’s Super Agent coordinates more than 150 specialized tools and that the company operates with roughly 50 engineers who write code through AI tools. | 中 | SO007 |
| CO036 | TechCrunch described Genspark as a small roughly 20-person Singapore- and Bay Area-based team in June 2024. | 中 | SO010 |
| CO037 | FundedIQ listed MainFunc at 51-100 employees in June 2026. | 低 | SO013 |
| CO038 | Silicon Valley Investclub estimated Genspark at roughly 143 employees by April 2026. | 低 | SO012 |
| CO039 | By June 2026 Main Func’s public strategy rested on an orchestration layer spanning OpenAI, Anthropic, Microsoft, and other frontier-model or cloud partners rather than on a proprietary foundation model. | 高 | SO001, SO006, SO007, SO022 |
| CO040 | Workspace 6.0 expanded the company from one-off artifact generation toward memory-rich enterprise software with SecondBrain, GenMail, GenTeam, and AgentBase. | 中 | SO019, SO020 |
| CM001 | Google defines agentic AI as systems that understand a goal, make a plan, and take actions across applications with human guidance and oversight. | 中 | SM001 |
| CM002 | Google says 2026 marks a behavioral shift from instruction-based computing toward intent-based computing for employees using agents. | 中 | SM001 |
| CM003 | Google reports that 52% of executives in gen-AI-using organizations already have AI agents in production. | 中 | SM001 |
| CM004 | Among organizations using agents, Google says 49% deploy them for customer service, 46% for marketing or security operations, 45% for tech support, and 43% for product innovation or productivity and research. | 中 | SM001 |
| CM005 | Anthropic’s 2026 survey says 57% of organizations deploy agents for multi-stage workflows and 16% already run cross-functional processes across multiple teams. | 中 | SM002 |
| CM006 | Anthropic’s survey says 81% of organizations plan to tackle more complex agent use cases in 2026, including multi-step and cross-functional processes. | 中 | SM002 |
| CM007 | Anthropic reports that 80% of organizations already see measurable economic returns from AI agent investments. | 中 | SM002 |
| CM008 | Anthropic identifies data analysis and report generation as a 60% high-impact use case, internal process automation at 48%, and research and reporting as a top area for next-year expansion. | 中 | SM002 |
| CM009 | The Business Research Company values the broad AI-in-workplace market at $421.09 billion in 2026 and defines it to include hardware, software, and services across many industries. | 中 | SM003 |
| CM010 | The broad AI-in-workplace category spans IT, BFSI, healthcare, retail, manufacturing, and other verticals, making it wider than Main Func’s direct software opportunity. | 中 | SM003, SM005 |
| CM011 | The Business Research Company defines AI productivity tools to include virtual assistants, document management, RPA, business intelligence, content creation, code assistance, project management, collaboration, and scheduling tools. | 中 | SM004 |
| CM012 | Data Bridge values AI knowledge work automation at $5.27 billion in 2025 and $14.86 billion by 2033 and defines the category to cover workflow management, enterprise search, customer support, content generation, and data analysis. | 中 | SM007 |
| CM013 | Intel Market Research projects AI knowledge automation at $14.3 billion in 2026 and describes it as a market spanning knowledge graphs, intelligent document processing, virtual assistants, and predictive analytics. | 中 | SM008 |
| CM014 | Mordor Intelligence values the standalone agentic AI market at $9.89 billion in 2026 with a 42.14% CAGR through 2031. | 中 | SM009 |
| CM015 | Mordor says large enterprises held 65.05% of agentic AI market share in 2025, cloud deployments held 59.72%, BFSI led with 19.12% share, and multi-agent systems held 53.30% share. | 中 | SM009 |
| CM016 | Gartner predicts that 40% of enterprise applications will feature task-specific AI agents by the end of 2026, up from less than 5% in 2025. | 中 | SM010 |
| CM017 | Gartner also predicts that by 2027 one-third of agentic AI implementations will combine agents with different skills to manage complex tasks inside applications and data environments. | 中 | SM010 |
| CM018 | Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027 because of escalating costs, unclear business value, or inadequate risk controls. | 中 | SM011 |
| CM019 | Gartner says the market is full of “agent washing” and estimates only about 130 of the thousands of agentic AI vendors are real. | 中 | SM011 |
| CM020 | Workday argues that bolted-on AI tools often fail to deliver enterprise-grade accuracy because they sit outside the data, compliance context, and business rules of core systems. | 中 | SM023 |
| CM021 | Workday’s Sana launch says Sana Self-Service Agent launches with 300+ skills and acts on HR and finance workflows using Workday’s existing security, permissions, and audit framework. | 高 | SM022, SM023 |
| CM022 | Workday says Sana Enterprise connects Workday with systems such as Gmail, Outlook, Salesforce, ServiceNow, SharePoint, Slack, and Zoom so agents can complete work across applications. | 高 | SM023, SM024 |
| CM023 | Salesforce positions Agentforce around customer service, contact center, field service, employee service, sales service, and IT service workflows. | 中 | SM018 |
| CM024 | Salesforce says Agentforce agents are built on CRM and external data and can be extended through Flows, MuleSoft APIs, Apex, and JavaScript. | 高 | SM018, SM019 |
| CM025 | Microsoft Copilot Studio is an end-to-end agent-building platform that lets organizations publish standalone agents or deploy them directly into Microsoft 365 Copilot. | 中 | SM020 |
| CM026 | SAP says Joule Agents automate workflows end to end by combining business-process expertise, SAP Knowledge Graph grounding, and centralized governance. | 中 | SM021 |
| CM027 | ServiceNow says its AI agents operate across IT, customer service, HR, CRM, risk, and application-development workflows and can be coordinated through an AI Agent Orchestrator and AI Control Tower. | 中 | SM025 |
| CM028 | UiPath says 90% of U.S. IT executives have business processes that would be improved by agentic AI, 87% say interoperability is essential or significant, and 52% say agentic AI will enable automation of complex business workflows. | 中 | SM014 |
| CM029 | UiPath positions Maestro as a control plane that orchestrates AI agents, robots, systems, and humans and argues most agents fail to reach production without governance and platform support. | 高 | SM014, SM015 |
| CM030 | Automation Anywhere says enterprise processes span systems such as Salesforce, ServiceNow, SAP, and custom applications and therefore need universal orchestration rather than isolated agents. | 中 | SM016 |
| CM031 | Automation Anywhere says its context graph improved agent accuracy by more than 30% in internal evaluations and that process simulation and governance are required before production deployment. | 中 | SM016 |
| CM032 | Across the analyst sources, the category’s main growth drivers are digital transformation, productivity optimization, workflow automation, and better integration of AI with enterprise platforms. | 中 | SM003, SM007, SM008 |
