humans&
顶尖 frontier-AI 团队,种子轮融资极其充足,但公开产品和客户验证仍远远落后。
humans& 拥有一线创始人-市场匹配和足以支撑前沿 AI 平台的资本;但产品、客户和财务证据仍落后于估值叙事,公开记录目前只支持继续研究。
封面要素
公司概况
humans& 是一家 San Francisco Bay Area 私营公司,2025 年 9 月由 Eric Zelikman 和一小群来自 xAI、OpenAI、Anthropic、Google 及 Stanford 相关研究社群的 frontier-AI 研究员创立。公开材料把公司定位在人类协作、沟通和集体决策,而不是泛化的单用户自动化;保留报道称,团队正围绕这一论点同步演进模型和产品。公司已在 2026 年 1 月拿到一笔异常庞大的 $480 million 种子轮融资,报道估值为 $4.48 billion;对一家约 20-28 人的公司而言,这给了它罕见的算力采购和招募能力。核心尽调限制在于:融资可见度仍显著高于经营可见度。保留下来的公开证据尚未显示 live customer 部署、实质性产品指标、详细治理条款或公开收入披露。
- 成立时间
- 2025-09-01
- 创始人
- Eric Zelikman, Andi Peng, Georges Harik, Yuchen He, Noah Goodman
- 创立地点
- San Francisco Bay Area, California, USA
- 总部
- San Francisco Bay Area, California, USA
- 产品
- 面向沟通和协作工作流的隐身研究与产品系统;公开材料强调人类-AI 协作、多智能体强化学习、记忆和用户理解,而不是已发布的通用 API 或聊天机器人。
- 客户
- 未披露;保留来源暗示企业和消费者协作用例,但公开信息尚未验证具名客户群或已部署细分市场。
- 商业模式
- 未披露;截至 2026-06-19,保留公开来源没有给出定价、变现机制或收入披露。
- 阶段
- Seed private company
- 融资情况
- 公开报道支持公司在 2026 年 1 月完成 $480 million 种子轮,报道估值为 $4.48 billion,但保留来源并未一致说明确切估值口径或证券条款。
执行摘要
主要优势
- 创始人-市场匹配异常强:Eric Zelikman 和已披露团队都与 xAI、OpenAI、Anthropic、Google、Stanford 的前沿推理、对齐和产品工作相关。
- $480 million 种子轮让 humans& 在早期就拿到稀缺算力、招聘能力和时间窗口,可以探索差异化的协作优先逻辑。
- 产品愿景不是泛化聊天机器人自动化;保留下来的来源一致把它描述为围绕沟通、信任、记忆和协同的「模型 + 产品」路线。
- 投资人质量和财团深度带来信号强度,即使商业化发布前也可能提供生态入口。
主要风险
- 公开商业化证据明显薄于融资证据:保留下来的公开来源没有验证具名客户、生产部署、定价或实时使用指标。
- 关键人和团队深度风险高,因为公司相对野心仍显得很小,公开控制信号也集中在 Eric Zelikman 身上。
- 算力稀缺、专业人才稀缺和前沿模型资本强度,可能比预期更快吃掉账上资金的有效价值。
- 公开基本面很难支撑报价估值,因为轮次基础、治理条款和运营指标仍不透明。
未决问题
- 确切已发布产品界面、路线图和架构,包括是否已有在线协作产品或 API。
- 具名客户、生产部署、留存信号和细分买方证据。
- 收入模式、定价、烧钱速度、现金跑道、毛利率及其他核心财务承销输入。
- 董事会构成、估值基础、所有权稀释、清算优先权及其他种子轮治理条款。
目录
01公司概览
1.1 身份、使命与地域足迹
humans& 于 2026 年 1 月 20 日公开亮相,自称是一家「以人为中心的 frontier AI lab」,核心判断是进步来自信任、连接和协作,而不是用孤立的自主系统取代人。公司自己的材料强调 AI 的连接组织作用:产品方向被表述为帮助组织和社区更好地一起工作,而不只是自动化单个任务。官方表述把这一论点同长周期和多智能体强化学习、记忆、用户理解,以及研究与产品紧密耦合的开发模式连在一起。这足以证明公司有真实研究议程,但不足以证明商业化准备就绪。 地点证据方向清楚,但并不完全干净。二级公司目录和 California registry mirror 指向 601 Marshall Street 的 Redwood City 邮寄和主要地址,而多家媒体和数据库资料使用更宽泛的「San Francisco」简称。后续章节最安全可复用的事实底座是:humans& 是一家 Bay Area 公司,公开备案地址在 Redwood City,运营叙事则落在 San Francisco/Palo Alto。确切总部标签因此是一个小但真实的尽调清理项,本章不应夸大。[CO001, CO002, CO003, CO004, CO005, CO006]
| 指标 | 数值 / 状态 | 日期 | 信心 | 缺口 |
|---|---|---|---|---|
| 成立日期 | September 2025 | 2025-09 | 中 | |
| 法律实体备案 | Humans& Ai, Inc 已在加州记录中备案 | 2025-10-27 | 中 | |
| 当前阶段 | 种子轮;产品披露仍接近隐身状态 | 2026-01-20 | 中 | |
| 公开轮次规模(USD M) | 480 | 2026-01-20 | 高 | |
| 公开轮次估值口径 | $4.48B 估值被广泛报道;口径未被一致说明 | 2026-01-20 | 中 | 需要领投方或公司确认投前与投后处理。 |
| 总部 / 主要地址 | Redwood City 备案地址;更广义的 Bay Area / San Francisco 叙事 | 2026-03-25 | 中 | 公开来源没有完全对齐邮寄地址与运营地点简称。 |
| 产品状态 | 截至 2026 年 1 月亮相时,没有公开记录显示已发布产品 | 2026-01-20 | 中 | |
| 收入 / ARR | 2026-06-19 | 低 | 没有留存公开来源披露收入、ARR 或变现规模。 | |
| 具名客户 | 2026-06-19 | 低 | 没有留存公开来源识别参考客户或生产部署。 | |
| 员工数 | 2026-06-19 | 低 | 公开来源提到发布时约 20 名员工,以及后续 28 名员工估计,但没有经审计的当前人数。 | |
| 董事会披露 | 2026-06-19 | 低 | 留存公开来源没有披露正式董事会名单或控制权。 |
空值标记留存公开来源仍未支撑的私人指标或治理项。
[CO007, CO008, CO010, CO011, CO012, CO023]使命、研究重点、人才、资本和落地风险如何拼成当前公司叙事。
[CO001, CO003, CO004, CO018, CO025, CO026]当前公开成熟度指标更强调资本和履历,而不是商业证明。
员工数作为方向性公开估计展示;收入和客户指标仍未披露。
[CO011, CO023, CO025, CO028, CO029, CO034]1.2 创始人、领导层与关键人物风险
当前对 humans& 最强的承保论点是 founder-market fit。Eric Zelikman 是 CEO 兼联合创始人,保留来源支撑一个一致画像:早期 xAI 贡献者、Stanford 博士候选人,以及参与 STaR 和 Quiet-STaR 推理工作的研究员。其余已披露创始阵容对种子期公司而言也异常强。Andi Peng 被呈现为 Anthropic 后训练和强化学习操盘手;Georges Harik 带来 Google 早期时代的深厚产品和商业化经验;Yuchen He 的官方简介把他同 xAI 及此前 OpenAI 工作相连;Noah Goodman 则补上 Stanford 学术和认知科学视角。纸面上,这套组合很贴合公司所说的尝试:把 frontier-model 研究同协作产品论点融合。 同一组证据也指向集中的执行风险。公开治理披露稀少;目前最好的 corporate-registry mirror 不仅把 Eric Zelikman 列为 CEO,也列为 CFO 和 Secretary。这对一家刚成立的 startup 并不罕见,但意味着关键人物依赖在公司刚完成 mega-seed 轮的时点异常高。尽调问题因此不是团队是否有才华,而是这种高度集中的创始结构能否承接多十亿美元估值随之而来的运营、治理和招聘要求。[CO011, CO012, CO013, CO014, CO015, CO016]
| 人物 | 角色 | 背景 | 创始人市场匹配或职能覆盖 | 关键人依赖 |
|---|---|---|---|---|
| Eric Zelikman | CEO、联合创始人 | 前 xAI 贡献者;Stanford 博士候选人;与 STaR 和 Quiet-STaR 研究相关 | 直接匹配公司的协作与推理研究论点及公开使命 | 高 |
| Andi Peng | 联合创始人 | 前 Anthropic 研究员,参与 Claude 后训练和 RL 工作 | 增加后训练和前沿模型行为专长 | 中 |
| Georges Harik | 联合创始人兼投资领投人 | Google 第 7 号员工,拥有早期广告、Gmail、Docs 和 Android 收购经验 | 带来种子阶段少见的商业化、网络和融资杠杆 | 中 |
| Yuchen He | 联合创始人 | 官方简介将他与 xAI 以及此前在 OpenAI 的后训练和记忆工作联系起来 | 连接模型训练与产品导向的记忆和评估工作 | 中 |
| Noah D. Goodman | 联合创始人 | Stanford 教授,认知 / AI 研究员 | 增加围绕人类理解和推理的学术可信度与研究深度 | 中 |
本表覆盖公开披露的创始人梯队,不是完整运营领导组织架构图。
[CO014, CO015, CO016, CO017, CO018, CO019]1.3 资本结构、阶段与覆盖缺口
humans& 最明确的外部事实,是 2026 年 1 月融资规模。多家主流来源一致称,公司以 $4.48 billion 估值完成 $480 million 种子轮融资;SV Angel 和联合创始人 Georges Harik 反复被点名为领投方,Nvidia、Jeff Bezos、GV 和 Emerson Collective 反复被列为投资方。公司官网还列出一长串其他支持者,包括 Forerunner、S32、DCVC、Human Capital、Liquid 2、Felicis 和 CRV。Crunchbase News 还报道称,联合创始人 Andi Peng 表示大部分新资金将用于算力,这符合公司训练新模型而非只是包装现有模型的计划。 公开记录尚未给投资人提供一轮如此大融资通常需要的披露层级。可访问来源没有一致说明 $4.48 billion 应视为 pre-money 还是 post-money;保留来源集合也未独立确认非正式投资人名单中流传的所有名字,包括 Abstract Ventures。同样,公开资料没有 cap table、没有披露董事会构成、没有收入或 ARR 披露、没有具名客户群,也没有经审计员工数。TechCrunch 的发布报道提到约 20 名员工,Tracxn 随后在 2026 年 5 月给出 28 人估计。这些是有用的规模信号,但不能替代直接尽调数据。[CO023, CO024, CO025, CO026, CO027, CO028]
| 利益相关方 | 角色 | 控制权或经济重要性 | 尽调要求 |
|---|---|---|---|
| SV Angel | 领投方 | 多次被点名为 $480M 种子轮领投方,且可能对未来银团信号至关重要 | 要求披露经济条款、董事会权利、按比例跟投权和任何附函。 |
| Georges Harik | 联合创始人兼联合领投投资人 | 创始人与融资领投人的罕见双重角色可能放大战略影响力 | 厘清创始人持股、出资规模,以及投资人角色附带的治理权。 |
| Nvidia | 战略投资人 | 既是资本提供方,也可能影响算力渠道 | 厘清投资是否包含算力分配、商业承诺或最惠条款。 |
| Jeff Bezos / Bezos vehicle(投资方) | 知名投资人 | 增加品牌和网络背书,但公开经济条款未披露 | 确认实体名称、出资规模,以及任何信息权或观察员权。 |
| GV | 机构投资人 | 显示顶级风投支持,并可能具备后续跟投能力 | 确认出资规模,以及 GV 是否拥有正式治理权或数据权。 |
| Emerson Collective | 机构投资人 | 在公开银团名单中反复出现,是可见支持方 | 确认参与规模和资本之外的战略相关性。 |
| 长尾种子轮银团 | 其他财务支持方 | 公司官网列出 Forerunner、S32、DCVC、Human Capital、Liquid 2、Felicis、CRV 等 | 将网站名单与完整股权表核对,并确认任何遗漏投资人。 |
| 创始团队 / 管理层 | 控制群体 | 运营控制权集中,因为公开治理披露很少,登记记录显示 Eric Zelikman 持有多个高管头衔 | 要求董事会名单、保护性条款、高管授权,以及补强财务 / 法务深度的招聘计划。 |
这是公开可见度下的利益相关方地图,不是股权表或确定所有权安排。
[CO028, CO029, CO030, CO031, CO032, CO033]1.4 里程碑、隐身状态与负面背景
时间线很短,但事件密集。保留公开来源支撑一条序列:2025 年 9 月成立,2025 年 10 月完成注册并出现融资传闻,2026 年 1 月出现条款和隐私通知,随后公开脱离隐身状态并正式宣布种子轮。媒体报道又补充了几个重要的发布后节点:2026 年初计划推出首款产品,2026 年 3 月确认法律实体仍处于 active 状态,2026 年 6 月第三方估计显示团队按已募集资本隐含标准看仍然很小。由于公开产品界面仍然很薄,除非出现更新的一手证据,本章应作为后续章节引用的标准时间线。 这条时间线也带有主要负面信号。Reworked 的怀疑性报道认为,humans& 在没有已发布产品的情况下以独角兽身份亮相,其估值即便放在 2026 年 AI 融资市场也不寻常,而且企业激励仍可能把「以人为中心」的系统拉向自动化和裁员。这并不能推翻公司的论点,但它是一个可信警示:资本和创始人履历到得比 product-market fit 证明更快。后续尽调因此不应重点问 humans& 能否融资,而应问它能否把协作式 AI 哲学变成客户真正部署并持续使用的产品。[CO011, CO012, CO013, CO027, CO030, CO034]
| 日期 | 事件 | 类型 | 金额 / 估值 / 状态 | 参与方 | 含义 |
|---|---|---|---|---|---|
| 2025-09 | humans& 成立 | 创立 | 公司成立 | 创始团队 | 设定本报告通篇使用的公司时间线起点。 |
| 2025-10-27 | Humans& Ai, Inc 备案并成为加州追踪的活跃实体 | 监管 | 实体活跃;文件 B20250359156 | Eric Zelikman;Telos Legal Corp.(法律实体) | 提供具体法律实体里程碑和公开备案地址。 |
| 2025-10-31 | Forbes 报道称 Eric Zelikman 正在洽谈为一家新的前沿 AI 实验室融资约 $1B | 融资 | 传闻目标 $1B,估值 $5B | Forbes;Eric Zelikman | 显示公开发布前已有大资本雄心。 |
| 2026-01-19 | 网站条款和隐私通知生效 | 治理 | 公开法律界面就位 | humans& ai, inc.(公司实体) | 表明公司在发布前夕正准备公开网页形象。 |
| 2026-01-20 | humans& 公开走出隐身状态,并将自己定位为以人为中心的前沿 AI 实验室 | 产品 | 公开亮相 | humans&;创始团队 | 创建首个权威使命陈述和产品论点。 |
| 2026-01-20 | 公司宣布 $480M 种子轮融资,估值 $4.48B | 融资 | $480M 种子轮;公开估值口径未被一致说明 | SV Angel;Georges Harik;Nvidia;Jeff Bezos;GV;Emerson Collective;其他 | 立即让公司跻身有记录以来最大 AI 种子轮之一。 |
| 2026-01-21 | TechFundingNews 称公司计划在 2026 年初发布首款产品 | 产品 | 首款产品仍待发布 | TechFundingNews;humans& 官网声明 | 显示资本早于公开产品发布。 |
| 2026-03-25 | 加州登记镜像更新实体为活跃状态,并列出 Redwood City 主要地址和邮寄地址 | 治理 | 活跃状态再次确认 | Bizprofile / California Secretary of State 数据 | 支撑持续公司活动和 Bay Area 足迹。 |
| 2026-06-10 | Tracxn 更新公司档案,标注种子轮状态和 28 人员工估计 | 规模 | 仅为第三方估计 | Tracxn | 表明相对于种子轮规模,团队仍然很小。 |
| 2026 | Reworked 将公司描述为一家无产品独角兽,其增强论点可能仍被自动化激励牵引 | 负面 | 可信质疑,不是正式程序 | Reworked;外部评论者 | 确立后续章节的主要负面尽调视角。 |
这是公开公司概览的唯一记录年表;缺少精确日期证据时使用月份级日期。
[CO011, CO012, CO013, CO023, CO027, CO028]从 2025 年 9 月创立,到 2026 年 6 月规模估计与质疑的公开里程碑。
在保留的公开来源没有给出精确发布日期时,使用月份级日期。
[CO011, CO012, CO013, CO023, CO027, CO028]02市场分析
2.1 市场边界:一个狭窄的监督与研究切口,而非泛化 AI TAM
对 humans& 而言,市场边界应围绕这样一类系统:把 frontier-model 推理同人工复核、评估、编排和领域判断结合起来。公开记录对这一框架的支撑,远强于对公司可覆盖整个 AI 经济的任何说法。HITL 来源把最接近、可观察的类别定义为把人工反馈嵌入模型训练、验证和决策的软件、硬件与服务,尤其适用于错误成本、可解释性或问责性重要的场景。较新的批评把边界收得更紧:许多所谓 HITL 部署本质上是 AI-in-the-loop 系统,人类仍是最终决策者,模型只是加速器。这一区分对尽调很重要,因为它表明买方通常为可信决策支持、可审计性和工作流可靠性付费,而不是为原始自主模型输出付费。结果是,市场边界落在协作研究、评估和质量保障工作流上;更广的基础设施和 consumer-AI 支出只能作为背景。[CM001, CM002, CM003, CM004, CM005, CM006]
| 细分 / 类别 | 纳入支出 | 排除支出 | 买方 / 付款方 | 与 humans& 的相关性 |
|---|---|---|---|---|
| 人机协作 AI 研究、评估和监督 | 模型评估、提示词 / 工作流管理、人工审核、QA、可观测性、治理,以及专家在环研究工作流 | 通用聊天机器人席位、纯基础设施、通用云 GPU 转售、消费 AI | CTO / CIO / CDAO、研究负责人、风险负责人 | 最接近且有直接证据的切口 |
| 企业 AI 智能体平台 | 领域智能体、编排、工作流自动化,以及绑定 CRM / ERP / IT / 知识工作流的 copilots | 没有企业工作流集成的简单 FAQ 机器人或消费者助手 | 业务职能负责人加中央 IT / AI 平台团队 | 重要的相邻分发层 |
| 模型和 AI 平台预算 | AI 模型、DS / ML 平台、应用开发工具、模型服务支持 | 半导体、通用设备、无关 SaaS 模块 | 中央 AI 平台、工程、开发者平台负责人 | 可为 humans& 类产品供资的上层预算池 |
| 广义 AI 基础设施和设备 | AI 优化 IaaS、服务器、网络结构、半导体、AI 设备 | 这一广义口径内无排除项 | Hyperscalers、OEM、基础设施买家 | 有用的背景上限,但大多不可触达 |
| 现状替代方案 | 内部分析师或研究团队、顾问、RPA、传统知识工具、人工审核队列 | N/A | 今天已经支付现有工作流的同一运营预算 | 早期试点的真实竞争集合 |
各行是边界视角,而非可相加子市场。本章有意区分可直接观察的监督 / 研究支出,与只存在部分可争夺性的更广义 AI 类别。
[CM001, CM003, CM004, CM005, CM006, CM007]从广义 AI 经济到最贴近 humans& 的狭义人工监督切片的嵌套视角。
所有数值均为十亿美元。各层是概念嵌套,不是可相加的会计类别;较低层依赖不同发布方和范围定义。
[CM006, CM008, CM013, CM017, CM018, CM020]2.2 规模测算视角:相邻大预算存在,但只有一部分可争夺
公开来源给出几种可用的市场视角,但它们描述的是不同支出层,不应混成一个 TAM 数字。Gartner 的万亿美元级 AI 预测可作为天花板参考,因为它显示资本流向哪里;但其中大部分是基础设施和供应商产能,并非可直接触达的推理工具支出。更接近 humans& 的是 Gartner 的模型层和平台层,两者合计意味着模型消费和开发者工具有数百亿美元预算池。IDC 补充了行业化企业视角,显示软件、银行和零售已在 AI 上大量支出,生成式 AI 已成为该预算的实质性部分。再窄一层,AI-agent 报告显示工作流层快速扩张;HITL 市场报告则为人工复核和模型质量类别给出最低的直接公开边界。这种矛盾有用而非麻烦:类别显然真实存在,但相关市场取决于买方把解决方案归为模型工具、agent 软件,还是可信人工监督基础设施。[CM008, CM009, CM010, CM011, CM012, CM013]
| 发布方 / 视角 | 年份 / 地理 | 数值 | 增长 / 节奏 | 为什么重要 | 关键限制 |
|---|---|---|---|---|---|
| Gartner 总 AI 支出上限 | 2026 / 全球 | $2.596T | 47% YoY | 显示整体 AI 资本流入规模 | 主要是基础设施和供应商主导的容量,不是 humans& 可触达支出 |
| Gartner AI 模型 | 2026 / 全球 | $32.6B | 2026 年展望为 110% 增长 | 最接近的单一公开模型层预算 | 仍比协作研究工具更宽 |
| Gartner AI DS/ML + 应用开发平台 | 2026 / 全球 | $38.3B | 平台随企业集成增长 | 捕捉可能为推理工具供资的开发者和平台预算 | 并非所有支出都会流向推理或评估用例 |
| Gartner 模型与工具组合外沿区间 | 2026 / 全球 | $70.9B | 由模型 + DS/ML + 应用开发平台推导 | 相比完整 AI TAM,更接近 humans& 的软件 / 工具预算合理外沿 | 推导估计,不是已发布的独立市场 |
| IDC 领先企业行业 | 2024 至 2028 / 全球 | 2024 年 $89.6B;到 2028 年近 $222B | 27% 五年 CAGR | 显示软件、银行和零售中的企业预算池 | 广义行业 AI 支出,不特指推理 |
| IDC 上述行业隐含 genAI 切片 | 2024 / 全球 | 估计约 ~$17.0B | GenAI 占所引行业支出的 >19% | 表明已有可观预算被拨给较新模型类别 | 简单比例估计,不是披露的产品类别 |
| MarketsandMarkets AI 智能体 | 2025 至 2030 / 全球 | $7.84B 至 $52.62B | 46.3% CAGR | 最佳公开相邻工作流软件视角,用于智能体系统 | 供应商定义类别,公开方法论有限 |
| The Business Research Company HITL AI(研究来源) | 2025 至 2030 / 全球 | $5.4B、2026 年 $6.73B、2030 年 $16.4B | 24.7% 至 24.9% CAGR | 与人工审核和监督相关的最窄公开下限视角 | 在许多用例中混合软件、服务和硬件 |
| MarketsandMarkets HITL 公开页面 | 2024 至 2029 / 全球 | 公开数值被隐去 | 公开页面仅提供定性增长 | 可用于判断类别形态和驱动因素 | 方法论不透明,不能单独锚定数值 TAM |
各行不可相加。每一行捕捉不同支出层:广义 AI 采购、模型 / 平台工具、智能体软件,或人工监督服务。
[CM008, CM009, CM010, CM011, CM012, CM013]可比公开视角从狭义 HITL 支出延伸到更宽的模型与工具预算带。所有数值均为十亿美元。
第 1 行使用已发布的 2025、2026 和 2030 年 HITL 数值;第 2 行使用已发布的 2025 和 2030 年 AI-agent 数值,并因没有公开中点而重复当前期数值作为中点;第 3 行把 AI 模型视为低端,把模型加 DS/ML 平台视为中点,把模型加两类平台视为高端。
[CM010, CM011, CM012, CM013, CM017, CM018]2.3 买方、用户与付款方:预算在 AI 从试点进入核心工作流时出现
采用广度已不再是核心问题,规模化企业价值才是。McKinsey 和 Deloitte 都显示 AI 使用已经广泛,但只有少数组织从实验推进到更深的企业转型。这一模式意味着 humans& 不太可能先卖进泛化创新预算。更可能的买方,是已经拥有重复、重判断工作流的负责人:中央 AI 平台团队、CIO 或 CTO 组织、CDAO 牵头的治理团队、R&D 负责人、知识管理负责人,以及 professional-services 运营方。用户是最接近模型失败模式和工作流瓶颈的人:研究员、开发者、分析师、复核员、合规团队,以及同时需要自动化和人工覆盖的运营人员。付款方会随用例变化。有些公司预算在中央 AI 或数据平台条线;另一些公司则由能把采购同更快研究、更低复核成本或更可靠决策挂钩的职能负责人出钱。最可能的早期采用路径,是从有边界的工作流试点,走向更深的工具集成,最后才进入多团队规模化部署。[CM021, CM022, CM023, CM024, CM025, CM026]
| 细分 | 主要买方 | 主要用户 | 可能付款方 / 预算所有者 | 工作流 | 采用触发因素 |
|---|---|---|---|---|---|
| 中央 AI 平台 / ML 工程 | CTO、CIO、VP Engineering | ML 工程师、AI 平台团队、开发者 | 中央数据 / AI 平台预算 | 模型路由、评估、可观测性、集成、工具使用可靠性 | 需要在大量内部用例中标准化工具 |
| 研发和产品开发 | Chief Product Officer、研发负责人 | 研究员、产品经理、科学家、工程师 | 业务单元创新或研发预算 | 深度研究、实验综合、设计空间探索 | 需要在不丢失专家审核的情况下压缩周期 |
| 风险、合规与质量职能 | 首席风险官、总法律顾问、合规负责人 | 审核员、审计员、政策团队、QA 团队 | 风险 / 合规或共享服务预算 | 人工监督、升级处理、审计轨迹、政策检查 | 需要把 AI 从试点推到有监控的生产环境 |
| 专业服务与知识工作 | 业务线负责人、COO、知识管理负责人 | 分析师、顾问、主题专家 | 业务线 P&L 或转型预算 | 文档分析、研究、起草、知识检索 | 人力开支大,重复审核环节多 |
| IT 与服务运营 | CIO、服务运营负责人 | IT 运营人员、支持团队、知识管理人员 | IT 运营预算 | 工单分流、服务台辅助、运行手册自动化 | 需要压低积压,并在不完全自治的前提下加快响应 |
| BFSI 和医疗等受监管行业 | 业务线负责人加风险负责人 | 核保员、理赔团队、临床人员、审核人员 | 带治理覆盖的业务线预算 | 高文档密度环境里的决策支持 | 既需要自动化,也需要可审计的人工覆盖 |
这张买方地图受到证据约束。公开来源能识别采用 AI 最快的职能和行业,但没有披露 humans& 类产品的定价或赢率数据。
[CM021, CM022, CM025, CM026, CM027, CM028]基于证据的序数判断:humans& 类协作推理系统最可能先落地在哪里。
矩阵单元格是序数判断,综合了关于 agent 和 AI 工作流已在哪些场景试点或扩张的调研与市场报告证据;它们不是来源报告的市场份额。
[CM022, CM026, CM027, CM028, CM029, CM031]2.4 增长驱动与约束:信任、工作流重构和算力纪律同需求一样重要
市场顺风真实存在:更多企业可用 AI,更多预算流向 agent 和模型工具,更多行业在测试可由推理模型增强人类专家的工作流。但公开证据同样清楚:增长不会自动转化为可规模化价值。治理成熟度仍低,监管和标准预期却在反向上升,推动买方走向可审计、有人监督的部署。数据访问和集成仍是持续阻碍;任何承诺跨系统研究或编排的产品都尤其受此影响。技能缺口和工作流重构也居于核心;瓶颈往往不是模型,而是包在模型外面的 operating model。Bain、BCG 和 RAND 从不同角度反复给出同一条信息:组织在自动化破碎流程、过度假设自主性,或低估上下文、数据和人工运营设计投入时,会错失目标。算力和能源约束又加上一道刹车:推理密集型工作负载成本上升,使效率、路由和评估纪律具备经济重要性。对 humans& 来说,这意味着可触达市场可以快速增长,但前提是产品帮助买方跨过信任、集成和 operating-model 门槛,而不是再增加一个投机性试点。[CM032, CM033, CM034, CM035, CM036, CM037]
| 驱动因素 / 约束 | 方向 | 时间 | 含义 | 尽调追问 |
|---|---|---|---|---|
| AI 采用面广,但规模化价值仍浅 | 混合 | 当前 | 带来大量试点机会,但转成耐久平台的速度较慢 | 追问管线里多少是试点工具,多少是生产关键支出 |
| 智能体式工作流扩张 | 驱动 | 2025-2027 | 推高对编排、审核和评估层的相邻需求 | 检验 humans& 的定位是智能体基础设施、研究副驾,还是托管服务 |
| 人工监督与合规要求 | 驱动 | 当前且趋严 | 可审计审核、日志和升级功能更值钱 | 梳理目标工作流里哪些按客户政策或监管口径属于高风险 |
| 技能缺口与工作流重塑不足 | 约束 | 当前 | 即便预算存在,也会拖慢部署 | 要求证明早期用户无需沉重变革管理就能吸收产品 |
| 数据访问与集成摩擦 | 约束 | 当前 | 阻断跨系统推理,并抬高实施成本 | 检查连接器、安全模型,以及客户数据上的首次可用时间 |
| 自治预期跑在现实前面 | 约束 | 当前 | 若产品按完全自治经济性售卖,买方可能抗拒 | 验证产品在稳定状态下假设多少人工审核 |
| 供应商锁定、IP 与信任顾虑 | 约束 | 当前至中期 | 大型买方可能推迟承诺,或要求可迁移性和强治理 | 追问模型可迁移性、审计轨迹、留存,以及输出权利 |
| 高级工作负载的计算与能耗强度 | 约束 | 中期 | 高效模型路由和 ROI 纪律更重要 | 要求单位经济性,并说明生产中如何控制重推理任务成本 |
驱动因素和约束有意混列,因为两者共同决定相邻 AI 预算能否被 humans& 类产品变现。
[CM021, CM022, CM025, CM032, CM033, CM034]企业采用通常从边界清晰的工作流问题开始,走向数据集成、人工审核设计,最后才进入规模化推出。
该漏斗是概念性而非数值化;它总结了 Bain、Deloitte、PwC 和 RAND 材料隐含的常见采用步骤,而不是已发布的转化基准。
[CM025, CM032, CM033, CM034, CM037, CM040]2.5 图表
03竞争对手
3.1 格局分层
humans& 不应只同某一家具名 startup 比较。真实格局至少分成四类方案。第一类是 Thinking Machines Lab 和 Safe Superintelligence 这样的 frontier neo-labs,它们同 humans& 在精英团队履历、巨额融资轮,以及先定义新模型架构再交付广泛部署企业产品的意愿上重叠。第二类是 Anthropic、OpenAI 和 xAI 等 incumbent model vendors,它们已经开放产品、定价、透明度和开发者界面。第三类是 Scale AI、Labelbox、Braintrust 和 Arize 等 enabling vendors;它们不需要赢得 frontier-model 竞赛也能拿到预算,因为它们今天就销售评估、RLHF、可观测性和人工反馈工作流。第四类是现状:围绕公开 API 和工具内部自建。最后一类重要,因为 humans& 公开材料强调新的协作中心模型,但买方已经可以把现有模型、eval stack 和工作流软件组合成可接受替代品。[CP003, CP010, CP015, CP018, CP019, CP022]
| 竞争对手 / 类别 | 类型 | 规模 / 融资信号 | 目标客群 | 差异化 | 局限 |
|---|---|---|---|---|---|
| humans& | 协作推理新型实验室 | $480M 种子轮,估值 $4.48B;顶尖前实验室团队 | 需要沟通与协同智能的团队 | 以人为中心、协作优先的叙事 | 已审阅材料中没有公开产品、定价、客户或合作伙伴证明 |
| Thinking Machines Lab | 前沿研究 + 产品实验室 | 约 $2B 融资,估值 $12B;后续曾探索更高一轮融资 | 研究人员、初创公司和未来前沿模型用户 | 人机协作叙事叠加前沿模型野心 | 相比现有巨头,公开产品界面仍处早期 |
| Safe Superintelligence | 研究优先的前沿实验室 | 此前一轮 $1B;后续讨论 $20B+ 估值,并被报道估值 $32B | 顶尖研究人员、长期 AGI 支持者 | 单一的安全超级智能使命 | 没有公开企业工作流产品 |
| Anthropic | 现有模型供应商 | 公开定价、系统卡、透明度中心 | 购买模型和治理式部署的企业 | 产品、治理和企业界面可见 | 不属于 humans& 意义上的协作专用 |
| OpenAI | 现有模型供应商 | 公开企业和 API 界面,并有评估文档 | 企业、开发者、内部构建团队 | 分发、企业打包、公开评估工具 | 旧版评估平台正在迁出 |
| xAI | 现有模型供应商 | 公开模型和开发者文档 | 接受 API 主导采用路径的开发者和买方 | 公开模型界面推进快,并带搜索工具 | 本材料中协作专用工作流护城河证据较少 |
| Scale AI | 人类数据 / 评估平台 | Meta 投资后估值 >$29B | 前沿实验室、企业、政府 | 专家标注员、评估、RLHF、红队测试 | 不是协作模型公司 |
| Labelbox | 人类数据 / 评估平台 | 免费入口叠加服务和企业销售 | AI 实验室和企业模型团队 | 偏好竞技场、评估、RLHF、专家网络 | 竞争点在工作流基础设施,不在基础模型 |