| CM033 | Main Func’s practical market boundary should exclude hardware, robotics, model-layer infrastructure, and the full AI-in-workplace umbrella because those categories overstate its direct software opportunity. | 中 | SM003, SM005, SM011 |
| CM034 | A defensible Main Func SAM is better framed by the overlap between the standalone agentic AI and knowledge-work automation lenses, implying a low-double-digit-billion software market rather than a $400B-plus umbrella. | 中 | SM007, SM008, SM009 |
| CM035 | The most plausible near-term buyer functions for Main Func are research and reporting, finance operations, customer support, sales enablement, HR, and internal operations rather than every enterprise department at once. | 中 | SM002, SM018, SM022, SM025 |
| CM036 | Anthropic identifies integration with existing systems (46%), data access and quality (42%), and change-management needs (39%) as the top scaling challenges for enterprise agents. | 中 | SM002 |
| CM037 | Mordor and Gartner both highlight governance, compute cost, interoperability, vendor lock-in, and transparency as the constraints that slow category diffusion. | 中 | SM009, SM011, SM013 |
| CM038 | Because agents are being embedded inside systems of record and workflow platforms, distribution power increasingly sits with incumbent enterprise-software vendors rather than with stand-alone model wrappers. | 中 | SM020, SM021, SM023, SM025 |
| CM039 | Multi-agent orchestration is moving toward category table stakes as Google, Gartner, Mordor, UiPath, ServiceNow, and Automation Anywhere all describe teams of agents or orchestrated workflows rather than isolated bots. | 高 | SM001, SM009, SM010, SM014, SM016, SM025 |
| CM040 | The net market picture is strong top-down growth but slower realized production adoption, which means valuation should weight proof of workflow completion and retention more heavily than TAM headlines alone. | 中 | SM002, SM011, SM013 |
| CP001 | Main Func positions Genspark as an AI workspace that gets work finished autonomously rather than only assisting with chat. | 高 | SP001, SP002 |
| CP002 | Genspark for Business advertises 70+ AI models, SOC 2 Type II and ISO 27001 certifications, and a $30-per-user team plan while preserving custom enterprise packaging. | 高 | SP002, SP025 |
| CP003 | UiPath says the platform orchestrates AI agents, robots, systems, and humans from a single control plane called Maestro. | 中 | SP003 |
| CP004 | UiPath explicitly argues that most agents never make it to production without platform-level governance, reliability, and security. | 高 | SP003, SP004 |
| CP005 | Automation Anywhere markets itself as the number-one provider of agentic automation and emphasizes secure, adaptive workflows across industries such as financial services, healthcare, IT, and manufacturing. | 中 | SP005 |
| CP006 | Automation Anywhere’s website and enterprise PR position the category around governed AI agents and a process reasoning layer rather than standalone chatbots. | 高 | SP005, SP006 |
| CP007 | Salesforce positions Agentforce as an AI agent platform for customer service, sales, field service, employee service, and IT service. | 中 | SP007 |
| CP008 | Salesforce exposes both action-based and conversation-based pricing for Agentforce, making it more transparent than many quote-only rivals. | 高 | SP007, SP008 |
| CP009 | Microsoft Copilot Studio is an end-to-end platform to create agents and publish them into Microsoft 365 Copilot, Teams, SharePoint, and other Microsoft surfaces. | 中 | SP009 |
| CP010 | Cognition positions Devin as the first autonomous software engineer, making it a specialist substitute for engineering workflows rather than a broad white-collar workspace. | 中 | SP019 |
| CP011 | 11x positions Alice and Julian as digital workers for SDR and phone-agent workflows, demonstrating a much narrower revenue-team wedge than Main Func. | 中 | SP022 |
| CP012 | Workday says Sana changes the user experience from sidecar copilots to action-taking agents embedded in HR and finance systems, using the same security, permissions, and audit framework customers already trust. | 高 | SP011, SP012 |
| CP013 | Workday says Sana Self-Service Agent launches with 300+ skills and is available through Workday Flex Credits instead of an extra standalone license. | 中 | SP012 |
| CP014 | Workday also says Sana Enterprise extends beyond Workday with connectors including Gmail, Outlook, Salesforce, ServiceNow, SharePoint, Slack, and Zoom, while the Microsoft 365 Copilot integration widens distribution further. | 高 | SP012, SP013 |
| CP015 | ServiceNow says its AI agents work across IT, customer service, HR, CRM, risk, security, and app development and can be coordinated through AI Agent Orchestrator, Fabric, and Control Tower. | 中 | SP014 |
| CP016 | ServiceNow explicitly claims its agents are “built-in, not bolted on,” highlighting the same embedded-context competitive advantage that disadvantages horizontal overlays like Main Func. | 中 | SP014 |
| CP017 | Microsoft says organizations can pay for Copilot Studio through $200 monthly packs of 25,000 Copilot Credits or a pay-as-you-go meter, tying cost directly to usage. | 中 | SP009 |
| CP018 | UiPath publishes survey evidence that 87% of U.S. IT executives see interoperability across AI technologies as essential or significant, reinforcing the importance of orchestration layers. | 中 | SP003 |
| CP019 | UiPath says 52% of respondents believe agentic AI will enable automation of complex business workflows, reflecting how incumbents are moving beyond narrow bot tasks. | 中 | SP003 |
| CP020 | SAP says Joule Agents automate workflows end to end using business-process expertise, SAP Knowledge Graph grounding, and centralized governance. | 中 | SP010 |
| CP021 | IBM positions watsonx Orchestrate as an open, hybrid, secure control plane for connecting agents, workflows, data, and applications across the business. | 中 | SP015 |
| CP022 | Google’s agent platform pricing page makes clear that hyperscaler alternatives can expose transparent token economics, which strengthens the internal-build substitute set for sophisticated buyers. | 中 | SP016 |
| CP023 | Sierra markets itself as a leading conversational AI platform for businesses with built-in guardrails, observability, customer context, and proactive engagement workflows. | 中 | SP017 |
| CP024 | Sierra’s product and fundraising narrative centers on customer experience and support interactions rather than the broad multi-workflow output layer Main Func is pursuing. | 高 | SP017, SP018 |
| CP025 | Ema markets AI employees for HR, IT, payroll, and finance with 1,000+ connectors and a governance-heavy enterprise posture, making it one of the closer horizontal-overlap startups in the set. | 中 | SP020 |
| CP026 | Ema also promotes multi-model fusion, cost efficiency, and on-prem or air-gapped deployment, competing directly on enterprise-control concerns that Main Func must address. | 中 | SP020 |
| CP027 | Clay publishes a free tier and paid plans from $167 per month and $446 per month, offering a much cheaper public starting point for GTM workflow automation than most enterprise agent platforms. | 中 | SP021 |
| CP028 | Clay also bundles actions, data credits, and its web-research agent Claygent, showing how specialist workflow tools can combine agentic behavior with immediately legible ROI. | 中 | SP021 |