| Braintrust / Arize | 可观测性和评估工具 | 公开定价和开源入口 | 构建自有栈的 AI 产品团队 | 低摩擦的可观测性和评估层 | 可与多家模型供应商拼装 |
| 内部构建 | 现状替代方案 | 使用公开 API 加现成工具 | 具备内部 ML 能力的产品和平台团队 | 最大灵活性和多归属 | 集成负担更重,迭代更慢 |
覆盖范围有意保持局部:表格聚焦证据最充分的替代选项,包括前沿实验室、现有模型供应商、评估基础设施和现状替代方案。
[CP003, CP010, CP015, CP018, CP019, CP022]与 humans& 竞争的主要解决方案类别序数图。
坐标轴是序数判断,综合了已审阅的产品和公司材料,而不是来源报告的基准分数。
[CP003, CP010, CP015, CP018, CP019, CP023]3.2 Frontier Labs 对比 Public Platforms
humans& 最接近的概念同行是 Thinking Machines Lab,其次才是 SSI。Thinking Machines 在明确的人类-AI 协作叙事上重叠,也在「frontier 能力和可用性应一起设计」的主张上重叠。SSI 在用例上重叠较少,但重要性在于:它争夺的是任何雄心勃勃的 frontier lab 都需要的稀缺研究员、投资人和算力关系。相比之下,Anthropic、OpenAI 和 xAI 从不同起点竞争:它们已经有公开产品、企业分发和有记录的工具链。因此,买方决定是否同 humans& 合作时,并不只是从 frontier 研究叙事中选择;它也在决定,是等待一家隐身实验室交付独特产品,还是现在就购买现有 API 或企业平台。这一分裂具有战略重要性,因为 humans& 的公开切口比泛化推理或 coding 市场更窄、更具体,但其商业化准备仍落后于已经向企业销售的 incumbents。[CP004, CP008, CP009, CP012, CP013, CP018]
| 采购标准 | humans& | Thinking Machines / SSI 新型实验室 | Anthropic / OpenAI / xAI(前沿模型公司) | Scale / Labelbox / Braintrust / Arize | 内部构建 |
|---|---|---|---|---|---|
| 公开产品可用性 | 无已确认的公开产品 | 部分:TML 有 Tinker;SSI 无公开产品 | 是 | 是 | 是,由公开组件拼装 |
| 协作专用推理叙事 | 高 | 中 | 中低 | 低 | 可定制,但团队必须自行设计 |
| 人类评估 / RLHF 运营 | Unknown | Unknown | 部分 | 高 | 随团队能力而变 |
| 治理 / 透明度界面 | Unknown | 中低 | 高 | 中 | 团队自有 |
| 公开定价 / 打包 | Unknown | Unknown | 中高 | 中高 | 内部成本模型可知 |
| 即时采购就绪度 | 低 | 中低 | 高 | 高 | 中 |
| humans& 可能面临的多归属风险 | 高 | 高 | 高 | 高 | N/A |
单元格只写已审阅材料能支持的内容;“未知”表示没有找到经过验证的公开证据,不代表能力不存在。
[CP005, CP012, CP018, CP019, CP020, CP022]基于已审阅公开证据,概览各类解决方案看起来最强的领域。
取值是基于已审阅语料的综合标签;“未知”表示未找到经验证的公开证据,并不意味着能力不存在。
[CP018, CP019, CP022, CP023, CP027, CP033]3.3 人类反馈与评估层
enabling stack 不是旁注;它是真实竞争集合,因为 humans& 想要的许多预算可能流向数据、评估或监控。Scale 把自己定位为 frontier model developers 的测试与评估伙伴,强调专家打分员、自定义 eval sets、RLHF 和 red teaming。Labelbox 用 arenas、benchmarks 和模型比较工作流销售类似的人类偏好与评估动作。Braintrust 和 Arize 推广可观测性和评估基础设施,让团队能检查 traces、给输出打分并迭代,而不必押注某一家隐身 foundation model 供应商。Humanloop 尤其有参考意义,因为它同时展示机会和整合:它做出一个带免费到企业梯度的 LLM eval 平台,随后宣布加入 Anthropic 并关闭独立平台。对 humans& 而言,这意味着周边层拥挤、模块化,并且越来越能同大实验室合作,而不是保持独立。若买方主要想要测量、偏好收集或部署反馈,它已经有替代方案,不必等待新的 frontier lab 成熟。[CP023, CP024, CP025, CP026, CP027, CP028]
| 供应商 / 类别 | 公开定价状态 | 商业模式 | 明确包含内容 | 含义 |
|---|---|---|---|---|
| humans& | 未审阅到公开定价 | 隐身 / 未知 | 未验证到公开商业套餐 | 当前采购需要私下尽调,不是网站自助下单 |
| Thinking Machines Lab | 未审阅到公开定价 | 早期产品 / 未知 | Tinker 已存在,但未审阅到公开商业打包 | 作为现成替代品仍不成熟 |
| SSI | 未审阅到公开定价 | 研究优先 / 未知 | 未审阅到公开企业产品 | 今天主要在人才、资本和算力上构成竞争 |
| Anthropic | 公开套餐可见;企业附加项未完全列明 | 自助服务加企业销售 | Claude 套餐和公开治理材料 | 对试点买方的摩擦低于隐身实验室 |
| OpenAI | 公开 API 定价加企业销售动作 | 按 token 计价的 API 和企业合同 | API 定价、企业产品、评估文档 | 内部构建团队的强默认选项 |
| Scale AI | 已审阅评估页面没有标价 | 定制企业 / 项目报价 | 评估、RLHF、监控、红队测试 | 预算可能通过销售主导合同落地 |
| Labelbox | 免费入口加企业 / 服务 | 自助服务与服务混合 | 模型评估、RLHF、专家网络 | 先解决工作流层,可能削弱模型实验室论点 |
| Braintrust | 公开入门版和 $249/月专业版 | 使用型平台费 | 追踪、评估、存储基础设施 | 为想自建的团队提供低价入口 |
| Arize Phoenix | 免费 / 开源加付费层 | 开源加 SaaS 企业版 | 智能体开发和评估工作流 | 让内部构建在经济性上保持可信 |
| Humanloop(历史) | 关闭前有免费和企业版 | SaaS 企业版 | LLM 评估平台 | 显示独立评估供应商可能被更大实验室整合 |
本表比较公开采购界面,不比较已成交企业合同经济性。定制报价行有意保持定性,因为已审阅页面没有发布标价。
[CP005, CP012, CP018, CP019, CP021, CP023]3.4 切换壁垒与负面观点
负面案例应说得直白。humans& 认为协作、沟通和长周期协调仍被今天的单用户聊天产品低估,这个方向可能是对的。但审阅证据尚未显示 humans& 拥有通向该结果的唯一路径。Incumbent model vendors 已经提供公开推理和企业界面,eval 与可观测性供应商则提供周边工作流基础设施。这让内部自建和多供应商并用都具备可信度。若 humans& 最终创造出黏性切换成本,最可能来自工作流记忆、决策上下文和嵌入式团队行为,而不是基础模型独占。该判断仍是一个假设,因为审阅语料没有公开客户或部署证明。在此之外,市场拥挤、资本充裕且受算力约束。Reuters 把 frontier-lab 招聘市场描述为人才战争,SSI 的 TPU 关系也显示算力访问本身会成为护城河。这些动态抬高了任何隐身实验室的门槛:它不仅要发明差异化产品,还要防守更富有的 incumbents 和模块化替代 stack。[CP005, CP017, CP034, CP039, CP040, CP041]
| 护城河主张或风险 | 威胁 | 严重程度 | 为什么重要 | 缓释 / 尽调追问 |
|---|---|---|---|---|
| 协作优先切入点 | 现有巨头加入记忆、协作或智能体式工作流功能 | 高 | 公开模型供应商已掌握分发,也能快速复制相邻 UX | 要求路线图证据,证明存在独特工作流数据或互动循环 |
| 工作流锁定假设 | 客户可在多个 API 和评估工具之间多归属 | 高 | 可组合工具让内部构建和平行供应商测试变得现实 | 追问留存、迁移摩擦和工作流状态可迁移性证据 |
| 人才护城河 | 前沿实验室从同一个小型研究人才池挖人 | 高 | Reuters 已把前沿劳动力市场描述为人才战争 | 审阅留存数据、组织设计和薪酬策略 |
| 算力护城河 | 更大实验室先拿到战略芯片关系 | 高 | SSI 的 TPU 获取和支持者显示,算力可能成为门槛输入 | 追问已承诺算力、模型训练计划和依赖图 |
| 评估栈吃掉预算 | Scale、Labelbox、Braintrust 和 Arize 无需新模型实验室也能解决相邻痛点 | 中高 | 买方可能先资助测量和 RLHF,再资助新的基础模型伙伴 | 检验 humans& 赢预算靠模型供应商身份,还是靠工作流层 |
| 证据稀缺 | 未审阅到公开客户、部署或合作伙伴证明 | 高 | 没有证明点,护城河主张仍是概念,而非已验证事实 | 承销差异化前,要求客户推荐、试点和部署材料 |
严重程度来自对已审阅证据集的竞争承销判断,不是公司发布的评分。
[CP017, CP025, CP031, CP034, CP039, CP040]紧凑记分卡,区分 humans& 今天真正已证明的部分与仍停留在假设的部分。
这些是基于已审阅语料得出的分析判断,而不是第三方评级。
[CP003, CP034, CP040, CP041, CP045, CP046]3.5 图表
04财务
4.1 收入模型:变现路径看得懂,当前收入看不见
humans& 对哲学化产品方向异常明确,对商业机制却异常沉默。官方首页把公司定位为一家以人为中心的 frontier AI lab,专注强化协作;Reuters 称首款产品预计在 2026 年初推出。TechCrunch 又给出更具体的产品草图,称其是一个用于协作的即时通讯 app 的 AI 版本。公开记录仍没有呈现故事下方最关键的财务层:没有公开定价页,没有企业价目表,没有 API rate sheet,没有收入披露,没有客户数量,也没有已实现合同结构的证据。 这意味着正确的财务框架是可选项,而非 traction。可行收入路径包括 seat subscriptions、按用量计价的 agents 或 APIs、定制企业部署,甚至未来的附属变现;但没有一条能被视为 humans& 当前事实。本章因此把可见的预期产品界面,同实际变现仍未披露的部分分开。实践中,最强的公开结论是:humans& 未来可能有几种可信计费模型,但今天没有公开证据显示其中任何一种已经上线或产生可持续收入。[CI001, CI002, CI003, CI008, CI009, CI010]
| 收入流 | 机制 | 单位 | 当前价值 / 状态 | 质量 | 尽调追问 |
|---|---|---|---|---|---|
| 协作软件订阅 | 面向人机协作的潜在付费工作区或席位套餐 | 席位 / 月 | 公开未披露;未找到 humans& 资费卡 | 今天偏低 | 要求定价材料、SKU 清单、发布计划和试点合同样本。 |
| 基于使用量的模型访问 | 可能按模型调用或智能体额度,为 API 访问或额度计费 | token / credit / action(计费单位) | 未找到 humans& 公开 API 定价;前沿可比对象显示使用量定价很常见 | 今天偏低 | 要求 API 架构、用量计量设计和毛利率假设。 |
| 企业定制合同 | 面向全组织协作工作流的谈判式部署 | 年度合同 / 最低承诺 | 没有当前真实企业合同或条款的公开证明 | 当前低 | 要求提供管线、试点转化数据、MSA 条款和最低承诺。 |
| 广告或赞助访问 | 未来可能为更广泛用户访问提供补贴 | 展示 / 赞助位 | humans& 未披露广告产品;OpenAI 的路径表明,前沿实验室后续可能叠加新的变现层 | 当前很低 | 向管理层确认广告或赞助是否被明确排除。 |
| 研究或战略伙伴资金 | 与算力或生态关系绑定的一次性战略支持 | 股权 / 额度 / 伙伴支持 | 种子轮融资已披露;经常性商业伙伴收入未披露 | 资本层面中等,收入质量层面低 | 董事会报告中应把一次性融资支持与经常性产品收入分开。 |
各行区分可能的变现路径和已验证的当前收入。公开证据只能支持协作愿景存在,不能证明已有商业收款。
[CI008, CI009, CI010, CI021, CI022, CI026]| 价格 / 单位 / 合同 | 标价 vs 实际价格 | 折扣 / 未知项 | 来源 |
|---|---|---|---|
| OpenAI API:GPT-5.5 每 1M tokens 输入 $5 / 输出 $30;GPT-5.4 为 $2.50 / $15 | 公开标价 | humans& 没有公开的同类价目表 | OpenAI API 定价 |
| Slack Pro $7.25/用户/月,按年付;Business+ $15/用户/月,按年付 | 公开标价 | 企业实际折扣未知 | Slack 定价 |
| Google Workspace Business Standard $14/用户/月;Business Plus $22/用户/月 | 公开标价 | 促销折扣和企业定制价格会变动 | Google Workspace 定价 |
| Claude Pro $20/月 | 公开标价 | 已审阅页面未披露企业合同价格 | Anthropic 定价 |
| Notion AI agents:免费试用后,每月每 1,000 credits 收费 $10 | 混合用量定价 | Human& 当前没有智能体计量口径 | Notion 产品 / 价格页面 |
| humans& 当前产品定价 | 未找到公开标价 | 实际价格、折扣和合同机制全部未知 | 本章审阅的 humans& 官方页面 |
本表只把第三方前沿软件价格作为市场锚点。任何一项都不应被误读为 humans& 当前实际价格。
[CI010, CI021, CI022, CI023, CI024, CI025]公开证据支持若干可能的变现路径,但尚不支持判断活跃商业组合。
[CI009, CI021, CI022, CI026, CI027, CI033]4.2 定价和单位经济代理指标:市场锚点存在,但 humans& 的经济性仍属私域
由于 humans& 没有发布自己的定价,公开层面推演变现的最好办法,是借相邻 frontier 和协作市场。OpenAI 证明 model-serving 业务可通过 token pricing 直接按用量变现。Slack、Google Workspace、Anthropic 和 Microsoft 证明 AI 增强的协作或生产力产品也可通过 per-seat plans、企业 upsells 和功能打包变现。Notion 则补充了混合模式:AI agents 和 credits 可以同更广的 workspace subscriptions 并列。这些外部基准不能证明 humans& 会如何收费,但它们确实说明更广市场已经接受 seat-based 和 usage-based 两种模型。 更重要的承保点是,市场定价代理指标解决不了 humans& 的单位经济问题。审阅公开记录仍缺少收入、gross margin、CAC、payback、NRR 和 sales-cycle 数据。因此,即便可比产品证明了商业先例,外部人士仍无法判断 humans& 能否把协作价值转化为足够强的定价,以抵消 frontier-model 成本。对这家公司而言,今天的公开单位经济故事主要是一组缺失字段,加上少数外部参考点,而不是可审计的 operating model。[CI010, CI021, CI022, CI023, CI024, CI025]
| 指标 | 数值 / null | 置信度 | 重要性 | 尽调要求 |
|---|---|---|---|---|
| 收入 / ARR | null | 高 | 没有任何收入基础时,承销判断必须先看现金消耗和上线准备度,而不是增长效率 | 要求提供月度收入桥、已签约 ARR 和已签试点金额。 |
| 毛利率 | null | 高 | 算力密集型产品的用量增长可能看起来亮眼,但贡献利润率仍然偏弱 | 要求按产品拆分毛利率,包括推理、存储和支持成本。 |
| 获客成本 | null | 高 | 如果采用依赖企业教育或创始人大量亲自销售,在可复制需求出现前,CAC 可能很高 | 要求按渠道拆分 CAC,并区分创始人主导销售与规模化销售动作。 |
| 回本周期 | null | 高 | 回本周期决定种子资金是否转化为可持续的市场进入能力 | 要求按队列拆分回本周期,以及试点转付费的时间。 |
| token / 算力经济性代理指标 | OpenAI 旗舰模型公开输出价格大约为每 1M 输出 tokens $15 到 $30 | 中 | 说明模型服务可以支撑定价,但前提是 humans& 能控制推理成本和利用率 | 要求提供每次模型调用的内部 COGS、缓存命中率和 GPU 利用率。 |
| 席位定价代理指标 | 协作套件公开价格集中在每用户每月高个位数到低两位数美元 | 中 | 这意味着,即使协作产品足够强,也可能需要高留存或混合用量费用,才能消化前沿模型成本 | 要求提供目标 ARPU、席位打包方式和高级 AI 功能附加率。 |
null 是有意保留,因为公司尚未公布真正承销单位经济性所需的私有运营指标。
[CI021, CI022, CI023, CI024, CI025, CI029]公开单位经济信号主要指向成本压力,而不是利润率证明。
定性流程,因为 humans& 尚未公布收入、成本或利润率输入。
[CI018, CI021, CI022, CI029, CI031]公开锚点显示市场能够支撑什么,而 humans& 自身已实现的数值大多仍不可得。
区间项目混合了已披露的固定数值、第三方定价锚点,以及用于标记核心指标不可得的零披露标记。
[CI004, CI005, CI021, CI023, CI024, CI030]4.3 资本充足性:种子轮巨大,但算力和招聘仍可能跑在它前面
这轮种子融资足够大,把财务问题从「公司能否启动」改写为「它能否在资本强度复合前,把异常充足的预发布资本转化为产品和可重复经济性」。Reuters 和 TechCrunch 均印证了 $480 million 规模和 $4.48 billion 估值;Crunchbase 称该轮全部为现金且无结构化条款,大部分资金将投入模型训练算力。这一组合很关键:它意味着 humans& 融资主要是为了购买时间、人才和基础设施,而不是扩张一个已经验证的业务。TechCrunch 和 Tracxn 的公开员工数信号也表明,团队很可能在融资后开始扩张,这会在收入可见前增加薪酬和协调成本。 因此,现金充足性不能从标题数字直接推出。一支不到 30 人的团队手握 $480 million 看起来资金极其充裕,但 frontier-model 预算会通过 GPU access、研究薪酬、数据基础设施和发布延迟快速吸收资本。公开记录没有披露现金余额、burn、runway、cloud commitments、partner credits、leases 或类债务义务。因此,财务上诚实的结论是:humans& 近期大概率有相当大的试验空间,但外部人士仍无法核验,在扣除算力和规模化承诺后,标题资本还剩多少真实 runway。[CI004, CI005, CI006, CI011, CI012, CI013]
| 在手现金 / 资本项目 | 月度烧钱 | 现金跑道(月) | 计划资金用途 | 下一轮触发因素 / 债务或项目融资备注 |
|---|---|---|---|---|
| 公开报道的 $480M 种子轮 | null | null | Crunchbase 称,大部分资金拟用于算力训练;员工扩张也很可能发生 | 下一轮融资时点取决于上线速度、算力承诺,以及在资本强度超过种子轮之前试点能否转化。 |
| 该轮被描述为全现金且无结构化条款 | null | null | 结构意味着灵活性,但不等于纪律;未公开披露董事会预算或分期拨付安排 | 询问是否有投资者附函、准备金或战略信用额度影响实际流动性。 |
| Nvidia 参投已披露 | null | null | 释放硬件供应和潜在生态支持具备战略重要性的信号 | 询问是否存在硬件 credits、最低承诺或排他条款。 |
| 交割后公开现金余额 | null | null | 已审阅来源未披露 | 要求提供交割声明、资金管理政策和受限现金明细。 |
| 债务 / 项目融资义务 | null | null | 已审阅来源未发现公开债务、租赁或项目融资包 | 要求提供所有云承诺、租赁和债务类算力安排。 |
历史融资时间线放在公司概况;本表只处理前瞻资本充足性,以及公开证据仍无法验证的事项。
[CI004, CI005, CI006, CI018, CI019, CI020]公开来源显示,这是一家商业化前公司,成本栈先于收入栈变得可读。
[CI008, CI010, CI013, CI018, CI019, CI031]4.4 财务结论:披露缺口是主要尽调障碍
上行案例在逻辑上很直接。humans& 拥有高资历团队、差异化的 augmentation-first 叙事、异常强的初始融资,以及相邻市场中可见的几种可信未来变现模板。若公司能发布协作产品,控制推理和研究成本,并找到一种把 seat value 与可防守 AI utility 结合起来的定价架构,那么这轮种子资金可能足以支撑公司走到有信息量的下一轮融资或早期商业化规模。 下行案例更迫近,证据也更充分。对一家估值处于中个位数十亿美元的公司而言,当前公开界面极薄:没有公开收入、没有定价、没有利润率、没有客户证明、没有 burn、没有 runway、没有合同细节,也没有在本次审阅公开材料中定位到 issuer-level 融资备案 URL。负面评论还凸显真实商业化风险:买方可能更偏好自动化经济性,而不是 humans& 正在倡导的增强哲学。从尽调角度,本章因此不能认可收入质量或 runway;它只能说,作为早期 frontier-AI lab,humans& 看起来资金显著充足,但对严肃财务承保而言披露显著不足。[CI017, CI028, CI030, CI031, CI032, CI034]
| 缺失的私有指标 | 影响 | 精确尽调路径 |
|---|---|---|
| 收入、ARR、客户数和收入结构 | 无法测试收入质量、集中度,也无法判断试点是否已转化为持久需求 | 要求提供月度收入桥、头部客户清单、按阶段拆分的管线,以及与已签合同绑定的客户证明。 |
| 按产品拆分的毛利率和推理成本栈 | 无法判断协作定价最终能否消化前沿模型成本 | 要求提供产品 P&L,并说明 GPU、存储、带宽和支持成本的分摊方法。 |
| 现金余额、烧钱速度和现金跑道 | 即便种子轮标题金额很大,仍无法验证资本充足性 | 要求提供交割现金、月度烧钱桥、已承诺 capex/opex 和 18 个月运营计划。 |
| 算力承诺和伙伴经济性 | 无法知道战略伙伴通过最低承诺降低还是增加实际现金消耗 | 要求提供云 / GPU 合同、credits、预付款、租赁和服务级别承诺。 |
| 定价架构和合同条款 | 没有标价、折扣和用量规则,就无法估算 ARPU 或回本周期 | 要求提供定价文件、报价模板、红线版 MSA 和折扣审批矩阵。 |
| 上线准备度的商业证明 | 上线时间公开,但客户准备度不公开,下一轮融资依赖项因此不透明 | 要求提供产品路线图、试点设计伙伴、上线日历和上线后 KPI 看板。 |
这是本章最重要的尽调材料,因为公开披露明显薄于估值标题。
[CI010, CI017, CI020, CI027, CI028, CI029]4.5 图表
05产品与技术
5.1 产品界面:协作系统已可见,但尚未公开打包
官方和第三方材料在一个重要点上收敛:humans& 并未公开兜售商品化模型 endpoint 或成型企业套件。公司首页把工作描述为一家以人为中心的 frontier AI lab,目标是强化组织和社区;TechCrunch 则给出目前最清楚的功能描述:humans& 正在围绕沟通和协作同时打造产品和模型,早期类比更接近共享沟通和文档界面,而不是独立助手。这些描述足够具体,可以定义可能交付的工作流类别——协调、共享上下文和群体决策支持;但还不足以命名已发布 SKU、受支持部署模型或可供买方采购的功能清单。这意味着本章应描述一个隐身协作 stack,而不是夸大为已发货软件产品。 因此,用户工作流最好从有证据支持的 jobs to be done 来写。问题陈述反复聚焦混乱的多人工作:让团队随时间保持一致,帮助人们表达相互竞争的观点,记住上下文,并改善人与人、人与 AI 工具的协作方式。公开来源不能证明 humans& 已在生产环境解决这些 jobs,但它们确实显示团队正努力让外界看懂:一个 model-plus-interface 层,能记住上下文、提出更好的问题,并充当人类工作流之间的连接组织。[CE001, CE002, CE003, CE004, CE007, CE008]
| 公开可见模块 / 资产 | 主要用户 | 证据支持的状态 | 已验证内容 | 差异化信号 | 尽调缺口 |
|---|---|---|---|---|---|
| 通信 / 协作产品界面 | 团队、群组,也可能包括消费者 | 概念已宣布;没有公开上线证明 | TechCrunch 称 humans& 正在构建一个以沟通和协作为核心的产品和模型 | 目标是占住协作层,而不是只接入现有工具 | 没有公开 demo、定价、SKU 清单或访问路径 |
| 社交智能模型 | 通过产品界面触达同一批用户 | 研究阶段 / 隐身 | 公开材料描述了一种面向社交智能和协调的新模型架构 | 定位从单用户助手扩展到多人上下文 | 未披露架构图、模型卡或评估栈 |
| 记忆和用户理解层 | 重复使用者和团队 | 声称具备能力,但不是已公开的产品功能 | 官网和 TechCrunch 都把记忆与用户理解列为核心 | 持久上下文被包装为更好协作的核心 | 未说明存储、留存、权限或删除控制 |
| 长周期和多智能体 RL 工作流 | 模型训练和研究团队 | 声称采用的训练方法 | 官网和联合创始人引述都提到长周期和多智能体 RL | 可能把该技术栈与单轮聊天机器人优化区分开 | 未公开训练配方、benchmark 套件或算力效率数据 |
| 开源 / 学术协作界面 | 研究人员和开发者社区 | 仅有意图信号 | 官网承诺回馈开源和学术研究;创始人 repo 历史公开 | 支撑招聘,也增强从业者眼中的技术可信度 | 未发现官方 humans& 组织 repo、SDK 或 docs 语料库 |
由于公司仍处于隐身状态,各行只覆盖公开材料直接点名或强烈暗示的模块或资产。本表应被理解为已验证的公开界面地图,而不是完整内部架构。
[CE003, CE006, CE007, CE011, CE012, CE038]| 用户任务 | 当前工作流 | humans& 方案 | 可衡量收益 | 限制 |
|---|---|---|---|---|
| 让群组长期保持一致 | 靠人工会议、聊天线程和反复共享上下文 | 模型加产品,意在协调人与人并保留上下文 | 官方目标是减少上下文重复输入,并提升连续性 | 没有公开证据证明已实际部署或带来可衡量提升 |
| 达成大群体决策 | 有人手工收集意见并协调不同阵营 | Zelikman 描述 AI 通过提问并平衡动机,服务整体利益 | 可能更快形成共识 | 没有公开评估证明决策更好或会议时间更短 |
| 帮助人类与 AI 工具协作 | 用户在不同助手和文档之间切换 | humans& 想做协作层,而不是独立助手插件 | 可能把团队互动和 AI 互动统一到同一界面 | 相对于 Slack/Docs/Notion 的产品边界仍未定义 |
| 保留关于人和项目的有用知识 | 重要上下文被反复输入提示词,或在会话之间丢失 | 记忆和用户理解是明确公开的优先事项 | 可能减少模型反复 onboarding | 没有公开记忆政策、权限模型或删除工作流 |
| 同时支持组织内和非工作场景协调 | 企业协作、消费者消息和计划分别使用不同工具 | TechCrunch 称团队暗示了企业和消费者应用 | 更宽的界面可能扩大可服务用例 | 公开分群、go-to-market 重点和买方证明仍缺失 |
收益仅具方向性。已审阅语料更清楚地解释了公司想解决的工作流问题,而不是已交付的客户结果。
[CE002, CE007, CE008, CE009, CE010, CE012]公开描述的隐身协作技术栈层次,并明确标出未披露的实现边界。
[CE003, CE007, CE009, CE011, CE012, CE032]有证据支撑的图景:humans& 希望协作工作如何穿过其模型加产品层。
[CE007, CE008, CE010, CE012, CE024, CE025]5.2 架构与研究 DNA:已验证要素是训练方法、记忆和社交智能目标
公开技术记录在研究 DNA 上异常丰富,在实现细节上异常稀薄。humans& 明确称该范式需要在长周期强化学习、多智能体强化学习、记忆和用户理解上创新;TechCrunch 也引用联合创始人的说法,描述一种让更多人类和 AI 一起互动的训练制度。这是在方法层面有意义的架构证据:公司在传递信号,核心系统应优化重复互动、群体协调和持久上下文,而不只是优化单轮回答质量。但这并不是模型架构、存储设计、retrieval stack、fine-tuning recipe、eval harness 或部署拓扑披露。 创始人和团队背景增强了这套方法 stack 的可信度,但不能证明商业化。Eric Zelikman 的公开履历把他同 Grok reasoning RL 以及 STaR 和 Quiet-STaR 相连;公司自己的隐藏团队简介显示,团队在 RL、inference、privacy、backend systems 和 data-center engineering 上都有深度。研究论文本身很重要,因为它们显示公司在自生成推理和 token-level internal thought 上有真实脉络;但在 humans& 披露架构、evals 和使用证据前,投资人仍应把从 frontier reasoning research 跨到可靠协作产品的跃迁视为未验证。[CE003, CE004, CE011, CE012, CE015, CE016]
| 层 / 组件 | 作用 | 依赖 | 公开证据质量 | 关键风险 |
|---|---|---|---|---|
| 用户交互 / 协作层 | 承载随模型共同演进的沟通和协调界面 | 产品设计加工作流研究 | TechCrunch 描述给出中等证据 | 没有公开 UI、访问路径或集成地图 |
| 核心社交智能模型 | 解读人、动机和群体上下文 | 创始人的模型训练经验和算力预算 | 官网文案和访谈给出中等证据 | 架构和 eval 标准仍未披露 |
| 记忆 / 用户理解层 | 延续个人、项目和群体上下文 | 依赖未公开的数据处理、检索和权限控制 | 官网文案和联合创始人引述给出中等证据 | 敏感数据留存和隐私设计未知 |
| 长周期 RL 循环 | 针对重复任务和长期结果优化 | 算力、训练数据和评估设计 | 作为既定方法的证据高;实现细节证据低 | 没有 benchmark,难以判断样本效率或稳定性 |
| 多智能体 / 多人训练设置 | 建模协作互动,而不是单用户互动 | 互动数据、仿真环境或人类反馈工作流 | 官网文案和访谈给出中等证据 | 未公开披露智能体、仿真器或 human-in-the-loop 协议 |
| 系统 / 基础设施班底 | 提供推理、kernel、后端和训练支持 | 覆盖 RL、GPU、后端和数据中心系统的专业人员 | 团队履历和员工页面给出中等证据 | 班底强不等于生产可靠性 |
这是基于公开证据拼出的运营模型,不是内部架构文件。各行区分已声明的方法和仍未公开的实现细节。
[CE003, CE004, CE011, CE012, CE015, CE016]公开论点在 humans& 能推出耐用协作产品之前似乎依赖的关键要素。
[CE011, CE015, CE027, CE030, CE035, CE038]5.3 成熟度、路线图与开发者信号:研究阵容强,公开产品完成信号弱
公开成熟度信号呈现出隐身 startup 可预期的混合状态。积极一面是,公司拥有一批研究员和工程师,他们在开源、论文和个人网站上留下了密集足迹。Zelikman 维护着包含 Quiet-STaR codebase 的公开 GitHub;Taylor Sorensen、Saurabh Shah、Alexis Ross 和 Niloofar Mireshghallah 的员工页面都明确把他们同 humans& 相连;至少一个外部 hackathon repo 自称是在 Humans& Product Hackathon 上围绕 collaborative AI agents 构建。这些都是正当的 developer-signal proxies,说明公司已经能引起技术从业者共鸣,而不只是融资壳。 弱项在产品完成度。TechCrunch 1 月 25 日画像仍称 humans& 尚无产品,仍在共同演进模型和界面。此后,审阅语料没有浮现公开 app、SDK、API reference、客户案例、benchmark dashboard 或 release log。因此,最有证据支撑的路线图结论很窄:到 2026 年 1 月,humans& 已公开宣布沟通与协作产品方向;但截至本轮,公开网络仍更像早期研究和招聘界面,而不是生产软件业务。[CE013, CE021, CE022, CE023, CE024, CE025]
| 日期 / 阶段 | 功能或里程碑 | 状态 | 含义 | 来源 |
|---|---|---|---|---|
| 2026-01-19 | 公共网站条款和隐私通知生效 | 已发布 | 基础法律和隐私界面在公开亮相前已就绪 | humans& 招聘 / 法律页面 |
| 2026-01-20 | humans& 公开介绍自己是一家以人为中心的前沿 AI 实验室 | 已发布信息 | 使命和技术论点已对外可见 | humans& 首页 |
| 2026-01-20 | TechCrunch 将目标描述为帮助人们协作的软件,类似 AI 即时通讯应用 | 第三方解读 | 提供公开记录中最清晰的工作流类比 | TechCrunch 种子轮报道 |
| 2026-01-25 | 联合创始人称 humans& 正在构建一个以沟通和协作为核心的产品和模型 | 已确认定位 | 模型和界面在共同开发,而不是分开推进 | TechCrunch 产品特写 |
| 2026-01-25 | 团队称,随着模型提升,产品和界面也在共同演进 | 开发中 | 表明发布周之后产品范围仍在流动 | TechCrunch 产品特写 |
| 截至 2026-06-19 审阅 | 未发现公开 SDK、API 文档、客户案例、安全白皮书或 benchmark 看板 | 公开层面仍缺失 | 公开产品成熟度落后于研究和融资能见度 | 本章已审阅来源语料 |
各行只追踪外部可见的产品开发里程碑。最后一行记录的是已审阅语料中有证据支持的缺失,而不是暗示内部没有动作。
[CE001, CE007, CE009, CE013, CE028, CE032]公开成熟度在论点和人才上最强,在出货证据和企业控制上较弱。
[CE013, CE021, CE023, CE028, CE032, CE040]5.4 信任、隐私与控制姿态:网站隐私披露明确,产品控制仍大多未披露
今天最好的公开信任证据在 marketing site,而不在产品文档里。humans& 称网站主要用于信息展示,不使用 analytics 或 ad-tracking pixels,不使用 tracking cookies 做画像,并把自动数据收集限制在用于可靠性和安全性的基本 server logs。同一通知解释说,求职申请经 Ashby 路由,Google Fonts 会带来有限第三方请求暴露。这些表述有用,因为它们显示公司至少在发布隐私底线并承认第三方处理方。团队构成也通过 Niloofar Mireshghallah 关于 contextual integrity、persistent memory 和 information-flow norms 的论文,包含明确隐私专长。 缺失的是企业买方真正会承保的产品控制层。审阅公开来源没有披露 SOC 2、ISO 27001、model-security whitepaper、data-processing agreement、客户记忆保留政策、abuse monitoring design、公开 uptime page,或量化安全与可靠性 benchmark。对尽调而言,这意味着网站隐私姿态应被视为关于心智模式的积极信号,而不是协作产品本身已可采购、可用于 security-sensitive 部署的证明。[CE027, CE028, CE029, CE030, CE031, CE032]
| 控制 / 质量项 | 状态 | 范围 | 重要性 | 缺口 |
|---|---|---|---|---|
| 网站无 analytics 或广告追踪 | 已验证 | 仅公共网站 | 限制随意网页追踪,也释放隐私意识信号 | 不能证明产品遥测实践 |
| 用于可靠性和安全的基础服务器日志 | 已验证 | 仅公共网站 | 表明公司承认最低限度的运营日志姿态 | 没有产品日志留存或客户隔离细节 |
| 管理、技术和组织保障措施 | 高层级声称 | 仅公共网站 | 表明存在一些基线治理语言 | 没有具名框架、审计或控制库 |
| Ashby 招聘处理方和 Google Fonts 披露 | 已验证 | 招聘流程和网站资源 | 表明隐私通知承认第三方处理方 | 未披露企业供应商清单或产品子处理方 |
| 团队隐私专业能力 | 已验证 | 人才 / 研究班底 | 提高认真处理记忆和信息流问题的概率 | 专业能力不等于已上线的隐私架构 |
| SOC 2 / ISO / DPA / uptime / 公共安全 benchmark | 已审阅语料未发现 | 产品和企业采购界面 | 这些项目是安全敏感买家的典型要求 | 承销生产准备度的重大尽调卡点 |
本表区分网站层面明确披露的内容,以及仍未披露的产品级控制。缺少公开披露并不能证明控制不存在,但确实让尽调负担保持在高位。
[CE027, CE028, CE029, CE030, CE031, CE032]5.5 图表
06客户
6.1 公开客户证明状态
审阅公开语料不支持「早期客户 traction 很强」这类常见 venture shorthand。humans& 官网仍是 announcement-style surface,而非 selling surface:首页标注「Announcing humans&」,文本强调哲学和以人为中心的 AI;按本次审阅,页面没有展示产品定价、客户故事、案例研究、绑定成果的 reference logos 或部署文档。独立发布报道方向一致。TechCrunch 描述公司仍没有公开产品,确切商业形态也尚不清楚;Crunchbase 称除了团队、融资和使命之外,公开可知内容不多。合在一起,最强的证据型表述不是 humans& 没有客户,而是截至 2026-06-19 审阅的公开材料,外部人士无法核验任何具名生产部署、pilot-to-production 转化或可衡量客户成果。这个区分很重要,因为隐身公司可以有私下 design partners,但缺少公开证明会实质性削弱对 reference 质量、采用耐久度和收入集中度的尽调。[CU001, CU002, CU005, CU006, CU007, CU008]
| 细分市场 | 买方 | 用户 | 付款方 | 使用场景 | 公开证据状态 | 缺口 |
|---|---|---|---|---|---|---|
| 企业知识工作团队 | CIO / COO / 职能负责人 | 管理者和跨职能贡献者 | 企业软件预算 | 共享上下文、决策记忆和沟通协同 | 仅为假设;尚未公开披露具名 humans& 企业客户 | 需要具名设计伙伴、工作流地图和生产环境参考访谈。 |
| 消费者或家庭协作小组 | 主要组织者或家庭决策者 | 家人或朋友小组成员 | 消费者订阅或免费增值升级(尚未公开上线) | 多人之间的小组规划、协同和记忆 | TechCrunch 称创始人暗示过消费者应用,但尚无公开上线证据或定价 | 如果消费者是近期真实细分市场,需要上线时间表、定价和使用证据。 |
| 使用多种 AI 工具的产品 / 研究团队 | 产品、研究或创新负责人 | 研究员、PM 和技术团队 | 创新或 R&D 预算 | 多智能体工作流、长周期规划和知识沉淀 | 官方和媒体表述强调模型训练、记忆和用户理解,而不是客户标识 | 需要证明该工作流不只是研究叙事,而是在解决一条现有预算线。 |
| 协同摩擦高的服务或运营团队 | 运营或服务负责人 | 坐席、主管或协调员 | 运营预算 | 升级路由、共享上下文和人在回路的决策支持 | 可比基准显示,高吞吐服务工作流存在需求,但 humans& 尚未披露这是一个已上线细分市场 | 需要试点证据,证明 humans& 能在不引发客户反弹的情况下提升受治理的解决效率。 |
各行是基于官方定位和上线报道推导出的、有证据支撑的买方假设;它们不是已验证的当前客户群。
[CU003, CU004, CU011, CU012, CU013, CU014]| 指标 | 数值 | 日期 | 来源 | 置信度 | 含义 | 缺失分母 |
|---|---|---|---|---|---|---|
| 公开具名客户披露 | 在已审阅语料中找到 0 个公开具名客户参考 | 2026-06-19 | 官方网站 + TechCrunch + Crunchbase | 中 | 牵引力目前缺乏外部验证 | 私下试点仍可能以非公开形式存在。 |
| 公开上线产品状态 | TechCrunch 称 humans& 仍没有产品,也未说明确切商业形态 | 2026-01-25 | TechCrunch | 高 | 客户证据可能仍处于上线前,而不是规模化商业采用 | 未披露候补名单、beta 或活跃账户数。 |
| 公开部署结果指标 | 已审阅的官方、上线和客户证据来源中均未公开披露 | 2026-06-19 | 已审阅语料 | 中 | 没有公开 ROI、使用或续约证据支撑耐久性 | 未给出试点、活跃团队或生产账户的分母。 |
| 企业 AI 生产环境基准 | Deloitte 称,至少 40% 项目进入生产环境的公司数量将在六个月内翻倍 | 2026 年报告 | Deloitte | 中 | 买方越来越要求生产环境证据,而不只看试点 | 该基准覆盖全行业,并非 humans& 专属。 |
| 工作流集中度基准 | Druid 遥测显示,生产环境 AI 使用集中在 FAQ、账户服务、帮助台和职场运营等前门工作流 | 2026 年基准 | Druid AI | 中 | humans& 可能需要一个清晰的工作流切入口,才能赢得预算 | 未披露 humans& 公开工作流遥测。 |
前三行衡量的是公开披露覆盖度,不是该公司实际私下客户数量。
[CU006, CU009, CU010, CU016, CU018, CU033]| 客户 | 细分市场 | 部署 / 使用场景 | 生产环境 vs 试点 | 结果 | 限制 |
|---|---|---|---|---|---|
| 未披露(官方表面) | Unknown | 官方网站讨论协作、沟通和以人为中心的 AI,而非具名部署 | 未说明 | 未公开披露客户结果 | 已审阅官方表面未显示案例研究、证言或部署文档。 |
| 未披露(上线报道) | 暗示过企业和消费者 | TechCrunch 将这一概念类比为 Slack / Google Docs / Notion 式协作上下文 | 未说明;TechCrunch 称当时尚无公开产品 | 未公开披露客户结果 | 上线和融资文章披露的是创始人和投资者,不是参考客户。 |
| 未披露(评测 / 采购证据) | Unknown | 已审阅的客户证据语料包含品类层面的采购基准,而非 humans& 公司评测档案 | 未说明 | 未公开披露有评测背书的部署结果 | 缺少保留下来的公司专属评测证据,并不能证明私下试点为零;它意味着外部参考质量目前还无法核查。 |
本枚举记录的是已审阅语料中缺少公开具名客户证据;不应解读为不存在私下试点的证据。
[CU007, CU008, CU010, CU032, CU033, CU040]公开证据质量在使命上最强,在命名部署、结果具体度和留存可见性上最弱。
单元格概括截至 2026-06-19 已审阅的公开语料;“否”表示保留来源中没有公开证据,不代表私有材料已证伪。
[CU001, CU006, CU007, CU008, CU010, CU020]6.2 买方假设与采购模式
即便存在客户证明缺口,可能买方假设仍然可读。TechCrunch 反复把 humans& 定位为协作和沟通层,而不是泛化单用户聊天机器人,并明确类比 Slack、Google Docs 和 Notion 式多人上下文。官网也用强化组织和社区、重新思考人们如何同 AI 互动,以及围绕长周期和多智能体强化学习、记忆、用户理解构建来呼应这一定位。这些信号首先指向企业知识工作团队:协调、共享上下文和决策记忆都是昂贵痛点;TechCrunch 还称创始人暗示企业和消费者应用并存,这扩大了长期 TAM,却让近期 go-to-market 故事更浑浊。公开招聘信号仍集中在研究、产品和财务,而不是明确投放的销售、客户成功或支持岗位,说明公司商业运营可能仍处于规模化前。实践中,本章因此把细分市场行视为由定位推导出的假设,而非已验证付费客户 cohorts。这对尽调重要,因为细分市场假设能解释产品可能适配哪里,却回答不了核心商业问题:谁签约、谁部署、谁续约、谁愿意做 reference。在这些答案公开或私下提供前,买方地图应被视为尽调的方向性脚手架,而不是可重复 go-to-market engine 的证明。[CU003, CU004, CU011, CU012, CU013, CU014]
| 基准信号 | 来源 | 2026 年观察 | 买方为何在意 | 对 humans& 的含义 |
|---|---|---|---|---|
| 生产环境门槛抬高 | Deloitte State of AI(研究) | 至少 40% 项目进入生产环境的公司数量将在六个月内翻倍 | 买方要看到 AI 摆脱试点泥潭的证据 | humans& 可能需要比普通隐身实验室更快拿出生产环境参考。 |
| 回报高度集中 | PwC AI fitness study(研究) | 受访公司中只有 20% 拿到 74% 的 AI 驱动回报 | 只有新奇 AI 还不够;可衡量价值很稀缺 | humans& 必须展示具体工作流的 ROI,而不是理念。 |
| 人类审批仍占主导 | Bain agentic AI survey(调研) | 目前只有 7% 的公司在生产环境运行完全自主智能体 | 企业买方仍预期要有护栏和人工升级 | 以人为中心的定位有帮助,但前提是转化为具体控制。 |
| 客户信任可能被侵蚀 | CNBC and AnswerConnect(来源) | 消费者反馈 AI 服务转移问题、信任降低,并在支持场景中强烈偏好真人 | 糟糕的自动化会毁掉可参考性和续约 | humans& 需要证明协作能改善结果,同时保留人工升级路径。 |
| 使用集中在前门工作流 | Druid production benchmark(生产基准) | 生产环境 AI 工作负载聚集在 FAQ、账户服务、帮助台和职场运营 | 买方通常从狭窄、高吞吐的工作流入手,而不是泛泛的通用转型 | humans& 在主张广泛协同领导力之前,可能也需要一个同样清晰的初始切入口。 |
这些行是可比采购基准,不是公司专属采用指标。
[CU018, CU019, CU025, CU026, CU027, CU028]基于 humans& 定位和企业 AI 基准,推断从协调痛点到治理化推出的企业采用路径。
这是综合推断的可能买方旅程,不是 humans& 披露的客户漏斗。
[CU003, CU004, CU020, CU022, CU024, CU034]强产品叙事与持久商业采用之间缺失的证明步骤。
节点代表从已审阅客户证明和企业采用基准推断出的尽调里程碑。
[CU007, CU008, CU010, CU016, CU017, CU030]6.3 耐久度、扩张与集中度
行业基准说明了 buyers 在 humans& 声称拥有 durable adoption 前很可能要求什么。G2 和 Capterra 等 review-platform 来源显示,协作和 AI 软件采购围绕具体工作流覆盖、用户评价、部署适配、访问控制和可衡量生产力提升。Deloitte、PwC、Bain、Druid、Microsoft、RAND 和 BCG 的生产基准都指向同一方向:许多组织在试验 AI,但相对少数能把 pilots 转成规模化、产生价值的生产系统;在高后果工作流中,人工监督仍居核心。CNBC 和 AnswerConnect 的负面客服证据进一步说明,若 AI 产品把用户挡回去、移除升级路径或无法交付可信结果,买方会惩罚。对 humans& 而言,这意味着商业瓶颈不太可能只是哲学差异化。闸门证明项更窄、更运营化:具名 design partners、生产 references、工作流特定成果、续约行为、升级设计,以及产品改善协调却不触发其他 AI 部署遭遇过的信任和采用反弹的证据。换句话说,下一步证明不是再写一篇关于团队协作未来的思想文章,而是一张能经受审查的客户事实短清单:账户名称、部署阶段、工作流前后对比、合同结构、续约时点,以及人类必须留在 loop 中的条件。[CU016, CU017, CU018, CU019, CU025, CU027]
| 指标 | 数值 | 细分市场 | 置信度 | 尽调要求 |
|---|---|---|---|---|
| NRR | null | 企业账户 | 低 | 要求按 cohort 以及从首个团队扩展到全组织的路径披露 NRR。 |
| GRR / 流失 | null | 企业账户 | 低 | 要求披露任何未扩展试点的 logo 流失、席位流失和流失原因。 |
| 合同期 / 续约时间 | null | 企业账户 | 低 | 要求提供 MSA/SOW 条款样本、续约日期和最低承诺。 |
| 可作参考客户程度 / 满意度 | null | 所有已上线账户 | 低 | 要求提供前十名可作参考的用户、如有跟踪则提供 NPS/CSAT,以及升级路径示例。 |
| 消费者重复使用(如适用) | null | 消费者或家庭用户 | 低 | 如果消费者是真实上线路径,要求提供 DAU/WAU/MAU、留存曲线和付费转化。 |
Null 表示该指标未在已审阅语料中公开披露,不表示底层数值为零。
[CU016, CU030, CU031, CU036, CU037]| 扩张驱动因素 | 集中度风险 | 影响 | 尽调路径 |
|---|---|---|---|
| 团队到组织的工作流扩张 | 尚无公开证据证明小团队试点会扩展为更广泛的企业部署 | 高 | 要求按账户披露试点到生产的漏斗、席位扩张和按工作流拆分的产品使用。 |
| 大型设计伙伴账户 | 如果早期收入集中在少数灯塔账户,收入集中度可能较高 | 高 | 要求提供按 ARR 或承诺支出排序的前十账户,包括最大客户收入占比。 |
| 平台 / 合作伙伴依赖 | GTM 可能依赖底层模型、云或工作流集成,而买方可能将这些视为替代品 | 中 | 要求提供依赖地图、排他性条款、转售安排和切换成本证据。 |
| 企业 vs 消费者拆分 | 同时服务企业和消费者,可能拖慢 ICP 清晰度和销售可复制性 | 中 | 要求管理层说明上线顺序、目标收入结构和按细分市场的资源配置。 |
| 信任和升级设计 | 客户反感治理不当的 AI 服务体验,可能拖慢续约和参考客户转化 | 中 | 要求提供升级政策、人工覆盖阈值,以及产品改善而非转移问题解决的案例。 |
风险以尽调假设表述,因为公开的集中度、续约和扩张数据缺失。
[CU017, CU019, CU025, CU027, CU028, CU029]07风险
7.1 商业化不透明是排名最高的投资人风险
核心承保问题不是 humans& 缺少履历或资本,而是公开记录对融资的证明远强于对产品采用的证明。Reuters 2026 年 1 月称公司预计当年早些时候发布产品,多家媒体确认 $480 million 种子轮和 $4.48 billion 估值。但本章审阅的官方界面主要仍是法律、隐私和招聘信息,而不是公开定价页、客户清单、产品手册或使用仪表盘。因此,Reworked 的怀疑性解读很重要:它认为 humans& 在没有产品的情况下达到独角兽地位,估值反映的是履历和战略定位,而不是已验证 market fit。这个缺口重要,因为在这个价格上,延迟并非中性。如果第一批可见商业化证据来得晚,或到来时没有客户证明,下行不只是执行变慢;而是一个已经按三个月大 frontier lab 的异常雄心定价的故事,面临 multiple compression。[CR001, CR002, CR003, CR004, CR005, CR006]
| 失效模式 | 可能性 | 严重性 | 缓解成熟度 | 剩余敞口 | 未解决缺口 |
|---|---|---|---|---|---|
| 商业上线延误,或上线后缺乏耐久客户证据 | 高 | 高 | 低 | 高 | 公开证据仍更多指向未来上线意图和融资动能,而不是已上线部署指标 |
| 产品质量或安全测试落后于模型野心 | 中 | 高 | 低 | 高 | 未看到 humans& 专属的公开评估制度、事件历史或安全框架 |
| 官方表面过于不透明,难以支撑企业尽调 | 高 | 高 | 低 | 高 | 已审阅官方页面强调法律 / 隐私基础,而不是架构、控制或使用证据 |
| 多智能体 / 长周期产品愿景比叙事显示的更难运营落地 | 中高 | 高 | 低中 | 中高 | 公开来源描述了论点,但没有给出基准路径或发布里程碑 |
| 运营工作量跑在不到 30 人的团队前面 | 中高 | 高 | 低 | 高 | 没有公开团队配置图显示产品、安全、GTM、法律和支持职能的覆盖情况 |
本登记表强调执行上可见的失效模式,而不是假设性的模型灾难,因为投资者眼前的风险是商业化延迟或疲弱。
[CR004, CR005, CR006, CR007, CR010, CR012]基于当前公开记录,按可能性、影响、缓释成熟度和剩余暴露对 humans& 主要风险排序。
高 / 中 / 低分档是基于已引用证据和缓释可见度的相对风险排序,不是点概率。
[CR004, CR005, CR007, CR012, CR019, CR020]7.2 对 Eric Zelikman 和微型精英团队的关键人物依赖让执行风险居高不下
创始阵容足够强,能解释投资人的兴奋,但也让公司变得脆弱。截至 2026 年 5 月下旬,公开员工数信号仍指向一家不到 30 人的公司,尽管公司的雄心覆盖前沿模型研究、产品设计、安全和商业化。Bizprofile 的加州登记镜像进一步显示,同一实体中 Eric Zelikman 同时担任 CEO、CFO 和 Secretary;即便这些头衔最终会在实践中拆分,这仍显示形式上的集中。OpenReview 将 Zelikman 标注为 Stanford 博士生,Reuters 又把他与 xAI 以及聚焦推理的强化学习工作联系起来,因此关键人风险不在于创始人质量,而在于创始人负荷。公开记录仍没有显示具名的合规负责人、商业运营负责人、安全负责人,或创始团队之外的深层管理梯队。对一家试图做长周期协作产品的前沿 AI 公司而言,这意味着少数人可能同时扛着产品愿景、模型战略、治理、融资和招聘。[CR013, CR015, CR016, CR017, CR018, CR019]
| 角色 / 职能 | 依赖或缺口 | 可能性 | 严重性 | 缓释措施 | 尽调路径 |
|---|---|---|---|---|---|
| Eric Zelikman / 创始人兼 CEO | 正式高管集中度和公开使命叙事都高度压在一个人身上 | 高 | 高 | 顶尖技术信誉和强大的联合创始人班底 | 索取组织架构、授权负责人、继任覆盖和董事会运作节奏 |
| 创始团队广度 | 相比公司愿景,公开可见班底仍偏创始人驱动 | 中高 | 高 | 联合创始人拥有 Anthropic、xAI、Google 和 Stanford 背景 | 索取直属汇报线,以及法务、安全、基础设施和 GTM 负责人名单 |
| 团队扩张 | 到 2026 年 5 月下旬,公开员工数信号仍低于 30 人,意味着运营底座依旧很小 | 高 | 高 | 种子轮资金可支撑招聘,Ashby 职位页面也显示公司在积极招人 | 索取按职能拆分的当前员工数和已接受录用邀约的候选人管线 |
| 合规 / 安全归口 | 公开层面看不到具名负责人 | 中 | 高 | 如果管理层优先推进,可借助外部框架和顾问 | 索取具名问责高管和当前评审委员会流程 |
| 专项人才招聘 | 前沿算力和数据管线人才市场薪酬上涨,且容易被挖角 | 高 | 中高 | 声望和资本能吸引部分候选人 | 索取留任计划、offer 接受率,以及对少数关键招聘的依赖 |
本表区分人才质量和组织韧性;风险不在于创始人弱,而在于带宽窄、执行负载集中。
[CR013, CR015, CR016, CR017, CR018, CR019]7.3 算力、Nvidia 和前沿人才市场带来集中的依赖风险
humans& 并没有把自己呈现成一款轻量级工作流应用;公开融资报道反复把它描述为一个算力密集型的前沿项目。Crunchbase News 报道称,公司计划把种子轮资本的大部分用于模型训练算力,Reuters 和其他媒体也突出提到 Nvidia 是投资方。TechStartups 更进一步称 humans& 将在硬件和软件两端与 Nvidia 紧密合作,但这一细节没有在公司官网得到印证,因此应谨慎看待。外部市场背景放大了风险。CNBC 2025 年关于人才战的报道称,前沿模型专家池子很小,如今能拿到数百万美元薪酬包,因为只有少数公司负担得起成本达数十亿美元的模型建设。Reuters 和 Forbes 还指出,雇主正在优先争夺细分 AI 技能,懂前沿实验室数据流水线的人才短缺。对 humans& 来说,含义很直接:即便种子资金充足,一旦算力获取、训练数据运营和顶尖技术招聘同时收紧,时间表仍会滑坡。[CR020, CR021, CR022, CR023, CR024, CR025]
| 依赖 | 交易对手 | 角色 | 集中度 | 失效场景 | 严重性 | 缓解措施 | 剩余敞口 |
|---|---|---|---|---|---|---|---|
| 训练算力 | Nvidia 和更广泛的 GPU 供应链 | 资本密集型模型训练和加速 | 高 | 获取或经济性收紧,拖慢模型进展或迫使重新排序优先级 | 高 | 大额种子轮提供购买力和战略投资者资源 | 高 |
| 硬件 / 软件协作主张 | Nvidia 或同等基础设施合作伙伴 | 支撑模型开发路线图 | 中高 | 合作伙伴支持比宣传显得更窄 | 中高 | 在看到合同前,保守看待未经佐证的合作细节 | 中高 |
| 招聘平台和外部供应商 | Ashby 和网站服务提供商 | 招聘工作流、托管、安全和运营 | 中 | 关键工作流位于核心领域之外,并在内部运营成熟前扩张 | 中 | 使用供应商很正常,公开法律页面也承认这一点 | 中 |
| 投资者生态预期 | SV Angel、Nvidia、Bezos、GV 及其他支持者 | 资本获取与战略信号 | 中高 | 支持方预期推动范围或节奏跑得比产品准备更快 | 高 | 知名投资人能帮助招聘和合作伙伴拓展 | 中高 |
| 模型 / 云栈披露 | 未披露供应商或内部技术栈 | 推理、训练、可靠性与合规 | Unknown | 管理层对第三方模型或云条款的依赖,高于公开材料所显示的程度 | 高 | 除高层愿景和资本配置线索外,公开层面看不到其他证据 | 高 |
最后一行刻意标为未知,因为已审阅的公开记录尚未披露 humans& 是在训练自己的前沿模型、实质性改造其他供应商模型,还是主要搭建应用层。
[CR020, CR021, CR022, CR023, CR024, CR025]梳理横在 humans& 与可持续商业执行之间的关键人才、基础设施和外部机构。
[CR016, CR019, CR020, CR021, CR022, CR024]7.4 前沿 AI 监管、安全预期和版权不确定性,可能超出公司公开控制面的承载范围
这里的监管风险并非基于任何已知针对 humans& 的执法行动,而是来自前沿模型雄心与当前可见合规证据之间的错配。公司的法律页面已经提到出口管制义务、知识产权限制、供应商处理以及公司交易中的转让权;这构成基础法律界面,但还不是前沿模型治理包。与此同时,外部门槛已经抬高。European Commission 称,GPAI 义务已于 2025 年 8 月 2 日开始适用,执法权将在 2026 年 8 月 2 日进一步强化,包括系统性风险模型通知和严重事故报告。NIST 的生成式 AI profile 聚焦治理、内容溯源、部署前测试和事故披露;2026 International AI Safety Report 则称,可靠的部署前测试正在变难,因为高级模型可以利用评估漏洞。U.S. Copyright Office 也仍在处理训练数据、合理使用和许可问题。如果 humans& 正在训练或实质性修改前沿模型,相对于 GPAI 提供方周围正在形成的安全、透明度和版权预期,其现有公开文档仍显单薄。[CR029, CR030, CR031, CR032, CR033, CR034]
| 风险 | 司法辖区 / 范围 | 状态 | 可能性 | 严重性 | 缓解措施 | 剩余敞口 | 尽调路径 |
|---|---|---|---|---|---|---|---|
| GPAI 透明度、事件报告和系统性风险合规 | 欧盟 / 任何触及 GPAI 规则的模型活动 | 已生效,并将在 2026 年前后收紧 | 中高 | 高 | 大额资本基础和更广泛执法前的准备时间 | 高,直到管理层说明其是否训练或实质性修改前沿模型,以及谁负责合规 | 要求提供 AI Act 适用性备忘录、模型分类、安全负责人和任何 AI Office 沟通材料 |
| 版权和训练数据模糊性 | 美国和全球权利人敞口 | 政策和诉讼环境已经活跃 | 中 | 高 | 未发现专门针对 humans& 的不利案件;该问题是全行业问题,并非特定事件 | 高,因为公开材料未解释训练数据来源、授权或权利升级流程 | 要求提供训练数据政策、退出处理、授权清单和外部律师备忘录 |
| 围绕以人为中心主张的营销 / 证实风险 | 美国 / 欧盟商业主张和采购尽调 | 当前风险 | 中 | 中高 | 相比部分 AI 炒作,公司使命表述较谨慎,且尚未公开过度承诺功能 | 中高,因为产品性能、客户结果和安全控制的公开证据仍然薄弱 | 要求提供上线指标、客户参考、基准包和明确的主张审核流程 |
| 出口管制、IP 和内容使用义务 | 美国法律条款和跨境使用 | 当前合同风险 | 中 | 中 | 条款和隐私页面已经承认出口管制和 IP 约束 | 中,因为法律表面存在,但未描述运营处理方式 | 要求提供出口筛查工作流、数据处理控制和事件 / 升级程序 |
各行按投资者剩余下行排序,而不是按法律新颖性排序;每一项都基于已审阅公开材料或清楚标注的尽调推断。
[CR007, CR008, CR009, CR029, CR030, CR031]展示证明、监管和依赖风险如何传导为发布放慢、转化走弱、资本强度上升和估值受损。
[CR006, CR012, CR019, CR028, CR030, CR031]7.5 公开缓释因素存在,但决定性证据仍藏在尽调要求和否决标准之后
文件里确实有实质缓释因素:humans& 融到异常大额资本,组建了高水准创始团队,提出差异化协作论点,维护法律 / 隐私界面,并保留外部招聘渠道。这些并不轻。但它们不同于硬证据,不能证明公司能够安全商业化、穿越人才战招聘、锁定耐久算力,并在规模化时满足正在成形的 GPAI 义务。因此,正确的尽调姿态应是有条件判断,而不是二元判断。投资人应坚持索取最新产品与部署备忘录、模型所有权和提供方栈地图、具名责任人与评估流程的安全合规包、覆盖 Eric Zelikman 之外继任安排的组织架构图,以及解释公司如何避免毛利或时间表冲击的算力采购视图。如果管理层无法拿出这些材料,审慎解读不是上行空间消失,而是估值已经要求一层公开记录尚未支撑的执行证明。[CR007, CR019, CR028, CR039, CR047, CR048]
| 风险 | 可监测触发项 | 阈值 / 事件 | 行动含义 |
|---|---|---|---|
| 商业化不透明 | 公开发布与客户证明 | 尽管预计 2026 年 1 月发布,管理层仍无法展示已上线产品、具名部署或使用 KPI | 将估值支撑视为破裂,把风险从执行风险下调为投资逻辑破裂风险 |
| 关键人物集中 | 班底扩张与授权归口 | 没有可信的组织架构、继任计划,或安全、GTM 和基础设施的非创始人运营负责人 | 不要只凭创始人履历承销规模化假设 |
| 算力依赖 | 采购与供应商多元化 | 没有具体算力计划、没有披露应急方案,或存在单一瓶颈供应商关系的证据 | 假设时间线和利润率敏感性明显差于表面现金余额所暗示的水平 |
| 安全 / 监管准备 | 治理材料包 | 没有模型分类、评估流程、事件响应路径或 AI Act 归口备忘录 | 在讨论价格前,把前沿 AI 合规列为阻断性尽调事项 |
| 估值预期 | 路线图到证据的桥 | 管理层无法把当前估值预期与发布、客户和商业化里程碑对齐 | 将下行情形建模为倍数压缩,而不只是收入时间推迟 |
| 合作伙伴 / 供应商暴露面 | 关键工作流韧性 | 关键招聘、托管或模型供应商流程由第三方承担,且没有书面备份计划 | 提高依赖折扣,并要求供应商风险审查 |
这些终止标准刻意设计成可监测项:每一项都要求文件、指标或负责人,而不是泛泛安抚。
[CR012, CR019, CR028, CR039, CR047, CR048]7.6 附录
08估值
8.1 投资论点、反论点与建议
humans& 作为故事很容易理解,作为业务却很难承保。多头论点对这么年轻的公司而言异常强:来自 Anthropic、xAI、Google、OpenAI 和 Stanford 的精英校友;$480 million 的全现金种子轮弹药;Nvidia 和其他明星投资人;以及一个显然奖励前沿 AI 期权价值的 2026 年市场。公开可比公司显示,在商业证明到来前,投资人曾愿意为稀缺创始人密度支付巨大价格。Thinking Machines Lab 在仍基本处于产品前阶段时,种子轮报价从报道的 $10 billion 移至 $12 billion;SSI 在没有商业发布的情况下达到 $32 billion。在这种背景下,humans& 的 $4.48 billion 估值不是类别错误;它是同一笔前沿实验室交易的较小版本。 反论点是,公开证据仍停在履历、理念和资本。humans& 没有留存公开收入、没有具名客户、没有公开董事会材料、没有定价、没有经审计员工数,也没有清晰公开条款书来解决所称 $4.48 billion 究竟是投前还是投后。负面报道直言,公司在证明产品前就以独角兽以上价格出现;公开 AI 公司可比案例也显示,支撑这个头条的运营证明很少。仅基于公开证据,建议是继续研究,中等信心,高风险,估值立场为偏高。这个价格或许可以用稀缺驱动的期权价值解释,但尚不能由牵引力支撑。[CV001, CV002, CV003, CV005, CV006, CV007]
| 建议 | 置信度 | 风险评级 | 估值立场 | 决策含义 |
|---|---|---|---|---|
| 继续研究 | 中 | 高 | 偏高 | 不要只凭公开证据承销 $4.48B;必须要求产品、客户和轮次文件证明,或显著更好的入场价格。 |
单行投委会摘要;决策含义明确反映公开牵引力和条款清单透明度的缺失。
[CV002, CV006, CV033, CV042, CV043]| 论点 | 类型 | 公开证据 | 什么会改变判断 |
|---|---|---|---|
| 顶尖创始人密度和投资人质量,创造了真实的前沿实验室期权价值。 | 正方 | 创始人来自主要实验室;头部投资人参与;2026 年市场仍愿意奖励人才驱动的前沿创业公司。 | 创始班底变弱或战略投资人流失,会显著削弱稀缺价值。 |
| $480M 全现金种子轮,显著提高了 humans& 购买算力和时间的能力。 | 正方 | Crunchbase 称大部分资金流向算力;2026 年市场报告显示电力、芯片和产能仍然稀缺。 | 如果算力稀缺很快缓解,或 humans& 缺少特权采购条款,这一溢价会收缩。 |
| humans& 的定价处在真实的超级种子前沿队列中,而不是孤立个案。 | 正方 | Thinking Machines 和 SSI 显示,投资人愿为尚未有牵引力的前沿实验室支付 $10B-$32B。 | 如果这些可比公司明显降级重估,humans& 的参照组会变弱。 |
| 公开层面仍没有收入、客户、定价或董事会披露。 | 反方 | 官方网站和负面报道展示的是使命和融资,而不是运营证明。 | 具名客户、试点、定价和治理材料会提高可支撑性。 |
| 保留下来的公开证据没有明确说明 $4.48B 报价是投前还是投后。 | 反方 | 保留来源重复了报道估值,但没有签署条款。 | 种子轮条款清单和股权结构表能厘清稀释和持股计算。 |
| 公开 AI 应用可比公司已经以更低市值披露收入。 | 反方 | C3.ai 在 FY2026 产生 $250.3M 收入,但公开市值约为 $1.49B。 | 如果 humans& 展示真实使用量和差异化产品,公开市场可比折价就没那么相关。 |
各行把愿意支付溢价的稀缺性理由,与仍让当前价格难以背书的公开证据缺口分开。
[CV004, CV010, CV011, CV014, CV020, CV021]创始人稀缺、算力稀缺、缺少牵引力和条款不透明,如何共同指向“继续研究”的建议。
[CV004, CV006, CV010, CV014, CV024, CV036]8.2 融资背景:先例融资、人才溢价与重置成本逻辑