| CP029 | Gartner predicts that 40% of enterprise applications will feature task-specific AI agents by the end of 2026, which structurally favors incumbents that can embed agents into existing software estates. | 中 | SP023 |
| CP030 | Main Func is broader than Sierra because Sierra is optimized around customer experiences and support journeys rather than general research, presentation, and multi-output knowledge work. | 中 | SP002, SP017, SP018 |
| CP031 | Main Func is broader than Cognition because Devin is focused on software engineering autonomy rather than general enterprise white-collar automation. | 中 | SP002, SP019 |
| CP032 | Main Func is broader than 11x because 11x concentrates on SDR and phone-based revenue workflows. | 中 | SP002, SP022 |
| CP033 | Main Func is broader than Clay because Clay is best understood as a GTM workflow and data-automation specialist rather than a general-purpose finished-work workspace. | 中 | SP002, SP021 |
| CP034 | Ema is the closest startup overlap among reviewed specialists because it also spans HR, IT, and finance, but its marketing still leans more toward AI employees inside enterprise processes than toward Main Func’s broad output creation layer. | 中 | SP002, SP020 |
| CP035 | Main Func’s public team-plan pricing is simpler than most quote-based enterprise agent platforms, but it does not by itself prove better economics at enterprise scale. | 中 | SP002, SP021 |
| CP036 | A buyer selecting between Main Func and incumbents is often choosing between finished-work breadth and trusted in-app context rather than between better or worse base models. | 中 | SP002, SP011, SP014 |
| CP037 | Opaque enterprise pricing remains a market-wide issue because only a few players, such as Salesforce, Microsoft, Clay, and Genspark’s team plan, expose enough public detail for direct comparison. | 中 | SP002, SP008, SP009, SP021 |
| CP038 | Gartner’s warning that over 40% of agentic AI projects will be canceled by end-2027 implies that competitive durability will favor vendors with measurable ROI and controls, not just feature breadth. | 中 | SP024 |
| CP039 | The strongest switching costs in the current enterprise-agent market come from embedded data context, identity, admin controls, and workflow ownership rather than from model access alone. | 高 | SP011, SP014, SP015, SP023 |
| CP040 | Main Func is vulnerable to multi-homing because teams can use it for exploratory or creative work while keeping customer, HR, ERP, or IT workflows inside incumbent platforms. | 中 | SP007, SP011, SP014, SP020 |
| CP041 | Main Func’s orchestration narrative is no longer unique because UiPath, Automation Anywhere, Workday, ServiceNow, SAP, IBM, and Google all now describe cross-system agent coordination or control planes. | 高 | SP003, SP005, SP010, SP012, SP014, SP015, SP016 |
| CP042 | The most convincing evidence of moat durability for Main Func would be proof that customers standardize multiple white-collar workflows inside Genspark instead of buying one specialist per function or defaulting to incumbent suites. | 中 | SP002, SP024 |
| CI001 | Genspark for Business publicly lists a Team Plan at $30 per user per month with 12,000 credits per seat, centralized billing, SSO/SAML, and connector management. | 中 | SI002 |
| CI002 | The Team Plan includes broad access to premium chat, image, video, and audio models, implying that list pricing bundles heterogeneous compute costs into one seat price. | 中 | SI002 |
| CI003 | Genspark Workflow lets users build scheduled or email-triggered automations across Gmail, Outlook, Google Workspace, Slack, Teams, Salesforce, Stripe, GitHub, and other systems without code. | 中 | SI003 |
| CI004 | The workflow product suggests Main Func monetizes not just discrete answers, but recurring automation and operational execution. | 中 | SI002, SI003 |
| CI005 | Main Func’s public commercial surface implies a mixed model of self-serve subscriptions, team plans, and custom enterprise agreements rather than a pure enterprise-only sales motion. | 中 | SI001, SI002, SI003 |
| CI006 | The privacy and terms pages reinforce an enterprise-ready packaging story through zero-retention language, plan distinctions, and legal controls, even though they do not disclose realized pricing. | 高 | SI004, SI005 |
| CI007 | OpenAI says Super Agent reached $36 million in ARR within 45 days of launch, with the growth achieved by a 20-person team and no paid advertising. | 中 | SI006 |
| CI008 | LG Technology Ventures and PR Newswire both say Genspark closed a $275 million Series B at a $1.25 billion post-money valuation in November 2025 while breaking $50 million in annualized run rate within five months. | 高 | SI008, SI009 |
| CI009 | Business Wire says Genspark surpassed $100 million ARR within nine months and had topped off its Series B to $300 million by January 2026. | 中 | SI010 |
| CI010 | Anthropic’s more-than-$250 million ARR disclosure should be treated as the upper bound of Main Func’s public 2026 revenue narrative rather than as a fully audited precision datapoint. | 中 | SI007, SI013 |
| CI011 | July 2026 coverage says Genspark had reached about $250 million ARR and more than 7,000 business clients. | 中 | SI013, SI020 |
| CI012 | Business Wire says more than 1,000 organizations had begun using the AI Workspace platform by late January 2026. | 中 | SI010 |
| CI013 | The public revenue sequence is directionally consistent even though the checkpoints use a mix of ARR and annualized run rate and come from company-controlled narratives or partners. | 中 | SI006, SI007, SI008, SI009, SI010, SI013 |
| CI014 | OpenAI’s case study says Super Agent orchestrates nine specialized large language models and more than 80 integrated tools, indicating nontrivial third-party model and tooling costs in service delivery. | 中 | SI006 |
| CI015 | Anthropic’s case study says Genspark’s Super Agent uses Claude to coordinate 150+ specialized tools and that roughly 50 engineers produce all company code through AI tools. | 中 | SI007 |
| CI016 | UiPath reported FY2026 GAAP gross margin of 83%, non-GAAP gross margin of 85%, revenue of $1.611 billion, and cash, cash equivalents, and marketable securities of $1.69 billion. | 中 | SI016 |
| CI017 | Workday reported FY2026 revenue of $9.552 billion, subscription revenue of $8.833 billion, operating cash flow of $2.939 billion, and cash plus marketable securities of $5.443 billion. | 高 | SI014, SI015, SI025 |
| CI018 | Salesforce reported FY2026 revenue of $41.5 billion, GAAP operating margin of 20.1%, and operating cash flow of $15.0 billion, while also saying Agentforce ARR reached $800 million. | 中 | SI017, SI024 |
| CI019 | These comparables show that scaled automation and workflow software can produce strong cash generation and high margins, but Main Func is likely earlier and more compute-heavy than those mature peers. | 中 | SI006, SI016, SI017 |
| CI020 | Main Func’s near-term gross margin is likely below mature peers because rich multimodal outputs, included credits, and third-party model orchestration make cost-to-serve variable and potentially heavy. | 中 | SI002, SI006, SI007 |