核心估值问题不是 humans& 是否融到了真金白银;它已经融到。更难的问题是,这轮钱代表什么。最佳公开证据显示,投资人买的是人才、时间和算力获取的组合,而不是收入或客户证明。Crunchbase 报道称种子轮为全现金,大部分资金将用于算力。J.P. Morgan 和 Colliers 都描述了一个 2026 年环境:推理需求、GPU 内存、电力和数据中心容量都稀缺,而且越来越多被提前锁定。这让一张大额前置资产负债表具备战略意义:在供给受限的前沿 AI 市场,资本不只是燃料,也是插队能力。 不过,重置成本逻辑本身无法一路推到 $4.48 billion。公开记录显示团队仍小——发布时约 20 人,5 月下旬估计 28 人——因此所称轮次隐含每名员工约 $160 million 至 $224 million 的异常价值区间。只有当投资人相信团队能把稀缺研究人才和算力采购能力足够快地转化为差异化前沿平台,这个数字才说得通。公开可比公司支持这种下注的存在,但不支持其确定性。Thinking Machines 和 SSI 说明,市场确实会为创始人密度和前沿实验室期权付高价;但缺少公开条款书、董事会结构和优先权栈,外部投资人仍无法准确判断,为支撑这个价格究竟买到了哪些经济权利。[CV004, CV005, CV008, CV016, CV020, CV021]
用敏感性条展示 humans& 相对公开可比公司、可支撑中点和前沿巨额种子轮标记的位置。
数值四舍五入至最接近的 $10M,用于显示相对位置,不是 EV 对账。
[CV002, CV010, CV011, CV014, CV028, CV029]用投委会风格记分卡评估 humans& 的人才、资本强度、产品证明、治理可见度、估值可支撑性和证据质量。
分数是基于留存公开证据的 1-5 启发式判断,不是外部评级。
[CV006, CV020, CV024, CV029, CV030, CV031]8.3 牵引力调整情景与公开可比交叉检查
没有留存来源能让本章把 humans& 与收入挂钩。这很关键,因为公开市场 AI 参照已经展示了收入披露后投资人可以买到什么。C3.ai 报告 2026 财年收入 $250.3 million,6 月中旬公开市值仍只有约 $1.49 billion。相比之下,CoreWeave 在有公开申报和公开市场流动性的情况下,市值约 $64 billion,显示一旦披露和规模存在,投资人会多么强烈地奖励稀缺 AI 基础设施。humans& 尴尬地夹在两端之间:明显贵过一家披露收入的上市应用 AI 厂商,明显便宜过一家公开基础设施赢家,同时也比两者都早得多。 这一缺口解释了为什么牵引力调整后的基准情景应低于所称轮次价格。合理的公开证据区间今天约为 $2.0 billion 至 $3.5 billion:高到足以尊重创始人质量、投资人质量和算力稀缺,低到足以折价缺失的产品、客户和治理证明。多头情景假设 humans& 交付一款差异化协作产品,把早期用户转化为真实设计伙伴或客户,并继续像稀缺前沿实验室一样被估值;只有到那时,当前报价才开始显得公平。空头情景更简单:如果公司到 2027 年仍处于产品前阶段,市场可能把它重估到公开 AI 应用可比公司加现金,而不是前沿神秘感。[CV006, CV012, CV015, CV025, CV026, CV027]
| 情景 | 假设 | 估值 / 回报逻辑 | 关键风险 | 概率信号 |
|---|---|---|---|---|
| 牛市 | humans& 在 2026 年推出差异化协作产品,获得可信试点或客户证明,并保持前沿实验室稀缺性支撑。 | $4.5B-$6.5B;如果 humans& 开始像一个规模更小的 TML 式平台押注,当前轮次价格可显得合理,甚至略保守。 | 需要产品市场信号、招聘成功和持续算力获取。 | 如果近期没有公开证明,概率偏低。 |
| 基准 | 未来 12-18 个月内,humans& 技术上仍令人印象深刻,但商业披露不足。 | $2.0B-$3.5B;缺少牵引力和文件,公开证据支持相对报价轮次给出实质折扣。 | 没有收入 / 客户证明;估值基础未解;治理仍不透明。 | 最符合今天的公开证据。 |
| 熊市 | humans& 到 2027 年仍未推出产品,或被迫在缺乏清晰牵引力时再次融资。 | $1.0B-$2.0B;市场把公司重估到更接近公开 AI 应用可比公司,加上现金和团队价值。 | 后续轮次低于报道标记、招聘放缓,或算力经济性偏弱。 | 如果里程碑生成滑坡,风险较大。 |
情景区间仅由公开证据推导,是作者估算。它们是可支撑性区间,不是对公司实际内部 409A 或下一轮定价的断言。
[CV024, CV029, CV030, CV038, CV039, CV040]| 可比对象 | 类型 | 指标 / 估值 | 相关性 | 局限性 |
|---|---|---|---|---|
| humans& 报价种子轮(2026 年 1 月) | 私有前沿实验室融资 | $480M 种子轮,报价估值 $4.48B | 被审资产的直接市场出清价格。 | 公开记录没有厘清投前 / 投后,或条款清单保护。 |
| Thinking Machines Lab(2025 年 6-7 月) | 私有前沿实验室融资 | $2B 种子轮,最初报道 $10B,后确认 $12B | 最接近的可比对象:由顶尖前 OpenAI 人才领导、仍处早期的前沿实验室。 | 绝对估值更大,创始人光环更广;产品路径仍不同。 |
| Safe Superintelligence(2025) | 私有前沿实验室融资 | $2B 轮次,估值 $32B | 显示投资人在发布前愿为稀缺前沿团队拉伸到多远。 | SSI 的使命更单一,并获得 Google Cloud 明确 TPU 支持。 |
| C3.ai(2026 年 6 月) | 公开 AI 应用公司 | FY2026 收入 $250.3M;市值约 $1.49B | 可用来校验公开市场如何给已披露 AI 软件收入定价。 | 公开市场应用厂商,不是发布前前沿实验室。 |
| CoreWeave(2026 年 6 月) | 公开 AI 基础设施公司 | 市值约 $64.35B,并有活跃 SEC 文件披露 | 显示当 AI 基础设施需求真实且已披露时,稀缺价值可以极高。 | 暴露在基础设施,而不是协作模型软件;运营证明多得多。 |
| 2026 年前沿融资环境 | 行业基准 | 2026 年 Q1 基础 AI 融资超过整个 2025 年的两倍,并集中在少数巨头 | 解释为什么投资人愿在 2026 年激进定价期权。 | 宏观背景不能替代公司自身牵引力。 |
本表混合了私有融资、公开公司和市场背景,因为对于一家发布前、以协作为先的前沿实验室,不存在单一完全同类可比对象。
[CV002, CV010, CV011, CV014, CV017, CV018]基于公开证据可支撑的 humans& 区间,与当前报价轮对照。
区间是作者基于留存公开证据得出的估算;公开记录没有披露融资条款经济性或牵引力指标,因此刻意保持较宽。
[CV033, CV038, CV039, CV040, CV041, CV042]8.4 退出准备度、否决触发器与仍缺失的证据
本章不应假装公开证据已把 humans& 收窄到一个精确内在价值。最重要的缺失项都在文件层面:种子轮条款书和股权表、董事会构成和保护性条款、算力采购合同或额度、产品使用证据,以及任何试点或客户推荐。没有这些,投资人无法判断所称轮次究竟只是昂贵的普通股文件,还是带有实质下行保护的结构化证券,抑或是一个假设未来商业里程碑的动量标记。 这也塑造退出逻辑。在所称价格下,最清晰的上行路径不是金融工程,而是创造里程碑。humans& 可能需要一个可见产品、协作式 AI 不只是哲学切口的证明,以及种子轮买到的不止时间的证据。否决触发器也相应清晰:长期无产品、后续轮次低于所称估值、无法锁定或高效使用算力,或有证据显示企业买家更偏好自动化经济性,而不是公司以人为中心的叙事。在这些问题厘清前,正确投资姿态是密切观察公司、要求更完整尽调,并避免仅凭公开证据就称当前价格公平。[CV006, CV024, CV033, CV034, CV037, CV042]
| 触发项 | 阈值 | 对投资逻辑的传导 | 行动含义 |
|---|---|---|---|
| 看不到产品或试点证明 | 下一融资窗口前,仍没有公开产品、设计伙伴或试点证据。 | 这一轮会变成纯粹的时间和人才押注,期权性收缩。 | 向下重估,避免按报价一级市场标记付费。 |
| 下轮降价或持平 | 下一次外部定价处在或低于报价 $4.48B 水平。 | 表明稀缺性溢价没有复利成证明。 | 视为全面估值重置,并从公开基本面重新承销。 |
| 算力获取不及预期 | 尽管计划重算力,却没有特权额度、供应获取或高效利用的证据。 | 削弱巨额种子轮最清晰的战略理由之一。 | 削减期权价值溢价,并突出现金消耗风险。 |
| 自动化经济性压过协作逻辑 | 买方反馈或早期用例显示,客户更偏好劳动力替代,而不是 humans& 的增强叙事。 | 削弱核心产品切入点,并压窄定价权。 | 除非客户证明反驳,否则把立场转向昂贵。 |
| 治理和纸面条款仍不透明 | 即便公司寻求更多资本,仍没有董事会、股权结构或条款清单可见度。 | 无法准确建模持股、稀释和下行保护。 | 无论技术多令人兴奋,建议都维持继续研究。 |
这些触发项是公开证据代理指标,用来判断稀缺性驱动的估值故事何时开始破裂。
[CV024, CV033, CV034, CV037, CV041, CV043]| 主题 | 缺失证据 | 为什么重要 | 负责人或尽调路径 |
|---|---|---|---|
| 轮次机制 | 已签署种子轮条款清单、精确投前 / 投后基础、期权池处理和投资人权利。 | 决定出让持股、稀释计算,以及报道价格是否高估普通股价值。 | 直接向管理层和领投方索取融资文件。 |
| 股权结构和治理 | 董事会构成、观察员权利、保护性条款和优先权结构。 | 需要用来理解控制权、清算瀑布,以及能否顺利募集后续资本。 | 索取股权结构表导出、章程、投资人权利协议和董事会材料。 |
| 算力经济性 | 云或芯片合同、额度、预留容量和目标利用率。 | 支撑或削弱重置成本与稀缺性溢价视角。 | 索取供应商协议、算力预算和基础设施路线图。 |
| 产品和客户证明 | 原型访问、产品路线图、试点管线,以及任何具名设计伙伴或付费用户。 | 没有这些,估值无法绑定商业证据。 | 索取演示、使用量仪表盘、试点合同和客户访谈。 |
| 运营模型 | 员工数计划、现金消耗、续航期和基于里程碑的招聘计划。 | 检验 $480M 是否买来足够时间,让公司在下一轮前跑出真实证明。 | 索取月度现金消耗桥、招聘计划和 12-18 个月里程碑地图。 |
每一行都是文件驱动的请求,目标是把今天的叙事估值转成可尽调的承销案例。
[CV006, CV024, CV033, CV034, CV035, CV044]免责声明
本尽调报告由 AI 研究代理基于截至 2026-06-19 的公开来源生成,不构成投资建议。humans& 是一家隐身期私营公司,重要承销输入仍未披露,包括当前产品指标、客户、收入、安全控制和详细融资条款;应直接通过管理层材料和一手尽调验证。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | humans& publicly introduced itself on January 20, 2026 as a human-centric frontier AI lab. | 高 | SO001, SO009 |
| CO002 | humans& says progress happens when people understand one another, build trust, make connections, and work together. | 高 | SO001, SO010 |
| CO003 | humans& says AI should serve as deeper connective tissue that strengthens organizations and communities. | 高 | SO001, SO009 |
| CO004 | humans& says its approach requires innovations in long-horizon and multi-agent reinforcement learning, memory, and user understanding. | 高 | SO001, SO009, SO015 |
| CO005 | humans& says it will tightly integrate science and product development. | 中 | SO001 |
| CO006 | The company website and legal notices show humans& ai, inc. operating a U.S.-based informational site and Ashby-hosted recruiting flow by January 2026. | 中 | SO002, SO003 |
| CO007 | Retained public sources consistently place humans& in the Bay Area even when the exact city label varies. | 中 | SO016, SO022, SO024 |
| CO008 | A California-registry mirror lists the principal and mailing address of Humans& Ai, Inc as 601 Marshall Street, Redwood City, CA 94063. | 中 | SO022 |
| CO009 | Other public profiles and coverage describe humans& more broadly as San Francisco-based. | 中 | SO016, SO025 |
| CO010 | The safest current location description is Bay Area company with a public Redwood City address and a broader San Francisco operating narrative. | 中 | SO016, SO022, SO024 |
| CO011 | Multiple retained sources say humans& was founded in September 2025. | 中 | SO010, SO014, SO021 |
| CO012 | Humans& Ai, Inc was filed on October 27, 2025 and remained active in March 2026 registry-mirror data. | 中 | SO022 |
| CO013 | Forbes reported on October 31, 2025 that Eric Zelikman was in talks to raise about $1 billion for a new frontier AI lab later known as humans&. | 中 | SO019 |
| CO014 | Eric Zelikman is CEO and co-founder of humans&. | 高 | SO001, SO004 |
| CO015 | Zelikman describes himself as an early xAI employee who contributed to Grok 2 pretraining data, Grok 3 reasoning reinforcement learning, and Grok 4 agent and tool reinforcement learning. | 中 | SO004 |
| CO016 | OpenReview lists Eric Zelikman as a Stanford PhD student with confirmed Stanford and x.ai email affiliations. | 中 | SO006 |
| CO017 | Zelikman says he wrote STaR while at Stanford, and retained research pages link him to both STaR and Quiet-STaR. | 中 | SO004, SO007, SO008 |
| CO018 | The publicly disclosed founding team consists of Eric Zelikman, Andi Peng, Georges Harik, Yuchen He, and Noah Goodman. | 高 | SO001, SO009, SO010 |
| CO019 | Andi Peng’s official bio ties her to Anthropic Claude behavioral and safety reinforcement learning plus post-training work. | 中 | SO001 |
| CO020 | Georges Harik’s official bio ties him to Google employee #7 status and early work on AdWords, AdSense, Gmail, Docs, and the Android acquisition. | 中 | SO001 |
| CO021 | Yuchen He’s official bio ties him to xAI and earlier OpenAI work on GPT-4x post-training, ChatGPT memory, and factuality. | 中 | SO001 |
| CO022 | Noah Goodman is presented as a co-founder, and retained sources tie him to Stanford academic work spanning natural and artificial intelligence. | 中 | SO001, SO021 |
| CO023 | TechCrunch reported that humans& had about 20 employees at its January 2026 launch. | 中 | SO009 |
| CO024 | Tracxn reported that humans& had 28 employees as of May 26, 2026. | 低 | SO024 |
| CO025 | Retained public sources do not disclose audited revenue, ARR, or named customer counts for humans&. | 中 | SO001, SO009, SO014 |
| CO026 | Retained public coverage describes the product thesis but does not document a launched product or public customer deployments at the January 2026 debut. | 中 | SO009, SO014, SO015 |
| CO027 | Tech Funding News reported that humans& planned to launch its first product early in 2026. | 中 | SO015 |
| CO028 | Multiple mainstream sources reported that humans& announced a $480 million seed financing in January 2026. | 高 | SO009, SO010, SO014, SO017 |
| CO029 | Multiple mainstream sources reported the round at a $4.48 billion valuation. | 高 | SO009, SO010, SO017 |
| CO030 | Reworked described the January 2026 financing as all-cash and unstructured. | 中 | SO014 |
| CO031 | Retained public investor lists consistently include SV Angel, Georges Harik, Nvidia, Jeff Bezos, GV, and Emerson Collective. | 高 | SO001, SO010, SO014, SO017 |
| CO032 | The company website also names Forerunner, S32, DCVC, Human Capital, Liquid 2, Felicis, and CRV among its investors. | 中 | SO001 |
| CO033 | The retained source set does not directly confirm Abstract Ventures as a January 2026 investor in humans&. | 中 | SO009, SO010, SO014, SO017 |
| CO034 | Accessible mainstream coverage usually states a $4.48 billion valuation without consistently clarifying whether the figure is pre-money or post-money. | 中 | SO009, SO010, SO014, SO017 |
| CO035 | Reworked argued that humans& reached unicorn status with no launched product and unusually opaque early proof points. | 中 | SO014 |
| CO036 | Reworked argued that enterprise incentives could still pull a human-centric AI product toward automation and headcount reduction. | 中 | SO014 |
| CO037 | Retained public sources do not disclose a formal board roster or clear governance-rights structure for humans&. | 低 | SO001, SO022 |
| CO038 | A California-registry mirror lists Eric Zelikman as CEO, CFO, and Secretary of Humans& Ai, Inc. | 中 | SO022 |
| CO039 | Company materials emphasize recruiting, indicating that capital deployment is still heavily oriented toward team buildout and product formation. | 中 | SO001, SO003 |
| CO040 | humans& should currently be treated as a well-capitalized but still stealth-adjacent research company whose funding visibility exceeds its product and customer visibility. | 中 | SO001, SO009, SO014, SO017 |
| CM001 | Human-in-the-loop AI integrates human judgment and feedback into model training, validation, and decision-making to improve accuracy and reliability in high-impact settings. | 中 | SM015, SM024 |
| CM002 | Recent HITL literature frames the category as a shift away from full autonomy and toward systems that enhance rather than replace human decision-making. | 中 | SM015 |
| CM003 | Some researchers argue many deployments described as human-in-the-loop are more accurately AI-in-the-loop because humans retain final decision authority. | 中 | SM016 |
| CM004 | The EU AI Act makes human oversight, logging, documentation, robustness, and data-quality controls explicit obligations for high-risk AI systems and GPAI models. | 中 | SM012 |
| CM005 | NIST, OECD, and EU policy sources all treat trustworthiness, evaluation, governance, and risk management as core deployment requirements for consequential AI use. | 高 | SM010, SM011, SM012, SM013 |
| CM006 | The narrowest directly observed public category adjacent to humans& is HITL AI, which spans software, hardware, and services for data labeling, model validation, human review, and governance. | 中 | SM024 |
| CM007 | A humans&-relevant market boundary excludes most device AI, semiconductors, generic hyperscaler infrastructure, and broad copilots unless spend is tied to reasoning workflows, evaluation, or human oversight. | 中 | SM001, SM022, SM024 |
| CM008 | Gartner forecasts worldwide AI spending at $2.596 trillion in 2026, but presents it as a broad market dominated by infrastructure and vendor-led capacity spending. | 中 | SM001 |
| CM009 | Gartner forecasts AI infrastructure spending of about $1.432 trillion in 2026, making infrastructure more than 45% of total AI spend. | 中 | SM001 |
| CM010 | Gartner forecasts AI model spending of about $32.6 billion in 2026. | 中 | SM001 |
| CM011 | Gartner forecasts AI platforms for data science and machine learning at about $29.9 billion in 2026. | 中 | SM001 |
| CM012 | Gartner forecasts AI application development platforms at about $8.4 billion in 2026. | 中 | SM001 |
| CM013 | Adding Gartner’s AI models, AI DS/ML platforms, and AI application development platforms yields an approximate $70.9 billion 2026 outer band for model-and-tooling budgets. | 中 | SM001 |
| CM014 | IDC says software and information services, banking, and retail alone account for about $89.6 billion of 2024 AI spending and nearly $222 billion by 2028. | 中 | SM003 |
| CM015 | IDC says generative AI already accounts for more than 19% of AI investment across those three leading industries. | 中 | SM003 |
| CM016 | Applying IDC’s stated 19% generative-AI share to its $89.6 billion three-industry total implies roughly $17.0 billion of 2024 spend in that slice. | 中 | SM003 |
| CM017 | MarketsandMarkets projects the AI agents market from $7.84 billion in 2025 to $52.62 billion by 2030 at a 46.3% CAGR. | 中 | SM022 |
| CM018 | The Business Research Company sizes the human-in-the-loop AI market at $5.4 billion in 2025, $6.73 billion in 2026, and $16.4 billion in 2030. | 中 | SM024 |
| CM019 | MarketsandMarkets’ public HITL market page provides qualitative segmentation and drivers but does not disclose a concrete public market value or CAGR, limiting its usefulness as a standalone sizing source. | 中 | SM023 |
| CM020 | These public estimates are not additive because they measure different layers of spend: infrastructure procurement, model-and-platform budgets, agent software, and human-oversight services. | 高 | SM001, SM003, SM022, SM024 |
| CM021 | McKinsey reports that 88% of surveyed organizations regularly use AI in at least one business function, but only about one-third have begun scaling AI across the enterprise. | 中 | SM008 |
| CM022 | McKinsey reports that 62% of respondents are at least experimenting with AI agents and 23% are scaling an agentic system somewhere in the enterprise. | 中 | SM008 |
| CM023 | Deloitte reports worker access to AI rose 50% in 2025 and that the number of companies with at least 40% of projects in production is set to double within six months. | 中 | SM006 |
| CM024 | Deloitte reports only 34% of organizations are deeply transforming the business with AI, while 37% still use it at a more surface level. | 中 | SM006 |
| CM025 | Bain finds only 7% of companies run fully autonomous agents in production today, while most still require human approval or guardrails. | 中 | SM020 |
| CM026 | The strongest near-term enterprise use cases for agentic or reasoning-heavy systems cluster in IT, knowledge management, customer support, supply chain, compliance, and R&D workflows. | 中 | SM006, SM008, SM022 |
| CM027 | MarketsandMarkets identifies BFSI as the largest 2025 end-user for AI agents and professional services as the fastest-growing end-user segment. | 中 | SM022 |
| CM028 | BCG finds that 62% of realized AI value comes from core business functions, with operations, sales and marketing, and R&D ahead of support functions. | 中 | SM018 |
| CM029 | PwC argues the highest-performing AI adopters win by pairing AI use with strategy, governance, data, workforce, and innovation foundations rather than proliferating disconnected pilots. | 中 | SM021 |
| CM030 | Microsoft’s 2025 diffusion data shows global generative-AI usage at about one in six people, but materially higher adoption in digitally prepared economies. | 中 | SM009 |
| CM031 | Deloitte predicts that 25% of enterprises already using generative AI will deploy AI agents in 2025, rising to 50% by 2027. | 中 | SM014 |
| CM032 | Governance and trust are first-order constraints because Deloitte sees low autonomous-agent governance maturity while NIST and EU frameworks are making oversight requirements more concrete. | 高 | SM006, SM010, SM012 |
| CM033 | Skills and workflow redesign remain gating factors because Deloitte describes the AI skills gap as the biggest integration barrier and notes most firms educate workers faster than they redesign roles. | 中 | SM006 |
| CM034 | Bain identifies data access and integration as the single biggest barrier to AI progress, cited by 41% of respondents ahead of compliance, budget, and skills gaps. | 中 | SM020 |
| CM035 | Bain reports that 44% of companies plan to fund the next AI wave from prior automation savings even though many earlier programs underdelivered their targets. | 中 | SM020 |
| CM036 | BCG argues that roughly 70% of AI implementation challenges come from people and process issues rather than algorithms. | 中 | SM018 |
| CM037 | RAND’s practitioner interviews identify wrong problem framing, insufficient data, weak infrastructure, and overconfidence in AI capabilities as leading failure modes for AI projects. | 中 | SM017 |
| CM038 | Deloitte’s enterprise scenarios warn that vendor lock-in, intellectual-property management, regulatory pressure, and trust erosion intensify as AI agents become embedded in workflows. | 中 | SM007 |
| CM039 | Deloitte predicts global data-center electricity use could roughly double to 1,065 TWh or 4% of total global energy consumption by 2030 as GenAI workloads scale. | 中 | SM014 |
| CM040 | The market layer most favorable to humans& is therefore the set of human-collaborative research, evaluation, orchestration, and oversight workflows where buyers accept ongoing human review instead of full autonomy. | 中 | SM015, SM016, SM020, SM021 |
| CM041 | Public sources do not cleanly isolate spending for reasoning-model evaluation, observability, and human-review tools from broader AI-agent, ML-platform, or HITL categories. | 中 | SM022, SM023, SM024 |
| CM042 | Public information does not disclose humans&’s initial customer segment, deployment model, or pricing basis well enough to isolate a responsible company-specific SAM or SOM. | 低 | |
| CM043 | No public contract-value or pricing benchmark cleanly matches a stealth human-collaborative AI research offering, so willingness-to-pay must be validated directly in diligence. | 低 | |
| CP001 | humans& publicly describes itself as a human-centric frontier AI lab that should act as connective tissue strengthening organizations and communities. | 中 | SP001 |
| CP002 | humans& says its agenda requires long-horizon and multi-agent reinforcement learning, memory, and user understanding. | 中 | SP001 |
| CP003 | humans& raised a $480 million seed round at a $4.48 billion valuation in January 2026. | 高 | SP002, SP003 |
| CP004 | humans& founders and early employees came from Anthropic, xAI, Google, OpenAI, Meta, Reflection, AI2, and Stanford-affiliated circles. | 高 | SP002, SP003 |
| CP005 | TechCrunch reported in late January 2026 that humans& still had no public product and no clear public description of exactly what it might launch. | 中 | SP003 |
| CP006 | humans& co-founders publicly framed the company as building a product and a model centered on communication and collaboration. | 中 | SP003 |
| CP007 | humans& hinted at both enterprise and consumer use cases, but its initial go-to-market focus was not publicly nailed down in the reviewed corpus. | 中 | SP003 |
| CP008 | Thinking Machines Lab describes itself as an AI research and product company focused on making AI more widely understood, customizable, and collaborative. | 中 | SP004 |
| CP009 | Thinking Machines Lab says it is building frontier-capability models with multimodality and infrastructure quality as top priorities. | 中 | SP004 |
| CP010 | Reuters reported that Thinking Machines raised about $2 billion at a $12 billion valuation in July 2025. | 中 | SP005 |
| CP011 | Reuters reported that nearly two-thirds of Thinking Machines Lab’s launch team came from OpenAI. | 中 | SP005 |
| CP012 | Reuters reported that Thinking Machines later explored a new round around a $50 billion valuation and had already launched a first product named Tinker. | 中 | SP006 |
| CP013 | SSI says it is a straight-shot lab with one goal and one product: safe superintelligence. | 中 | SP007 |
| CP014 | SSI says its business model is designed to keep safety, security, and progress insulated from short-term commercial pressures. | 中 | SP007 |
| CP015 | Reuters reported that SSI discussed a funding round at at least a $20 billion valuation after previously raising $1 billion at a $5 billion valuation. | 中 | SP008 |
| CP016 | Reuters reported that SSI was recently valued at $32 billion and attracted Alphabet and Nvidia as investors. | 中 | SP009 |
| CP017 | Reuters reported that Google agreed to supply SSI with significant TPU capacity, showing that compute access is part of the frontier-lab competitive stack. | 中 | SP009 |
| CP018 | Anthropic publishes public pricing, model system cards, and transparency materials, signaling a visible enterprise and compliance surface. | 高 | SP010, SP011, SP012 |
| CP019 | OpenAI sells ChatGPT Enterprise and maintains public eval learning materials and API guidance. | 高 | SP013, SP014, SP015 |
| CP020 | OpenAI’s legacy Evals platform is scheduled to become read-only on October 31, 2026 and shut down on November 30, 2026. | 中 | SP015 |
| CP021 | OpenAI exposes public API pricing, unlike the stealth frontier labs in this peer set. | 中 | SP016 |
| CP022 | xAI’s docs expose public model selection guidance and optional search-enabled tooling around Grok. | 中 | SP017, SP018 |
| CP023 | Scale Evaluation sells detailed model analysis, custom evaluation sets, expert human raters, and monitoring for frontier model developers. | 高 | SP019, SP022 |
| CP024 | Scale says its Data Engine powers frontier AI through RLHF, evaluation, safety, and alignment services. | 高 | SP020, SP021 |
| CP025 | Scale created SEAL to build evaluation and red-teaming products, underscoring buyer demand for external safety infrastructure. | 中 | SP022 |
| CP026 | Scale’s June 2025 Meta transaction valued the company at over $29 billion and moved founder Alexandr Wang to Meta while Scale remained independent. | 中 | SP023 |
| CP027 | Labelbox offers frontier model evaluation, human-preference arenas, RLHF workflows, and side-by-side model comparison tooling. | 高 | SP025, SP026, SP027, SP028 |
| CP028 | Labelbox exposes a free self-serve entry point while steering larger teams toward services and enterprise packages. | 中 | SP024 |
| CP029 | Braintrust markets observability and evaluation in one platform and publishes a free starter tier plus a $249 per month pro plan. | 高 | SP029, SP030 |
| CP030 | Arize Phoenix markets itself as an open-source platform for agent development and evaluation while Arize also lists paid AX tiers. | 高 | SP031, SP032 |
| CP031 | Humanloop previously sold a free-plus-enterprise eval platform and then announced that it was joining Anthropic and sunsetting that platform. | 高 | SP033, SP034 |
| CP032 | The neo-lab set around humans&, Thinking Machines Lab, and SSI is competing for the same capital and talent pool before it is competing on broadly deployed enterprise tooling. | 中 | SP002, SP005, SP008, SP009 |
| CP033 | humans&’s most explicit public wedge is collaborative and social-intelligence reasoning rather than single-user assistance or generic benchmark leadership. | 中 | SP001, SP003 |
| CP034 | humans& differentiation remains unproven because the reviewed public corpus still lacks a launched product, customers, deployments, or a price surface. | 中 | SP003 |
| CP035 | Thinking Machines Lab overlaps with humans& on human-AI collaboration rhetoric but is farther along in public product articulation because it has already described principles and launched Tinker. | 中 | SP004, SP006 |
| CP036 | SSI competes more for researchers, investors, and compute than for current workflow budget because its public posture remains singularly research-first. | 中 | SP007, SP008, SP009 |
| CP037 | Anthropic, OpenAI, and xAI are more enterprise-ready substitutes today because they already expose public model, tooling, and documentation surfaces. | 中 | SP010, SP012, SP013, SP015, SP016, SP017, SP018 |
| CP038 | Scale, Labelbox, Braintrust, Arize, and formerly Humanloop compete on the enabling layer around human feedback, evals, observability, and monitoring rather than on base-model research itself. | 中 | SP019, SP025, SP029, SP032, SP033 |
| CP039 | The eval and human-data layer is crowded and already consolidating, as Humanloop joined Anthropic while Scale deepened its strategic relationship with Meta. | 中 | SP023, SP034 |
| CP040 | Compute access is a structural barrier in this market because strategic chip relationships themselves are being used as competitive leverage. | 中 | SP009 |
| CP041 | Reuters characterized the frontier-lab hiring market as an escalating talent war when describing Thinking Machines Lab’s financing. | 中 | SP005 |
| CP042 | Internal build remains a credible status-quo substitute because buyers can compose public frontier-model APIs with independent evaluation and observability tools. | 中 | SP013, SP015, SP017, SP029, SP032, SP028 |
| CP043 | Public pricing is concentrated in tooling and API vendors, whereas humans&, Thinking Machines Lab, and SSI expose no reviewed public commercial packaging. | 中 | SP003, SP006, SP007, SP010, SP016, SP024, SP030, SP031, SP033 |
| CP044 | Because incumbent APIs and eval vendors are modular and separately purchasable, many buyers can multi-home rather than accept an end-to-end lock-in bet. | 中 | SP015, SP024, SP029, SP031, SP032 |
| CP045 | If humans& wins accounts, its strongest potential switching barrier would be workflow-specific collaboration memory and decision context rather than base-model scarcity. | 中 | SP001, SP003, SP015, SP032 |
| CP046 | Without public customers or deployment evidence, any claim that humans& already has durable switching costs should be treated as a hypothesis, not an established fact. | 中 | SP003 |
| CI001 | humans& publicly describes itself as a human-centric frontier AI lab. | 中 | SI001 |
| CI002 | The official humans& homepage says the company wants AI to serve as a deeper connective tissue that strengthens organizations and communities. | 中 | SI001 |
| CI003 | The official homepage says the technical agenda includes long-horizon and multi-agent reinforcement learning, memory, and user understanding. | 中 | SI001 |
| CI004 | TechCrunch reported on January 20, 2026 that humans& raised $480 million in seed funding at a $4.48 billion valuation. | 中 | SI003 |
| CI005 | Reuters likewise reported on January 20, 2026 that humans& raised $480 million in seed financing at a $4.48 billion valuation. | 中 | SI004 |
| CI006 | Crunchbase News reported that the round was raised all cash and unstructured. | 中 | SI005 |
| CI007 | Crunchbase News reported that co-founder Georges Harik and SV Angel led the round, with Nvidia, Jeff Bezos, GV, Emerson Collective, Forerunner, S32, DCVC, Human Capital, Felicis, and CRV also participating. | 中 | SI005 |
| CI008 | Reuters reported that humans& expected to launch a product early in 2026. | 中 | SI004 |
| CI009 | TechCrunch described the intended product as software that helps people collaborate with each other, likening it to an AI version of an instant messaging app. | 中 | SI003 |
| CI010 | No reviewed humans& official source disclosed public pricing, subscription tiers, API rates, customer counts, or revenue metrics. | 高 | SI001, SI002, SI027 |
| CI011 | TechCrunch said the company had about 20 employees at launch. | 中 | SI003 |
| CI012 | Tracxn listed humans& as a seed-stage company with 28 employees as of May 26, 2026. | 中 | SI011 |
| CI013 | The difference between the roughly 20 employees reported at launch and the 28 employees later listed by Tracxn is directionally consistent with an early hiring ramp after the seed round, but the exact current headcount is not independently verified. | 中 | SI003, SI011 |
| CI014 | BizProfile says Humans& Ai, Inc. was officially filed in California on October 27, 2025 under document number B20250359156 and is formed in Delaware. | 中 | SI021 |
| CI015 | The official humans& terms identify the website operator as humans& ai, inc. | 中 | SI001 |
| CI016 | The SEC Form D datasets page says notices of exempt offerings filed with the Commission are updated quarterly through March 2026. | 中 | SI007 |
| CI017 | During this run, the reviewed SEC EDGAR search surface exposed company-name and filing-type search capability but no issuer-level humans& filing URL was located from the public materials gathered here. | 中 | SI007, SI008 |
| CI018 | Crunchbase News reported that humans& plans to spend the majority of the capital on compute for training models. | 中 | SI005 |
| CI019 | Reuters reported Nvidia participated in the seed round, reinforcing the view that compute access is strategically important to the company. | 中 | SI004 |
| CI020 | TechStartups reported humans& would work closely with Nvidia on both hardware and software, although this partnership detail was not corroborated in the official homepage reviewed for this chapter. | 低 | SI009 |
| CI021 | OpenAI API pricing shows frontier-model outputs can be monetized through usage-based token pricing, with GPT-5.5 listed at $30 per 1M output tokens and GPT-5.4 at $15 per 1M output tokens. | 中 | SI014 |
| CI022 | OpenAI, Anthropic, Google Workspace, Slack, and Microsoft all publish per-user or plan-based pricing for AI-assisted collaboration or productivity products. | 高 | SI015, SI016, SI017, SI019, SI024 |
| CI023 | Slack publicly lists paid collaboration plans at $7.25 per user per month annually for Pro and $15 per user per month annually for Business+, illustrating a plausible seat-pricing reference range for collaboration software. | 中 | SI019 |
| CI024 | Google Workspace publicly lists Business Standard at $14 per user per month and Business Plus at $22 per user per month, showing that mainstream collaboration suites can sustain double-digit monthly seat pricing. | 中 | SI017 |
| CI025 | Anthropic publicly lists Claude Pro at $20 monthly, another benchmark that AI productivity tools can monetize through seat subscriptions even before fully custom enterprise contracts are considered. | 中 | SI016 |