| CI021 | Because Workflows can be triggered by schedule and email and can act across third-party systems, Main Func has a credible path to expansion revenue through recurring automation, not only through casual prompting. | 中 | SI003 |
| CI022 | Business Wire highlights an 80% reduction in data-analysis and document-creation workloads at ADK Marketing Solutions, offering at least one concrete ROI proof point for enterprise adoption. | 中 | SI010 |
| CI023 | The jump from 1,000-plus organizations in January to 7,000-plus business clients in July implies either very fast enterprise-logo growth, a broader business-client definition, or both. | 中 | SI010, SI013, SI020 |
| CI024 | Public evidence does not disclose churn, gross retention, net retention, or customer concentration, so the durability of the ARR story cannot yet be judged externally. | 中 | SI006, SI007, SI010, SI013 |
| CI025 | Gensparks business page combines high-cost capabilities such as premium image, video, audio, and large-model access inside one bundle, increasing the importance of routing efficiency and supplier discounts. | 中 | SI002 |
| CI026 | Salesforce, Microsoft, and Clay all expose agent-related public pricing signals, which suggests market pricing pressure is moving toward more explicit usage-based economics. | 中 | SI021, SI022, SI023 |
| CI027 | Main Func’s $30 seat anchor looks inexpensive relative to horizontal agent platforms, but it cannot be compared directly without understanding included usage and support costs. | 中 | SI002, SI021, SI022 |
| CI028 | No public source reviewed provides CAC, payback, gross margin, or contribution margin by modality, which is the main reason this chapter cannot move from growth narrative to quality-underwritten financial model. | 中 | SI002, SI006, SI007, SI010, SI013 |
| CI029 | The April 2026 extension was reported at $385 million total Series B financing and about a $1.6 billion valuation. | 中 | SI011 |
| CI030 | The June 2026 extension was reported at $100 million more, bringing total funding above $645 million and valuation to $2.6 billion. | 中 | SI012 |
| CI031 | Business Wire’s January 2026 wording that the company had topped off a $300 million Series B conflicts slightly with the November 2025 $275 million announcement and later $385 million figure. | 中 | SI008, SI009, SI010, SI011 |
| CI032 | The June 2026 SaaS News coverage says the company did not disclose the specific use of funds for the $100 million extension. | 中 | SI012 |
| CI033 | The visible capital base is large for a 2023-founded software company and should be sufficient to fund product development, geographic expansion, model procurement, and enterprise support in the near term. | 中 | SI008, SI011, SI012, SI013 |
| CI034 | At the same time, because no public cash balance or burn disclosure exists for Main Func, investors cannot calculate runway from funding headlines alone. | 中 | SI008, SI010, SI011, SI012 |
| CI035 | Rapid valuation step-ups from $1.25 billion to roughly $1.6 billion and then $2.6 billion raise the performance bar for future growth and margin delivery. | 中 | SI008, SI011, SI012 |
| CI036 | TechCrunch’s 2024 skepticism about Genspark’s search-era business model and publisher dynamics is partly outdated post-pivot, but it remains a reminder that early engagement does not automatically equal durable monetization. | 中 | SI019 |
| CI037 | Gartner’s warning that over 40% of agentic AI projects will be canceled by end-2027 implies Main Func’s forward revenue could be more fragile than current ARR headlines suggest if deployments fail to show durable ROI. | 中 | SI018 |
| CI038 | The company’s financial story currently rests more on velocity metrics, financing support, and partner narratives than on the standard recurring-revenue quality metrics used in late-stage SaaS diligence. | 中 | SI006, SI007, SI010, SI012, SI018 |
| CI039 | A minimum underwriting package should include monthly ARR bridges, cohort retention, customer mix, workload economics, burn, and supplier concentration before investors treat the current valuation as fully de-risked. | 中 | SI018, SI019, SI016, SI017 |
| CI040 | Financially, Main Func looks like a promising but still partially un-underwritable asset: likely strong near-term growth and adequate capital, but insufficient public data on revenue quality, margin path, and runway to support a full conviction model. | 中 | SI010, SI012, SI016, SI017, SI018 |
| CE001 | Main Func publicly positions Genspark as an all-in-one AI workspace where work gets finished autonomously rather than as a narrow assistant. | 高 | SE001, SE024 |
| CE002 | The visible Genspark suite now includes AI Slides, AI Docs, AI Sheets, GenMail, Workflows, AgentBase, SecondBrain, Skills, and Call For Me around the Super Agent core. | 中 | SE002, SE003, SE004, SE005, SE006, SE007, SE008, SE010, SE011, SE012 |
| CE003 | AI Slides is presented as an intelligent presentation agent that researches, structures, designs, and exports complete decks while supporting template following, speaker notes, and direct on-canvas edits. | 中 | SE010 |
| CE004 | AI Slides can run code for calculations and charts, uses professional or creative modes, and exposes quality tiers such as Standard and Ultra. | 中 | SE010 |
| CE005 | AI Docs is positioned as a document creation and editing agent with rich text and markdown support plus export to HTML, Word, and PDF. | 中 | SE004 |
| CE006 | AI Sheets is positioned as an autonomous spreadsheet agent that can gather data, run SQL and statistical analysis, and export editable XLSX outputs. | 中 | SE012 |
| CE007 | GenMail is an AI-powered email and calendar client that connects Gmail and Outlook, categorizes emails, drafts replies in a learned voice, and unifies calendars across accounts. | 中 | SE005 |
| CE008 | SecondBrain is a personal memory system that syncs emails, calendars, files, chats, CRM tools, and productivity suites to give Super Agent deeper personalized context. | 中 | SE007 |
| CE009 | AgentBase turns prompts into dashboards, CRMs, trackers, and other lightweight business systems, extending Genspark from content generation into operational software creation. | 中 | SE006 |
| CE010 | Skills package reusable expert workflows and style logic so users can reuse not just visuals but structured reasoning patterns across repeated work. | 中 | SE008, SE010 |
| CE011 | Call For Me shows that Main Func is pushing beyond digital-only outputs into real phone-based execution with transcripts, scheduling, and task history. | 中 | SE011 |
| CE012 | Mainfunc.ai says Genspark processes information using a Super Agent that orchestrates 30+ AI models through a mixture-of-agents system. | 中 | SE001 |
| CE013 | OpenAI says the April 2025 Super Agent launch orchestrated nine specialized large language models and more than 80 integrated tools. | 中 | SE013 |
| CE014 | Anthropic says the later Super Agent uses Claude to coordinate more than 150 specialized tools inside a single agent. | 中 | SE014 |
| CE015 | LG and PR Newswire describe the AI Workspace launch as orchestrating 30+ leading models, 150+ in-house tools, and 20+ premium datasets. | 高 | SE015, SE017 |
| CE016 | OpenAI describes a dual-layer voice system for Call For Me in which the Realtime API manages live dialogue while a shadow model monitors and guides the interaction. | 中 | SE013 |
| CE017 | Anthropic describes the core Super Agent loop as model-agnostic by design, with Claude deciding which tool to call next, when to gather more information, and when to stop. | 中 | SE014 |
| CE018 | TMCnet and FinancialContent say Workspace 6.0 introduced context-centric modules including SecondBrain, GenMail, GenTeam, and AgentBase. | 中 | SE018, SE019 |
| CE019 | The visible architecture therefore has at least five layers: user-intent surfaces, an orchestration runtime, model and tool substrates, a context/memory layer, and enterprise control wrappers. | 中 | SE001, SE007, SE009, SE013, SE014 |
| CE020 | SecondBrain plus Skills plus Workflow history suggest the product is trying to build compounding context and reuse, not just one-off generation. | 中 | SE003, SE007, SE008 |
| CE021 | The architecture is directly dependent on third-party model vendors and external connectors, which means supplier economics and API reliability are product-critical. | 中 | SE003, SE013, SE014, SE020 |
| CE022 | Main Func’s technical differentiation lies more in orchestration, context, and output assembly than in training a proprietary frontier foundation model. | 中 | SE001, SE013, SE014, SE020 |
| CE023 | The Team and Enterprise documentation says both plans provide centralized billing and admin controls, connector management, and SAML SSO, while preserving independent workspaces per user. | 中 | SE009 |
| CE024 | Enterprise plan documentation includes 25,000 credits per seat, 36-month typical initial terms, 12-month auto-renewals, 99.9% uptime SLA, 4-hour critical response, and dedicated support. | 中 | SE009 |
| CE025 | Enterprise governance controls include agent-level permissions, AI model restrictions, organization-wide credit caps, login and session logs, configurable residency, dedicated VPC, and custom compliance addendums. | 中 | SE009 |
| CE026 | Genspark for Business claims SOC 2 Type II certification and ISO 27001 certification, with ISO 42001 and GDPR work in progress. | 中 | SE002 |
| CE027 | The Team and Enterprise documentation says organizations are opted out of model training by default and admins cannot view individual members’ project content. | 高 | SE009, SE021 |
| CE028 | The privacy and support materials imply the product is prepared for higher-governance buyers, but public evidence does not independently verify how these controls perform under large-scale regulated deployments. | 中 | SE009, SE021, SE022 |
| CE029 | Business Wire says AI Workspace 2.0 added Speakly, AI Inbox 2.0, upgraded Slides, and media agents, evidencing fast release cadence across modalities. | 中 | SE025 |
| CE030 | Workflow documentation includes test runs, pending confirmation states, and run histories, which shows the product has explicit supervised-execution surfaces rather than only black-box autonomy. | 中 | SE003 |
| CE031 | AI Slides exposes Guide Mode, Professional versus Creative modes, and Standard versus Ultra quality tiers, which suggests the company is productizing trade-offs between speed, cost, and output quality. | 中 | SE010 |
| CE032 | GenMail’s Email Brain learns from past messages and edits, which can improve personalization but also raises a higher trust threshold for users delegating communication work. | 中 | SE005 |
| CE033 | SecondBrain requires explicit authorization for connected sources and lets users disconnect them, which is an important privacy control for a memory-heavy product. | 中 | SE007 |
| CE034 | AI model vendor restrictions in enterprise settings show that Main Func expects some customers to constrain which providers can be used for compliance or policy reasons. | 中 | SE009, SE010 |
| CE035 | The product has moved quickly from search and deep research into a broader workspace with memory, email, dashboards, and voice, indicating unusually high surface-area expansion in roughly two years. | 中 | SE016, SE018, SE019, SE025 |
| CE036 | Main Func’s best technical claim is that it turns chat into finished outcomes by combining planning, tool use, retrieval, code generation, and editable output surfaces in one runtime. | 中 | SE001, SE010, SE012, SE013, SE014 |
| CE037 | The July 2026 context-centric release pushes the product closer to an operating system for knowledge work rather than a one-off generative app. | 中 | SE018, SE019, SE020 |
| CE038 | TechTimes and the partner case studies make clear that Main Func is deeply dependent on OpenAI, Anthropic, and other model providers for core capabilities. | 中 | SE013, SE014, SE020 |
| CE039 | Anthropic’s case study quotes Kay Zhu saying “nobody really has a moat anymore” and that execution speed is the moat, which is an unusually candid statement about technology commoditization risk. | 中 | SE014 |
| CE040 | The overall technology verdict is positive on product ambition and visible execution depth, but conditional on whether the orchestration layer remains meaningfully better than what buyers can assemble from the same model ecosystem. | 中 | SE014, SE018, SE020 |
| CU001 | By July 2026 Main Func publicly positioned Genspark as serving both individual users and enterprise clients worldwide. | 高 | SU003, SU007 |
| CU002 | The company’s visible customer footprint includes US headquarters plus Singapore and Tokyo offices, while Business Wire said it was expanding support across North America, Europe, and Asia and formally into Japan. | 高 | SU002, SU007 |
| CU003 | Team plan is self-serve and aimed at groups of 2 to 150 people with centralized billing, seats, connectors, and other admin controls. | 高 | SU006, SU017 |
| CU004 | Enterprise plan is aimed at organizations with 151-plus users or buyers needing custom contract terms, data residency, DPA, dedicated support, and advanced governance. | 中 | SU006 |
| CU005 | The buyer-user-payer model differs by segment, with self-serve individuals entering directly while team admins or enterprise sponsors increasingly become the commercial buyer inside larger organizations. | 中 | SU001, SU006, SU017 |
| CU006 | Public customer examples span consulting, advertising, startup, real-estate-analysis, and executive workflows rather than one narrow industry vertical. | 中 | SU001, SU002, SU005 |
| CU007 | Hub gives teams shared files, shared conversation history, shared instructions, and project continuity inside a persistent workspace. | 中 | SU008 |
| CU008 | Teams adds direct messaging, group chat, project sharing, and cross-organization contact requests, making collaboration native inside Genspark instead of purely external. | 中 | SU009 |
| CU009 | GenTeam extends the collaboration layer into persistent human-plus-agent channels, tasks, threads, and DMs, indicating a push toward deeper organizational workflow embedding. | 中 | SU003, SU010 |
| CU010 | Business Wire said that by late January 2026 more than 1,000 organizations across consulting, advertising, and other industries had begun using Genspark’s Team and Enterprise plans. | 中 | SU002 |
| CU011 | By July 2026 TMCnet and a mirrored FinancialContent article said Genspark served more than 7,000 business clients. | 中 | SU003, SU004 |
| CU012 | OpenAI said Super Agent reached $36M ARR in 45 days, indicating strong early monetization pull but not distinguishing business from consumer usage. | 中 | SU018 |