| CI026 | Notion markets AI agents for teamwork and says usage-based AI agents cost $10 per 1,000 monthly Notion credits, illustrating that agent-oriented collaboration tools can also mix seat and consumption pricing. | 高 | SI020, SI025 |
| CI027 | Because humans& has not published a pricing page or product documentation, any current revenue model beyond a future collaboration product remains unverified. | 高 | SI001, SI002, SI027 |
| CI028 | The company is best understood as pre-launch and likely pre-revenue from the public record reviewed here, because Reuters said launch was still ahead and no public revenue disclosure was found. | 中 | SI004, SI001, SI002, SI027 |
| CI029 | The public record reviewed here does not disclose gross margin, CAC, payback, NRR, or sales-cycle metrics. | 高 | SI001, SI002, SI003, SI004, SI027 |
| CI030 | The public record reviewed here does not disclose cash on hand after close, monthly burn, or runway months. | 高 | SI003, SI004, SI005, SI001 |
| CI031 | A $480 million seed round provides a very large starting balance relative to a sub-30-person team, but capital adequacy still depends on how quickly humans& scales compute commitments and research hiring. | 中 | SI003, SI005, SI011 |
| CI032 | Reuters and Crunchbase together support the view that investors funded humans& primarily as a frontier-AI research buildout rather than against demonstrated commercial traction. | 高 | SI004, SI005 |
| CI033 | OpenAI’s January 2026 advertising announcement shows that even scaled frontier labs can keep broad access affordable by layering new monetization on top of subscriptions and API sales, highlighting one possible but unproven future path for humans&. | 中 | SI026 |
| CI034 | AI Funding Tracker argues that enterprise incentives usually favor cost reduction and headcount elimination, which is adverse to humans&’s augmentation-first positioning. | 中 | SI013 |
| CI035 | AI Funding Tracker also frames the company as competing in a crowded field against incumbent productivity suites and major AI labs, which raises go-to-market and differentiation risk. | 中 | SI013 |
| CI036 | TechCrunch noted that the largest seed round on record belongs to Thinking Machines Lab and that pedigree plus capital do not guarantee immediate success, which is relevant downside context for humans&. | 中 | SI003 |
| CI037 | The public surface for humans& appears intentionally thin: a manifesto-style homepage, an X account, and a jobs page exist, but detailed product, customer, and pricing materials do not. | 高 | SI001, SI002, SI027 |
| CI038 | This chapter cannot underwrite revenue quality or runway from public evidence; the strongest public conclusion is that humans& is pre-commercial with substantial capital but unusually large disclosure gaps for a company priced at a multibillion-dollar valuation. | 高 | SI003, SI004, SI005, SI010, SI013 |
| CE001 | humans& publicly describes itself as a human-centric frontier AI lab. | 中 | SE001 |
| CE002 | The official mission centers on helping people understand one another, build trust, make connections, and work together. | 中 | SE001 |
| CE003 | humans& says the next paradigm requires innovations in long-horizon reinforcement learning, multi-agent reinforcement learning, memory, and user understanding. | 中 | SE001, SE006 |
| CE004 | humans& says it will tightly integrate science and product development. | 中 | SE001, SE006 |
| CE005 | The official site frames the team as having built seminal work in reasoning, behavioral training, agents, and alignment across major labs. | 中 | SE001 |
| CE006 | The recruitment copy promises contributions back to open source and academic research. | 中 | SE001 |
| CE007 | TechCrunch reported that humans& is building both a product and a model centered on communication and collaboration. | 中 | SE005 |
| CE008 | The clearest public analogies put the product in shared communication or collaboration contexts such as messaging, docs, or Notion-like teamwork rather than a generic single-user assistant. | 中 | SE005, SE006 |
| CE009 | humans& told TechCrunch it wants to own the collaboration layer and is co-evolving product interface and model together rather than merely plugging a model into existing collaboration tools. | 中 | SE005 |
| CE010 | Co-founders described a training direction with more humans and AIs interacting and collaborating together. | 中 | SE005 |
| CE011 | Yuchen He said the model will be trained using long-horizon and multi-agent reinforcement learning. | 中 | SE005 |
| CE012 | Public materials make memory and user understanding central to the product idea. | 中 | SE001, SE005 |
| CE013 | As of the January 25 TechCrunch profile, humans& still did not have a publicly launched product and was still shaping what it would be. | 中 | SE005 |
| CE014 | The official website includes a cultural-dynamics simulation about interacting agents, reinforcing that the public narrative emphasizes social interaction and coordination rather than generic chatbot UX. | 低 | SE001 |
| CE015 | Official team bios show expertise across reasoning RL, inference, privacy, backend systems, GPU kernels, and data-center engineering. | 中 | SE001 |
| CE016 | Eric Zelikman's public bio ties him to Grok 2 pretraining data, Grok 3 reasoning RL, and Grok 4 agent/tool RL before humans&. | 中 | SE003 |
| CE017 | Eric Zelikman's public materials connect him to both STaR and Quiet-STaR. | 中 | SE003, SE001 |
| CE018 | STaR introduced a loop where a model improves reasoning by generating rationales, retrying with the correct answer, and fine-tuning on rationales that lead to correct answers. | 中 | SE010, SE011 |
| CE019 | Quiet-STaR generalizes STaR by teaching language models to generate token-level internal rationales that improve predictions and downstream question answering. | 中 | SE009 |
| CE020 | Quiet-STaR's author list links current humans& personnel including Eric Zelikman, Georges Harik, and Varuna Jayasiri to recent reasoning research. | 中 | SE009, SE001 |
| CE021 | Zelikman's Quiet-STaR GitHub repository was publicly visible with 739 stars and 92 forks at fetch time, showing real external developer engagement. | 中 | SE014 |
| CE022 | Zelikman's GitHub profile showed 28 public repositories and 211 followers at fetch time, indicating a meaningful pre-humans& developer footprint. | 中 | SE013, SE024 |
| CE023 | Taylor Sorensen's public materials identify him as a humans& researcher and describe the company as a small, high-agency place that combines fundamental research with product design. | 中 | SE015, SE016, SE017 |
| CE024 | Taylor Sorensen's essay frames the mission around plural human perspectives, collective decision-making, and preserving human agency rather than optimizing for one-shot preferences. | 中 | SE015 |
| CE025 | Saurabh Shah's site says he works at humans& training AI systems to work with people, not replace them. | 中 | SE018 |
| CE026 | Alexis Ross's public site describes her as a founding researcher at humans& working on human-AI collaboration. | 中 | SE019, SE020 |
| CE027 | Niloofar Mireshghallah's public site ties humans& to explicit privacy, contextual-integrity, memory, and long-horizon interaction expertise. | 中 | SE021 |
| CE028 | The humans& privacy notice says the public site is primarily informational and does not use analytics services, advertising pixels, or tracking cookies for profiling. | 中 | SE002 |
| CE029 | The privacy notice says the site may process only basic server logs such as IP address, user-agent, timestamps, page requests, and error/performance data for reliability and security. | 中 | SE002 |
| CE030 | The privacy notice discloses that job applications are handled through Ashby and that Google Fonts can receive standard request data. | 中 | SE002, SE025 |
| CE031 | humans& says it uses reasonable administrative, technical, and organizational measures to protect information processed through the site, but gives no detailed product-security architecture. | 中 | SE002 |
| CE032 | No reviewed public source disclosed SOC 2, ISO 27001, a public DPA, a product security whitepaper, or an uptime/status surface for the collaboration product. | 低 | SE001, SE002, SE005, SE025 |
| CE033 | No reviewed public source disclosed named customers, production deployments, or quantified product performance benchmarks. | 低 | SE001, SE005, SE008 |
| CE034 | Reworked argues that even a genuinely collaborative AI product could drift toward automation because enterprise incentives usually reward efficiency extraction over augmentation. | 中 | SE008 |
| CE035 | Crunchbase News reported that most of the capital would go to compute, implying that humans& is financing a heavy training effort ahead of public deployment proof. | 中 | SE007 |
| CE036 | A third-party GitHub repo built at a Humans& Product Hackathon framed AI agents as collaborators rather than assistants, a weak but real practitioner proxy for how the brand is being interpreted. | 低 | SE023 |
| CE037 | The public recruiting and homepage surfaces emphasize world-class, cross-disciplinary builders and researchers, but the reviewed jobs surface does not expose a detailed role matrix or technical stack taxonomy. | 低 | SE001, SE025 |
| CE038 | The most supportable product definition today is a stealth research-and-product system for communication and collaboration workflows, not a launched general-purpose API or commodity chatbot. | 中 | SE001, SE005, SE006 |
| CE039 | Across official and third-party materials, the moat ambition is framed around social intelligence, memory, and coordination rather than raw question-answering alone. | 中 | SE001, SE005 |
| CE040 | Despite the open-source intent and founder repo history, the reviewed public corpus does not show an official humans& GitHub organization, SDK, or docs center. | 低 | SE001, SE013, SE024 |
| CU001 | The official humans& site is presented as an announcement surface rather than a commercial product surface. | 中 | SU001 |
| CU002 | The reviewed official surface does not expose product pricing, customer stories, case studies, or deployment documentation. | 中 | SU001 |
| CU003 | Public positioning points first toward collaboration and communication workflows rather than a generic single-user chatbot. | 高 | SU001, SU003, SU004, SU005 |
| CU004 | TechCrunch compared the concept to AI versions of instant messaging and multiplayer collaboration contexts such as Slack, Google Docs, and Notion. | 高 | SU003, SU004 |
| CU005 | Crunchbase said not much was publicly known about humans& beyond the team, funding, and mission. | 中 | SU005 |
| CU006 | TechCrunch reported that humans& still did not have a public product as of late January 2026. | 中 | SU004 |
| CU007 | No named customer reference was found in the reviewed official, launch, or funding sources retained for this chapter. | 高 | SU001, SU003, SU004, SU005 |
| CU008 | No public customer outcome, deployment metric, or reference quote was found in the retained customer-proof corpus for humans& itself. | 中 | SU001, SU003, SU004, SU005, SU019, SU020, SU021, SU022 |
| CU009 | The best-supported adoption statement today is absence of public disclosure, not evidence of scaled traction. | 中 | SU001, SU004, SU005 |
| CU010 | Launch and funding coverage identify founders, investors, and mission but do not identify named customers or production deployments. | 高 | SU003, SU004, SU005 |
| CU011 | TechCrunch said the founders hinted at both enterprise and consumer applications, which broadens long-run opportunity but weakens near-term ICP clarity. | 中 | SU004 |
| CU012 | The official site frames the company around strengthening organizations and communities, implying multi-person coordination as the core job to be done. | 中 | SU001 |
| CU013 | The public jobs board exposed research, product, and finance openings rather than a broad public sales or customer-success surface. | 中 | SU002 |
| CU014 | The visible public jobs board titles included Finance Generalist, Member of Technical Staff, and Product Engineer (Member of Technical Staff). | 中 | SU002 |
| CU015 | Taken together, public positioning and hiring support a view that humans& is still earlier in product and research buildout than in scaled customer operations. | 中 | SU002, SU003, SU004 |
| CU016 | No public retention, contract-length, or renewal metrics were identified in the reviewed corpus. | 中 | SU001, SU003, SU004, SU005, SU019, SU020, SU021, SU022 |
| CU017 | No public customer concentration, top-account share, or channel-dependence metric was identified in the reviewed corpus. | 中 | SU001, SU003, SU004, SU005 |
| CU018 | Druid’s production telemetry shows that real AI usage often concentrates in narrow, front-door workflows rather than broad transformation stories. | 中 | SU018 |
| CU019 | Deloitte said the number of companies with at least 40% of projects in production is set to double in six months, raising the proof bar buyers will expect. | 高 | SU008, SU023 |
| CU020 | G2 and Capterra show that buyers in adjacent collaboration and AI categories evaluate workflow coverage, collaboration features, and user satisfaction rather than pure narrative differentiation. | 高 | SU019, SU020, SU021, SU022 |
| CU021 | G2’s collaboration category requires chat, document sharing, calendars, and task collaboration features, which makes practical workflow depth central to category fit. | 中 | SU020 |
| CU022 | Capterra’s collaboration guidance emphasizes work style, deployment model, and trial evaluation, reinforcing that buyers expect proof of fit before expansion. | 中 | SU022 |
| CU023 | Microsoft’s 2026 Work Trend Index found that 49% of AI-supported conversations are cognitive work and 19% involve working with people, which is directionally consistent with humans& collaboration thesis. | 中 | SU013 |
| CU024 | Deloitte’s human-AI interaction research says organizations are twice as likely to beat AI ROI expectations when they intentionally redesign human-machine work interactions. | 中 | SU010 |
| CU025 | PwC found that 20% of surveyed companies capture 74% of AI-driven returns, showing that value creation is concentrated rather than automatic. | 中 | SU011 |
| CU026 | BCG reported that 74% of companies had yet to show tangible value from AI, which supports skepticism toward pre-proof commercialization claims. | 中 | SU016 |
| CU027 | Bain reported that only 7% of companies run fully autonomous agents in production today, with human approval and guardrails far more common. | 中 | SU024 |
| CU028 | RAND said more than 80% of AI projects fail by some estimates, with common causes including wrong problem framing, weak data, and poor workflow fit. | 中 | SU025 |
| CU029 | WEF argued that effective AI work design separates repeatable tasks for machines from judgment, relationships, and trade-offs that stay with humans. | 中 | SU014 |
| CU030 | CNBC reported that nearly one in five consumers who used AI for customer service saw no benefit and described chatbot loops as frustrating. | 中 | SU006 |
| CU031 | AnswerConnect said 85% of surveyed consumers would rather speak to a real person than AI for customer service, and 57% would trust a business less if it mostly used AI support. | 中 | SU007 |
| CU032 | The public evidence reviewed here supports no named production deployment for humans& as of 2026-06-19. | 中 | SU001, SU003, SU004, SU005, SU019, SU020, SU021, SU022 |