| CU013 | Business Wire said Genspark crossed $100M ARR by January 2026, which suggests business packaging translated into meaningful revenue quickly. | 中 | SU002 |
| CU014 | TMCnet, FinancialContent, and Anthropic all pointed to roughly $250M run-rate scale by July 2026, implying heavy customer workload but still limited public detail on customer quality. | 中 | SU003, SU004, SU005 |
| CU015 | The Genspark for Business page includes a testimonial from Spyglaz AI founder Neeraja Rasmussen describing a 50-page slide deck produced in 25 minutes and broader acceleration in time to market and project delivery. | 中 | SU001 |
| CU016 | The same business page attributes to GEOPARK’s CIO a progression from searching for better AI presentations to users requesting enterprise access after broader platform testing. | 中 | SU001 |
| CU017 | Business Wire cited ADK Marketing Solutions in Japan as achieving roughly an 80% reduction in data-analysis and document-creation workloads over the prior few months. | 中 | SU002 |
| CU018 | Anthropic’s case study adds anonymized customer anecdotes, including a New York real estate analyst producing investor decks faster and Japanese seafood CEOs using Genspark for demand analysis and lead generation. | 中 | SU005 |
| CU019 | The named-customer evidence base is much smaller than the headline customer-count claims, so public proof remains concentrated in testimonials and selected anecdotes. | 中 | SU001, SU002, SU003, SU005 |
| CU020 | TechCrunch’s early 2024 coverage underscored that Main Func’s public narrative initially emphasized product ambition more than a clearly proven business model, which remains relevant when judging customer-quality transparency. | 中 | SU019 |
| CU021 | Main Func publicly discloses no NRR, GRR, logo churn, renewal rate, average contract term, or cohort-retention data. | 中 | SU001, SU002, SU003, SU006 |
| CU022 | Public evidence also does not reveal customer concentration, top-account revenue share, or contract-size distribution. | 中 | SU001, SU002, SU003, SU005 |
| CU023 | Meeting Notes auto-joins supported meeting platforms, records recurring meetings, and can automatically share notes with participants, which creates a clear repeat-usage surface. | 中 | SU011 |
| CU024 | Admin features such as seat management, user analytics, usage logs, SSO, login history, connector controls, and credit-pack assignment create visible mechanisms for rollout and expansion inside business accounts. | 中 | SU006 |
| CU025 | Hub’s shared context and GenTeam’s persistent channel memory suggest Main Func is intentionally designing for multi-project continuity rather than one-off prompt sessions. | 中 | SU008, SU010 |
| CU026 | The credit system and per-seat allocations imply that revenue can expand not only by adding logos or seats but also by driving heavier workload intensity among existing users. | 中 | SU006, SU012, SU017 |
| CU027 | Cross-organization messaging and share-link workflows indicate Main Func is comfortable with collaborative adoption that may start inside one organization and spread to adjacent stakeholders. | 中 | SU009, SU016 |
| CU028 | Enterprise packaging also implies procurement friction, because larger buyers are asked to evaluate residency, support, SSO, contract terms, and governance before broad rollout. | 中 | SU006, SU023, SU024 |
| CU029 | Geographic expansion into Japan with local support resources shows willingness to localize customer success, but public data does not show whether this translates into durable regional revenue density. | 中 | SU002 |
| CU030 | The strongest customer evidence today is breadth of deployment surfaces and a handful of recognizable proof points, not independently verified retention metrics. | 中 | SU001, SU002, SU006, SU008, SU010 |
| CU031 | Main Func appears to have progressed from consumer-style AI curiosity into real business usage faster than most software startups of similar age. | 中 | SU002, SU003, SU005, SU018 |
| CU032 | The company’s product arc toward shared memory, teams, and persistent agents is consistent with a strategy to become a default work layer rather than a point AI tool. | 中 | SU003, SU008, SU010, SU013 |
| CU033 | Publicly available evidence is not strong enough to prove that customer deployments are deeply embedded, multi-year, or diversified across many large accounts. | 中 | SU001, SU002, SU003, SU021, SU022 |
| CU034 | Main Func’s customer chapter therefore supports a positive breadth verdict but only a provisional durability verdict. | 中 | SU021, SU022, SU002 |
| CU035 | As of 2026-07-29, direct diligence should prioritize cohort retention, top-customer concentration, enterprise deployment depth, and the share of revenue coming from true organizational customers versus self-serve usage. | 中 | SU021, SU022, SU002 |
| CR001 | The EU AI Act’s transparency rules take effect in August 2026 and require humans to be informed when they are interacting with AI systems such as chatbots. | 高 | SR020, SR021 |
| CR002 | The EU AI Act also frames higher-risk AI around human oversight, robustness, cybersecurity, accuracy, post-market monitoring, and incident reporting. | 中 | SR020 |
| CR003 | General-purpose AI governance already applies under the EU AI Act timeline before some later high-risk obligations phase in. | 中 | SR020, SR021 |
| CR004 | California privacy law grants rights to know, delete, opt out of sale or sharing, correct inaccurate personal information, and limit the use of sensitive personal information. | 高 | SR024, SR031 |
| CR005 | Genspark’s privacy policy says the company may share personal data included in prompts with third-party AI service providers including OpenAI and Anthropic solely to generate responses. | 中 | SR008 |
| CR006 | Genspark’s terms say users cannot rely on service content for accuracy and that the company disclaims warranties as to completeness, accuracy, and currency. | 中 | SR007 |
| CR007 | Genspark’s terms also place responsibility on users for third-party IP claims connected to their use of generated or received content. | 中 | SR007 |
| CR008 | The FTC’s AI enforcement page highlights cases involving deceptive AI-generated reviews and misleading AI business-opportunity claims, showing that aggressive AI marketing and low-integrity output can attract enforcement scrutiny. | 中 | SR023 |
| CR009 | NIST’s AI Risk Management Framework provides a widely recognized voluntary benchmark for governing and measuring AI risk, raising the diligence bar for enterprise platforms even when not legally binding. | 中 | SR022 |
| CR010 | Together, the public legal record implies that Main Func’s enterprise sales process will increasingly require AI-governance review rather than standard SaaS procurement alone. | 中 | SR020, SR024, SR007, SR008 |
| CR011 | Main Func’s product surface now spans phone calls, meeting capture, email and calendar context, workflow automation, and persistent memory, increasing the number of ways autonomous errors can create real customer harm. | 中 | SR001, SR004, SR005, SR009, SR027 |
| CR012 | Team and Enterprise plan documentation says members are automatically opted out of model training and that admins cannot view individual members’ project content. | 高 | SR002, SR003 |