| CU033 | The public evidence reviewed here supports no quantified customer outcome or adoption metric for humans& as of 2026-06-19. | 中 | SU001, SU003, SU004, SU005, SU019, SU020, SU021, SU022 |
| CU034 | The likeliest initial paying-buyer hypothesis is an enterprise team with costly coordination problems, while consumer usage remains a plausible but unproven second path. | 中 | SU001, SU003, SU004, SU013 |
| CU035 | Category-level review platforms and benchmark sources provide procurement proxies, but they do not substitute for a humans& company-specific review footprint or reference customer. | 中 | SU018, SU019, SU020, SU021, SU022 |
| CU036 | Public sources do not disclose actual contract length, renewal timing, or seat-expansion behavior for any humans& account. | 中 | SU001, SU003, SU004, SU005 |
| CU037 | Public sources do not disclose any NRR, GRR, churn, or cohort curve for humans&. | 中 | SU001, SU003, SU004, SU005 |
| CU038 | If early revenue exists, concentration risk could be high because no public customer mix or top-account data is disclosed. | 中 | SU003, SU004, SU005, SU017 |
| CU039 | Trying to address both enterprise and consumer contexts before public product-market proof could slow ICP clarity and sales repeatability. | 中 | SU004, SU011, SU025 |
| CU040 | Absence of public customer proof should be interpreted as a diligence limit, not as proof that humans& has zero private pilots or design partners. | 中 | SU001, SU004, SU019, SU021 |
| CU041 | The highest-priority diligence asks are a named customer list, pilot-to-production funnel, outcome evidence, retention metrics, concentration data, and live reference calls. | 中 | SU010, SU018, SU024, SU025 |
| CR001 | TechCrunch reported on 2026-01-20 that humans& raised $480 million in seed funding at a $4.48 billion valuation. | 中 | SR004 |
| CR002 | Reuters likewise reported on 2026-01-20 that humans& raised $480 million in seed financing at a $4.48 billion valuation. | 中 | SR027 |
| CR003 | Crunchbase News reported that the seed round was raised all cash and unstructured. | 中 | SR008 |
| CR004 | Reworked said humans& debuted at unicorn status with no product. | 中 | SR005 |
| CR005 | Reworked quoted startup attorney Lindsey Mignano saying humans&'s funding and valuation were highly unusual even by the standards of the AI boom because of their scale and opacity. | 中 | SR005 |
| CR006 | Reuters said humans& expected to launch a product early in 2026. | 中 | SR027 |
| CR007 | The reviewed official humans& surfaces exposed legal, privacy, and recruiting information but no public pricing page, customer list, product manual, or usage dashboard. | 高 | SR001, SR002, SR003, SR030 |
| CR008 | The terms pages say humans& may change, suspend, or discontinue the website or its content without prior notice or liability. | 高 | SR002, SR003 |
| CR009 | The privacy notice says information may be transferred if humans& enters a merger, acquisition, financing, or sale of assets. | 高 | SR001, SR003 |
| CR010 | Built In said the lofty cost of using existing AI techniques to train automated models in new ways was a driving factor behind the company’s large raise. | 中 | SR009 |
| CR011 | SiliconANGLE described humans& as newly launched when it announced the $480 million round. | 中 | SR010 |
| CR012 | As of 2026-06-19, the public commercialization proof is materially thinner than the public financing proof. | 高 | SR001, SR002, SR003, SR004, SR005, SR027 |
| CR013 | TechCrunch listed the disclosed founding team as Andi Peng, Georges Harik, Eric Zelikman, Yuchen He, and Noah Goodman. | 中 | SR004 |
| CR015 | Tracxn listed humans& at 28 employees as of 2026-05-26. | 中 | SR029 |
| CR016 | Bizprofile’s California-registry mirror listed Eric Zelikman as CEO, CFO, and Secretary of Humans& Ai, Inc. | 中 | SR028 |
| CR017 | OpenReview identified Eric Zelikman as a PhD student at Stanford University. | 中 | SR026 |
| CR018 | Reuters said Zelikman previously worked at xAI and had a research background in reasoning-focused reinforcement-learning methods. | 中 | SR027 |
| CR019 | The combination of officer-title concentration and a team that still appears subscale relative to the ambition creates meaningful key-person and bench-depth risk. | 高 | SR004, SR026, SR028, SR029 |
| CR020 | Crunchbase News reported that humans& planned to spend the majority of its capital on compute for training models. | 中 | SR008 |
| CR021 | Reuters highlighted Nvidia as a participant in the seed round and said it was taking stakes in companies that rely heavily on its computing hardware. | 中 | SR027 |
| CR022 | TechStartups reported that humans& would work closely with Nvidia on both hardware and software, although that detail was not corroborated in the official company pages reviewed here. | 低 | SR011 |
| CR023 | The SaaS News said humans& would use the funding to advance multi-agent reinforcement learning, memory, long-horizon planning, and user-centered product development. | 中 | SR013 |
| CR024 | CNBC reported that the AI talent war is dangling millions of dollars in front of a small talent pool of specialists. | 中 | SR015 |
| CR025 | CNBC said only a handful of companies can afford to build frontier models because doing so is highly capital-intensive and requires billions of dollars. | 中 | SR015 |
| CR026 | CNBC warned that startups can get left behind while tech giants use outsized compensation to hoard AI talent. | 中 | SR015 |
| CR027 | Forbes reported a shortage of talent that understands the pipelines connecting frontier labs to human-data suppliers. | 中 | SR016 |
| CR028 | humans& appears exposed to simultaneous compute scarcity and specialist labor scarcity. | 中 | SR008, SR015, SR016, SR027 |
| CR029 | The reviewed legal pages impose export-control compliance obligations and broad IP and content-use restrictions. | 高 | SR002, SR003 |
| CR030 | The European Commission says the AI Act entered into force on 2024-08-01, that prohibited-practice rules started in February 2025, and that governance rules plus GPAI obligations became applicable on 2025-08-02 ahead of broader 2026 applicability. | 高 | SR022, SR023 |
| CR031 | The Commission’s GPAI guidelines say enforcement powers enter into application on 2026-08-02 and that systemic-risk model providers must notify the AI Office and report serious incidents. | 高 | SR023, SR022 |
| CR032 | The U.S. Copyright Office says it released a pre-publication version of Part 3 on Generative AI Training on 2025-05-09 and has not yet issued the final version. | 高 | SR020, SR021 |
| CR033 | The Copyright Office’s Part 3 report says generative AI training raises questions about copying, fair use, and licensing arrangements, and that the analysis is constrained by rapidly evolving technology and markets. | 高 | SR021, SR020 |
| CR034 | NIST says the AI RMF is intended to improve trustworthiness across the design, development, use, and evaluation of AI products, services, and systems. | 高 | SR017, SR018 |
| CR035 | NIST’s generative-AI profile says organizations should govern, map, measure, and manage generative-AI risks across the lifecycle, including issues involving cloud-based services or acquisition, with particular attention to governance, content provenance, pre-deployment testing, and incident disclosure. | 高 | SR018, SR019 |
| CR036 | The 2026 International AI Safety Report says frontier general-purpose AI risks include misuse, issues of control, cybersecurity risks, malfunctions, and systemic disruption. | 中 | SR024 |
| CR037 | The 2026 International AI Safety Report says reliable pre-deployment safety testing has become harder because advanced models can distinguish test settings from real deployment and exploit evaluation loopholes. | 中 | SR024 |
| CR038 | OpenAI’s safety page shows a frontier lab treating safety as ongoing work documented through recurring safety practices and multiple system cards, not as a one-time disclosure. | 中 | SR025 |
| CR039 | If humans& is training or materially modifying frontier models, the compliance expectations around transparency, incident reporting, copyright, and safety testing are rising faster than its public documentation suggests. | 高 | SR019, SR021, SR022, SR023, SR024, SR025 |
| CR040 | The privacy notice says vendors who help host, secure, and operate the site may process basic logs and technical data on humans&'s behalf. | 高 | SR001, SR003 |
| CR041 | The official legal pages say external links include a jobs page hosted by Ashby and social-media properties, confirming that some recruiting and engagement surfaces sit outside the core domain. | 高 | SR001, SR002, SR003 |
| CR042 | Reuters said humans& was building human-centric AI tools for communication and collaboration and quoted Zelikman saying the model would coordinate with people and other AIs where appropriate. | 中 | SR027 |
| CR043 | Reworked argued that the human-centric thesis only holds if deployment incentives align toward augmentation rather than efficiency extraction. | 中 | SR005 |
| CR044 | Reworked said the steeper uphill battle is whether enterprise conditions will preserve collaboration rather than push the product toward automation and headcount reduction. | 中 | SR005 |
| CR045 | Forbes reported before launch that Zelikman was in talks to raise about $1 billion at a $5 billion valuation for a new frontier lab. | 中 | SR007 |
| CR046 | The Economic Times described Nvidia as a key backer of AI startups that rely heavily on its computing hardware. | 中 | SR012 |
| CR047 | The financing story already embeds very high expectations for launch quality, commercialization proof, and follow-on execution. | 中 | SR005, SR007, SR027 |
| CR048 | No reviewed public source disclosed named customers, production deployments, revenue, retention, current safety-evaluation results, or a public incident history for humans&. | 高 | SR001, SR002, SR003, SR004, SR027 |
| CR049 | SiliconANGLE said the first product was planned for early 2026 and noted that reinforcement-learning-based reasoning development can still be capital-intensive because it often needs large numbers of graphics cards. | 中 | SR010 |
| CR050 | Tech Funding News said humans& planned to launch its first product early in 2026 while inviting comparisons with Thinking Machines Lab. | 中 | SR006 |
| CR051 | Reuters and Reuters-linked hiring coverage show that next-generation AI growth is increasingly tied to specialized skills rather than broad workforce expansion. | 中 | SR014, SR027 |
| CR052 | A Wayback-captured Ashby page confirms humans& maintained a separate jobs portal by May 2026, but the archived public text revealed only the brand name rather than useful public detail about openings or org depth. | 低 | SR030 |
| CR053 | Public mitigants include unusual seed capital, elite founders, basic legal/privacy controls, and evidence of active recruiting, but none of those by themselves proves product-market fit or safety readiness. | 中 | SR002, SR003, SR004, SR027, SR030 |
| CR054 | If management still cannot show live product metrics, named deployments, or customer proof despite the January 2026 launch expectation, commercialization risk should be treated as thesis-breaking at the current valuation narrative. | 中 | SR005, SR006, SR027 |
| CR055 | If compute access or critical frontier hires tighten, humans&'s timelines likely slip because public sources already point to compute and specialist labor as core bottlenecks. | 中 | SR008, SR015, SR016 |
| CR056 | If management cannot provide a current model-governance and safety pack covering testing, incident response, and compliance ownership, frontier-AI regulatory risk remains a blocking diligence item. | 中 | SR019, SR023, SR024, SR025 |
| CR057 | If management cannot reconcile current valuation expectations with a concrete roadmap to launch, customer proof, and commercialization, the main downside becomes multiple compression rather than a modest schedule delay. | 中 | SR005, SR007, SR027 |
| CV001 | humans& publicly describes itself as a human-centric frontier AI lab focused on collaboration rather than replacement. | 中 | SV001 |
| CV002 | Multiple independent January 2026 reports agree that humans& raised a $480 million seed round at a quoted $4.48 billion valuation. | 中 | SV002, SV003, SV004 |
| CV003 | Reuters said the round was led by SV Angel and co-founder Georges Harik, with Nvidia, Jeff Bezos, and GV among participants. | 中 | SV003, SV004 |
| CV004 | Crunchbase News reported that humans& described the seed round as all-cash and said most of the capital would be spent on compute for training models. | 中 | SV004 |
| CV005 | TechCrunch described humans& as having roughly 20 employees at launch, while Tracxn later showed a 28-employee estimate dated May 26, 2026. | 中 | SV002, SV006 |
| CV006 | The public materials retained for this chapter do not disclose revenue, customer names, pricing, or a board roster for humans&. | 中 | SV001, SV005 |
| CV007 | Reworked argued that humans& reached unicorn status before shipping a product and that the human-centric pitch may still be pulled toward automation economics in enterprise settings. | 中 | SV005 |
| CV008 | Forbes reported in October 2025 that the founders were discussing a round as large as $1 billion at a $5 billion valuation before the company launched publicly. | 中 | SV007 |
| CV009 | Thinking Machines Lab describes itself as an AI research and product company aimed at making AI more understandable, customizable, and collaborative. | 中 | SV010 |
| CV010 | TechCrunch reported in June 2025 that Thinking Machines Lab closed a $2 billion seed round at a $10 billion valuation roughly six months after founding. | 中 | SV011 |
| CV011 | TechCrunch later reported that Thinking Machines Lab officially closed the same seed round at a $12 billion valuation. | 中 | SV012 |
| CV012 | As of May 2026, TechCrunch still described Thinking Machines' flagship interaction model as a research preview rather than a public product. | 中 | SV013 |
| CV013 | SSI's official site says the company has one goal and one product—safe superintelligence—and that its business model is insulated from short-term commercial pressures. | 中 | SV014 |
| CV014 | Built In, The Economic Times, and Computing all reported that SSI reached a $32 billion valuation in a 2025 financing round. | 中 | SV015, SV016, SV017 |
| CV015 | Built In said SSI had not commercially launched a product when it reached the $32 billion valuation. | 中 | SV015 |
| CV016 | The Economic Times reported that Google Cloud gave SSI access to TPUs, showing that strategic compute access can sit alongside equity financing for frontier labs. | 中 | SV016, SV017 |
| CV017 | Crunchbase said Q1 2026 global venture investment hit $300 billion and that AI absorbed 80% of the total, with OpenAI, Anthropic, xAI, and Waymo taking 65% of quarter funding. | 中 | SV019 |
| CV018 | Crunchbase's frontier-lab snapshot said foundational AI funding in Q1 2026 alone doubled all of 2025 and was concentrated in a handful of giants. | 中 | SV020 |
| CV019 | Lower-tier market trackers also described 2026 AI funding as a barbell where giant frontier labs and specialized infrastructure captured most capital. | 低 | SV021, SV022 |
| CV020 | State of AI 2025 said OpenAI retained only a narrow lead at the frontier and that competition intensified, supporting the view that elite team density itself has become a scarce asset. | 中 | SV023 |