| CR013 | Team and Enterprise admins can still control billing, seats, connectors, SSO, usage analytics, logs, and login history, which is helpful for governance but also expands the admin-control surface that must be operated correctly. | 中 | SR003 |
| CR014 | Team and Enterprise docs warn that requiring SSO for all members can lock out an entire organization if the configuration is wrong because there is no in-app fallback. | 中 | SR003 |
| CR015 | Connector settings currently apply org-wide rather than per member, which can create blunt governance tradeoffs for mixed-sensitivity organizations. | 中 | SR003 |
| CR016 | SecondBrain concentrates email, calendar, file, chat, CRM, and meeting context in one personal memory layer, which increases both usefulness and the blast radius of any permission or privacy failure. | 中 | SR004 |
| CR017 | SecondBrain documentation says connected integrations require explicit authorization and can be disconnected at any time, which is a real mitigation but not proof of flawless access governance. | 中 | SR004 |
| CR018 | Meeting Notes auto-joins supported online meetings via connected calendar links, uses credits billed by recording time, and can fail when calendar access tokens expire or recording/upload issues occur. | 中 | SR005 |
| CR019 | Meeting Notes says Genspark does not save or provide access to original audio files, which narrows one privacy surface but can also limit downstream forensic review if a recording dispute arises. | 中 | SR005 |
| CR020 | GenTeam says agents that act as their creator can send email, post socially, message other platforms, or make phone calls only when the creator asks, and hard-to-reverse actions require explicit approval. | 中 | SR006 |
| CR021 | Public materials still do not disclose incident rates, false-action rates, workflow completion rates, or independent reliability benchmarks for autonomous execution. | 中 | SR002, SR003, SR009, SR010, SR014 |
| CR022 | Main Func’s public positioning depends materially on outside model vendors, with mainfunc.ai, OpenAI, and Anthropic all describing Genspark as an orchestration layer over multiple frontier models and tools. | 中 | SR001, SR009, SR010, SR027 |
| CR023 | OpenAI’s service terms make beta services available as-is and explicitly disclaim warranties that they will be generally available, uninterrupted, error-free, or secure. | 中 | SR025 |
| CR024 | OpenAI’s service terms also show that output indemnity has important exclusions, especially where users ignore safety features, mix outputs with third-party systems, or use third-party offerings. | 中 | SR025 |
| CR025 | Anthropic’s commercial terms allow service suspension where law, attacks, cost issues, or vendor suspensions interfere, which illustrates how upstream dependency can cascade into customer-facing disruption. | 中 | SR026 |
| CR026 | Anthropic’s terms also say third-party features are not Anthropic services and Anthropic is not responsible for them, underscoring layered accountability gaps in composite AI stacks. | 中 | SR026 |
| CR027 | SecondBrain’s supported integrations include Gmail, Google Calendar, Outlook, Slack, Notion, HubSpot, Salesforce, Google Workspace, and Microsoft 365, making connector reliability a central product dependency rather than a peripheral feature. | 中 | SR004 |
| CR028 | Salesforce, Microsoft, Workday, ServiceNow, and UiPath all market agentic or AI-agent automation from inside existing enterprise ecosystems, creating strong bundle-and-context distribution pressure against Main Func. | 中 | SR015, SR016, SR017, SR018, SR019 |
| CR029 | The credits model means heavier usage can raise revenue and compute spend simultaneously, which creates margin uncertainty unless the company proves strong cost control and pricing power. | 中 | SR029, SR003, SR011 |
| CR030 | Public customer metrics still do not show churn, NRR, or top-customer concentration, which makes it hard to separate real revenue quality from fast but shallow adoption. | 中 | SR011, SR012, SR013 |
| CR031 | TechCrunch’s early skepticism about business model, safety, and publisher implications remains relevant because Main Func is still asking enterprises to trust a broad autonomous-work layer. | 中 | SR014 |
| CR032 | Taken together, Main Func faces a compounded dependency problem in which vendor risk, connector risk, and incumbent distribution risk all attack the same customer-value chain. | 中 | SR022, SR025, SR026, SR028 |
| CR033 | OpenAI described early Super Agent growth as being achieved with a 20-person team, suggesting unusually high execution leverage but also a potential mismatch between product breadth and operating depth. | 中 | SR009 |
| CR034 | Anthropic later described roughly 50 engineers and quoted Kay Zhu saying “nobody really has a moat anymore” and that execution speed is the moat. | 中 | SR010 |
| CR035 | Business Wire said Genspark had established a local team with dedicated customer support and customer success resources in Japan, which is a positive mitigation but also evidence that support scaling is now a real operating challenge. | 中 | SR011 |
| CR036 | Public evidence does not yet reveal incident history, support backlog, implementation times, or formal succession depth, leaving people-and-process resilience under-documented. | 中 | SR003, SR011, SR012, SR013 |
| CR037 | The most important monitorable indicators are reliability outcomes, customer retention quality, gross-margin behavior, supplier concentration, and enterprise security-review performance. | 中 | SR021, SR022, SR030 |
| CR038 | A formal privacy or AI-governance investigation, a severe autonomous-action failure, or weak enterprise retention would each materially change the investment case. | 中 | SR020, SR023, SR024, SR030 |
| CR039 | If Main Func can prove durable enterprise retention, low false-action rates, and manageable model-driven gross margins, the current risk profile would compress quickly. | 中 | SR011, SR012, SR022 |
| CR040 | As of 2026-07-29, Main Func’s risk ranking is led by enterprise trust/compliance, supplier dependence, and distribution incumbency, with execution depth and margin opacity close behind. | 中 | SR020, SR025, SR028, SR031 |
| CV001 | Main Func has a credible bullish narrative because it appears to have converted fast product iteration into unusually rapid commercial traction for an AI-native workflow platform. | 中 | SV004, SV005, SV007, SV008 |
| CV002 | The anti-thesis is that public evidence still proves momentum more clearly than durable economics or defensible moat. | 中 | SV007, SV011, SV028, SV029 |
| CV003 | Main Func’s public product arc moved from 2024 search-era experimentation to autonomous agents and then to a broader AI workspace platform by 2026. | 中 | SV001, SV008, SV005 |
| CV004 | Public financing anchors stepped from a $275M Series B at a $1.25B valuation in November 2025 to later 2026 reporting around a $2.6B valuation and more than $645M total funding. | 高 | SV009, SV010, SV005, SV006 |
| CV005 | Business Wire said Genspark had crossed $100M ARR by January 2026 while TMCnet and FinancialContent later paired roughly $250M ARR or annualized run rate with the July 2026 story. | 高 | SV004, SV005, SV006 |
| CV006 | OpenAI’s case study added an earlier $36M ARR-in-45-days milestone, reinforcing that the company’s revenue narrative is unusually front-loaded. | 中 | SV008 |
| CV007 | Anthropic’s case study framed Main Func as a company where execution speed rather than moat purity is the central strategic claim. | 中 | SV007 |
| CV008 | TechCrunch’s early skepticism on business model and product trust remains relevant because later valuation upside still depends on durability rather than novelty. | 中 | SV011 |
| CV009 | The company’s public enterprise packaging and controls are real enough that the opportunity should not be treated as a pure consumer-AI story. | 中 | SV002, SV003, SV028, SV029 |
| CV010 | The best high-level frame is therefore conditional optimism rather than unqualified conviction. | 中 | SV001, SV004, SV011, SV029 |
| CV011 | A ~$2.6B valuation against a ~$250M ARR or annualized run-rate anchor implies roughly a ~10x revenue multiple, while the same valuation against $100M ARR would imply roughly ~26x. | 中 | SV004, SV005, SV006 |
| CV012 | That multiple compression from ~26x on the January ARR anchor to ~10x on the July run-rate anchor shows how sensitive the valuation is to which growth milestone is treated as durable. | 中 | SV004, SV005, SV006 |
| CV013 | Multiples.vc said July 2026 public software valuations showed wide dispersion and cited approximately 6.1x EV / NTM revenue for horizontal SaaS reference points. | 中 | SV021 |
| CV014 | Multiples.vc also argued that public investors were segmenting software by AI application, technical complexity, specialization, and disruption risk rather than by TAM alone. | 中 | SV021 |
| CV015 | The BVP Nasdaq Emerging Cloud Index is designed to track emerging public cloud-software companies, making it a useful public-market mood indicator even if it is not a direct comp set for Main Func. | 中 | SV020 |
| CV016 | Main Func probably deserves some premium to broad horizontal SaaS averages if the later 2026 revenue and customer claims are durable, but the public record does not support an unlimited AI premium. | 中 | SV005, SV007, SV021 |
| CV017 | Public software market data in 2026 therefore acts more as a guardrail than as a direct pricing answer for Main Func. | 中 | SV020, SV021 |
| CV018 | UiPath’s FY2026 results disclosed $1.853B ARR, 85% GAAP gross margin, and positive cash flow, showing the level of economic visibility public automation leaders provide. | 中 | SV012, SV022 |
| CV019 | Workday’s FY2026 results disclosed $9.552B revenue, 29.6% non-GAAP operating margin, 11,500-plus customers, and large subscription backlog, illustrating the disclosure standard of trusted enterprise platforms. | 中 | SV013, SV023 |
| CV020 | Salesforce’s FY2026 results disclosed $35.1B current remaining performance obligation and $14.4B free cash flow, highlighting just how much public proof lies behind premium enterprise-software valuations. | 中 | SV014, SV024 |
| CV021 | Main Func offers none of that disclosure depth publicly today, which should directly reduce valuation confidence even if growth is exceptional. | 中 | SV004, SV005, SV018, SV019, SV020 |
| CV022 | The public evidence supports a recommendation of research-more / pursue selectively rather than an unconditional invest or pass. | 中 | SV004, SV005, SV011, SV021 |
| CV023 | Confidence in that recommendation should be medium because the company’s momentum is strong but the price-driving private variables remain under-disclosed. | 中 | SV004, SV007, SV021, SV029 |
| CV024 | The appropriate risk rating is high because valuation support depends on retention, margin, governance, and term details that are still mostly private. | 中 | SV011, SV028, SV029 |
| CV025 | At a valuation near the latest public ~$2.6B mark, Main Func may still be investable if private diligence proves strong economics and clean terms; at materially higher pricing the margin of safety deteriorates quickly. | 中 | SV005, SV006, SV021 |
| CV026 | A supportable bear case centers on valuation compressing toward roughly $1.4B-$2.0B if growth quality, margins, or enterprise durability prove weaker than the public narrative implies. | 中 | SV011, SV018, SV019, SV021 |
| CV027 | A supportable base case centers on roughly $2.2B-$3.0B if the July 2026 scale claims are mostly real but not yet fully de-risked by public retention or margin proof. | 中 | SV004, SV005, SV006, SV021 |
| CV028 | A supportable bull case centers on roughly $3.5B-$5.0B only if private diligence shows durable enterprise retention, strong margins despite model costs, and limited preference overhang. | 中 | SV007, SV013, SV014, SV021 |
| CV029 | The next round can still be unattractive even if the company is attractive if terms or price pull too much future success into the present. | 中 | SV009, SV010, SV029 |
| CV030 | Final diligence should therefore focus on cap table, preference stack, revenue quality, margins, concentration, and reliability rather than on product vision alone. | 中 | SV021, SV028, SV029 |
| CV031 | The company’s public product and customer claims are strong enough that a total pass would be premature. | 中 | SV002, SV004, SV005, SV007 |
| CV032 | The lack of public filings means Main Func cannot currently be valued with the same confidence interval as mature public software companies. | 中 | SV022, SV023, SV024, SV025, SV026, SV027 |
| CV033 | UiPath, Workday, and Salesforce together show that public software valuations are ultimately earned through transparency around margins, renewals, backlog, and cash generation, not just growth narrative. | 中 | SV012, SV013, SV014, SV022, SV023, SV024 |
| CV034 | The presence of strong incumbent agent platforms from Salesforce, Workday, ServiceNow, Microsoft, and UiPath should raise the discount rate applied to Main Func’s long-term share assumptions. | 中 | SV015, SV016, SV017, SV018, SV019, SV030 |
| CV035 | Main Func’s enterprise packaging and trust surfaces are better than a typical early AI startup, which modestly supports the right to use premium-software rather than consumer-app comparables. | 中 | SV002, SV003, SV028, SV029 |
| CV036 | Even so, trust surfaces are not the same as public proof of enterprise durability, so they cannot close the valuation gap by themselves. | 中 | SV003, SV011, SV029 |
| CV037 | Unknown gross margin remains a central valuation variable because multimodal outputs, phone calls, and model routing can make delivery costs structurally heavier than mature SaaS averages. | 中 | SV003, SV007, SV008, SV021 |
| CV038 | Supplier and model-vendor dependence should influence multiple selection because Main Func’s product quality and unit economics depend on capabilities it does not fully control. | 中 | SV001, SV007, SV008, SV029 |
| CV039 | Exit readiness on public evidence is still limited; the more realistic near-term frame is a private financing or strategic optionality story rather than imminent IPO-grade readiness. | 中 | SV004, SV005, SV029 |
| CV040 | As of 2026-07-29, the final valuation verdict is to keep Main Func in the investable universe but require disciplined price, clean terms, and proof-heavy diligence before committing capital. | 中 | SV022, SV023, SV024, SV025, SV026, SV027 |