| CV021 | J.P. Morgan said agentic and inference-heavy usage can require 10x to 100x more compute per user than earlier AI workloads. | 中 | SV024 |
| CV022 | J.P. Morgan also said supply shortages now span chips, power, and data-center infrastructure, with HBM suppliers sold out for 2026. | 中 | SV024 |
| CV023 | Colliers reported more than $580 billion of global data-center investment in 2025, with more than 90% of new capacity pre-leased and power consuming 40%-50% of project costs. | 中 | SV025 |
| CV024 | The combination of a large all-cash round and unusually scarce compute inputs means humans&' war chest is economically more valuable than a normal seed software balance sheet. | 中 | SV004, SV024, SV025 |
| CV025 | CompaniesMarketCap showed CoreWeave at roughly $64.35 billion of market capitalization in June 2026. | 中 | SV026 |
| CV026 | CoreWeave had already filed a Form 10-Q for the quarter ended March 31, 2026, underscoring how much more public operating disclosure exists for scaled AI infrastructure than for humans&. | 中 | SV027 |
| CV027 | C3.ai reported $250.3 million of fiscal 2026 revenue in its June 2026 results release. | 中 | SV028 |
| CV028 | CompaniesMarketCap showed C3.ai at roughly $1.49 billion of market capitalization in June 2026. | 中 | SV029 |
| CV029 | At the quoted $4.48 billion round price, humans& was valued at roughly 3.0 times C3.ai's June 2026 market cap despite humans& having no public revenue disclosure. | 中 | SV002, SV028, SV029 |
| CV030 | Relative to other 2025-2026 frontier-lab financings, humans& sits well below SSI's $32 billion and Thinking Machines Lab's $10-12 billion marks, but still inside the same mega-seed cohort. | 中 | SV002, SV011, SV012, SV015, SV016, SV017 |
| CV031 | Using the public headcount signals of roughly 20 employees at launch and 28 by late May, the quoted $4.48 billion valuation implies about $160 million to $224 million of value per reported employee. | 中 | SV002, SV006 |
| CV032 | AI Funding Tracker characterized humans& as a three-month-old startup with no product that became a unicorn on day one. | 中 | SV030 |
| CV033 | The retained public sources usually state a $4.48 billion valuation but do not provide a retained term sheet clarifying whether that figure is pre-money or post-money. | 中 | SV002, SV003, SV004, SV030 |
| CV034 | Because valuation basis and security terms are undisclosed, public evidence cannot precisely quantify seed ownership, liquidation preferences, or future dilution. | 中 | SV001, SV002, SV003, SV004 |
| CV035 | No retained public source ties humans&' quoted valuation to revenue multiples because no retained public source discloses revenue or named customer traction. | 高 | SV001, SV005 |
| CV036 | The strongest public support for humans&' price is option value around elite founders, investor quality, and the ability to buy scarce compute and time before launch. | 中 | SV003, SV004, SV011, SV015, SV024, SV025 |
| CV037 | The strongest public challenge to the price is opacity: no launched product, no disclosed customers, no public operating metrics, and no visible round documents. | 中 | SV001, SV002, SV005, SV030 |
| CV038 | A traction-adjusted base case below the quoted $4.48 billion is more defensible on public evidence than treating the headline price as already fundamental. | 中 | SV005, SV024, SV025, SV028, SV029 |
| CV039 | A plausible supportable public-evidence range is roughly $2.0 billion to $3.5 billion: above listed AI-application comps because of frontier optionality, but below the quoted round because of missing traction and documentation. | 低 | SV024, SV025, SV028, SV029 |
| CV040 | The bull case is that humans& ships a differentiated collaboration product and attracts early customer proof, allowing the current $4.48 billion mark to look like an early entry into a TML- or SSI-style scarcity premium. | 低 | SV010, SV011, SV012, SV014, SV015, SV016 |
| CV041 | The bear case is that humans& remains pre-product into 2027, in which case the market may re-rate it closer to public AI application vendors plus cash rather than to frontier-lab narratives. | 中 | SV005, SV028, SV029 |
| CV042 | On public evidence alone, the current pricing looks stretched rather than attractive or fair. | 中 | SV005, SV024, SV025, SV028, SV029, SV030 |
| CV043 | The appropriate recommendation on public evidence is research-more with high risk because the upside case depends on milestones the public record still does not verify. | 中 | SV001, SV005, SV024, SV025, SV028, SV029 |
| CV044 | The diligence items most likely to move the call are the signed seed term sheet, cap-table and board package, compute procurement contracts or credits, pilot/customer evidence, and product usage data. | 中 | SV001, SV002, SV003, SV004, SV024, SV025 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | humans& | humans& | Today we introduce humans&, a human-centric frontier AI lab. |
| SO002 | humans& | humans& / careers and site legal notices | Effective: January 19, 2026. |
| SO003 | Ashby | humans& Jobs | |
| SO004 | Eric Zelikman | Eric Zelikman | I'm CEO and co-founder of humans&. |
| SO005 | Eric Zelikman | Eric Zelikman | Publications | |
| SO006 | OpenReview | Eric Zelikman profile | PhD student, Stanford University. |
| SO007 | arXiv | Quiet-STaR: Language Models Can Teach Themselves to Think Before Speaking | |
| SO008 | Google Research | STaR: Self-Taught Reasoner Bootstrapping Reasoning With Reasoning | |
| SO009 | TechCrunch | Humans&, a 'human-centric' AI startup founded by Anthropic, xAI, Google alums, raised $480M seed round | The company’s 20-odd employees also come from OpenAI, Meta, Reflection, AI2, and MIT, according to the company. |
| SO010 | Crunchbase News | Humans& Raises Huge $480M Seed Round At $4.48B Valuation For ‘Human-Centric AI Lab’ | Peng told Crunchbase News via email that humans& will spend the majority of the capital on compute for training models. |
| SO011 | Crunchbase | humans& - Crunchbase company profile | |
| SO012 | Bloomberg | Nvidia, SV Angel Set to Back Humans& at $4.48 Billion Valuation | |
| SO013 | Bloomberg | Humans& AI Inc - Company Profile and News | |
| SO014 | Reworked | Humans& Bets $480M That AI Can Be Human-Centric | An AI startup has emerged from stealth with one of technology's largest seed funding rounds, though whether its vision survives contact with enterprise reality remains an open question. |
| SO015 | Tech Funding News | Ex-OpenAI researchers’ Humans& raises $480M: Can it beat Thinking Machines Lab? | |
| SO016 | The SaaS News | humans& Raises $480M Seed at $4.48B Valuation | |
| SO017 | The Economic Times | AI startup Humans& raises $480 million at $4.5 billion valuation in seed round | |
| SO018 | The New York Times | Humans& funding coverage | |
| SO019 | Forbes | xAI Researcher In Talks To Raise $1 Billion For New Frontier Lab Humans& | xAI Researcher In Talks To Raise $1 Billion For New Frontier Lab Humans& |
| SO020 | No Priors | humans& and Eric Zelikman episode page | |
| SO021 | AI Funding Tracker | Humans& Raises $480M Seed Round (2026) | At $4.48 billion post-money valuation, Humans& became a unicorn on day one. |
| SO022 | Bizprofile | Humans& Ai, Inc Redwood City, CA - filing information | Officially filed on October 27, 2025, this corporation is recognized under the document number B20250359156. |
| SO023 | PitchBook | Humans& 2026 Company Profile: Valuation, Funding & Investors | |
| SO024 | Tracxn | humans& | humans& has 28 employees as of May 26. |
| SO025 | AI Market Watch | Humans& - AI Startup Profile | AI Market Watch | |
| SM001 | Gartner | Gartner Forecasts Worldwide AI Spending to Grow 47% in 2026 | |
| SM002 | Gartner | Gartner Says Worldwide AI Spending Will Total $1.5 Trillion in 2025 | |
| SM003 | IDC | IDC’s Worldwide AI and Generative AI Spending – Industry Outlook | |
| SM004 | Stanford Human-Centered Artificial Intelligence | The 2026 AI Index Report | |
| SM005 | Stanford Human-Centered Artificial Intelligence | AI Index Report 2026 PDF | |
| SM006 | Deloitte | The State of AI in the Enterprise - 2026 AI report | |
| SM007 | Deloitte | Four futures of generative AI in the enterprise: Scenario planning for strategic resilience and adaptability | |
| SM008 | McKinsey & Company | The state of AI in 2025: Agents, innovation, and transformation | |
| SM009 | Microsoft AI Economy Institute | Global AI Adoption in 2025 | |
| SM010 | National Institute of Standards and Technology | AI Risk Management Framework | |
| SM011 | National Institute of Standards and Technology | AI Standards | |
| SM012 | European Commission | AI Act | |
| SM013 | OECD | OECD.AI | |
| SM014 | Deloitte | Deloitte Global’s 2025 Predictions Report: Generative AI: Paving the Way for a transformative future in Technology, Media, and Telecommunications | |
| SM015 | Entropy (MDPI) | Human-in-the-Loop Artificial Intelligence: A Systematic Review of Concepts, Methods, and Applications | |
| SM016 | arXiv | Human-in-the-loop or AI-in-the-loop? Automate or Collaborate? | |
| SM017 | RAND | Why AI Projects Fail and How They Can Succeed | |
| SM018 | Boston Consulting Group | AI Adoption in 2024: 74% of Companies Struggle to Achieve and Scale Value | |
| SM019 | Deloitte | AI ROI: The paradox of rising investment and elusive returns | |
| SM020 | Bain & Company | Your AI Budget Is Growing. Your Returns Aren't. Here's Why. | |
| SM021 | PwC | Want ROI from AI? Go for growth | |
| SM022 | MarketsandMarkets | AI Agents Market Report 2025-2030, by Application, Geo, Tech | |
| SM023 | MarketsandMarkets | Human in the Loop Market Revenue Trends and Growth Drivers | |
| SM024 | The Business Research Company | Global Human-In-The-Loop Artificial Intelligence (AI) Market Report 2026 | |
| SM025 | IEEE Spectrum | Stanford's AI Index for 2026 Shows the State of AI | |
| SP001 | humans& | humans& | At its best, AI should serve as a deeper connective tissue that strengthens organizations and communities. |
| SP002 | TechCrunch | Humans&, a 'human-centric' AI startup founded by Anthropic, xAI, Google alums, raised $480M seed round | |
| SP003 | TechCrunch | Humans& thinks coordination is the next frontier for AI, and they're building a model to prove it | We are building a product and a model that is centered on communication and collaboration. |
| SP004 | Thinking Machines Lab | Thinking Machines Lab | Instead of focusing solely on making fully autonomous AI systems, we are excited to build multimodal systems that work with people collaboratively. |
| SP005 | Reuters (via Yahoo Finance) | Mira Murati's AI startup Thinking Machines valued at $12 billion in early-stage funding | The massive funding round for a company launched only in February, with no revenue or products yet, underscores Murati's ability to attract investors in a sector where top executives have become coveted targets in an escalating talent war. |
| SP006 | Reuters (via Yahoo Finance) | Mira Murati's Thinking Machines seeks $50 billion valuation in funding talks, Bloomberg News reports | |
| SP007 | Safe Superintelligence Inc. | Safe Superintelligence Inc. | We have started the world’s first straight-shot SSI lab, with one goal and one product: a safe superintelligence. |
| SP008 | Reuters (via Yahoo Finance) | Exclusive-OpenAI co-founder Sutskever's SSI in talks to be valued at $20 billion, sources say | |
| SP009 | Reuters (via Yahoo Finance) | Exclusive-Alphabet, Nvidia invest in OpenAI co-founder Sutskever's SSI, source says | |
| SP010 | Anthropic | Plans & Pricing | Claude by Anthropic | |
| SP011 | Anthropic | Model system cards | |
| SP012 | Anthropic | Anthropic’s Transparency Hub | |
| SP013 | OpenAI | ChatGPT for enterprise | |
| SP014 | OpenAI Developers | Evals | OpenAI Developers | |
| SP015 | OpenAI API | Working with evals | OpenAI API | |
| SP016 | OpenAI API | Pricing | OpenAI API | |
| SP017 | xAI Docs | Overview | xAI Docs | |
| SP018 | xAI Docs | Models | xAI Docs | |
| SP019 | Scale AI | Scale AI | Evaluation and monitoring of enterprise-grade model builders | |
| SP020 | Scale AI | Scale Data Engine | AI Training Data at Scale | |
| SP021 | Scale AI | Quality RLHF Data For Natural Language Generation & Large Language Models | Scale AI | |
| SP022 | Scale AI | Our plan to build a robust test & evaluation platform | |
| SP023 | Associated Press via Yahoo Finance | Meta invests $14.3B in AI firm Scale and recruits its CEO for 'superintelligence' team | Scale said the $14.3 billion investment puts its market value at over $29 billion. |
| SP024 | Labelbox | Plans & Pricing | Labelbox | |
| SP025 | Labelbox | Evaluate and rate frontier AI models | |
| SP026 | Labelbox | Labelbox Evals | |
| SP027 | Labelbox | Reinforcement learning from human feedback (RLHF) | |
| SP028 | Labelbox Docs | LLM human preference - Labelbox | |
| SP029 | Braintrust | Braintrust - The AI observability platform for building quality AI products | |
| SP030 | Braintrust | Pricing - Braintrust | |
| SP031 | Arize AI | Pricing | |
| SP032 | Arize AI | Phoenix | |
| SP033 | Humanloop | Humanloop: LLM evals platform for enterprises | |
| SP034 | Humanloop | Humanloop joins Anthropic | As we sunset the Humanloop platform, we will continue to work closely with our customers to make their transition as smooth as possible. |
| SP035 | Stanford HAI | The 2026 AI Index Report | Stanford HAI | |
| SI001 | humans& | humans& | Today we introduce humans&, a human-centric frontier AI lab. |
| SI002 | Ashby | humans& Jobs | |
| SI003 | TechCrunch | Humans&, a 'human-centric' AI startup founded by Anthropic, xAI, Google alums, raised $480M seed round | Humans&, a startup with a philosophy that AI should empower people rather than replace them, has raised $480 million in seed funding at a $4.48 billion valuation. |
| SI004 | Reuters / U.S. News | AI Startup Humans& Raises $480 Million at $4.5 Billion Valuation in Seed Round | raised $480 million in a seed financing round, which values the company at $4.48 billion |
| SI005 | Crunchbase News | Humans& Raises Huge $480M Seed Round At $4.48B Valuation For Human-Centric AI Lab | Peng told Crunchbase News via email that humans& will spend the majority of the capital on compute for training models. |
| SI006 | Forbes | Humans& Raises $480 Million, Eleven Labs Launches AI Music, Xreal Sues Viture | Three month old AI startup Humans& raises a $480 million seed round at a $4.48 billion valuation, one of the largest seed financings on record. |
| SI007 | Securities and Exchange Commission | Form D Data Sets | The Form D Data Sets below provide the structured data from Notices of Exempt Offerings of Securities filed with the Commission. |
| SI008 | Securities and Exchange Commission | EDGAR Full Text Search | Company name, ticker, CIK number or individual's name |
| SI009 | TechStartups | AI startup Humans& raises $480M seed at $4.48B valuation as former OpenAI and Google researchers launch frontier AI lab | The company is based in San Francisco and confirmed it will work closely with Nvidia on both hardware and software. |
| SI010 | Built In | New AI Startup Humans& Raises $480M Seed at $4.8B Valuation | The lofty cost of using existing AI techniques to train automated models in new ways is the driving factor behind the company’s sizeable capital raise. |
| SI011 | Tracxn | humans& | humans& has 28 employees as of May 26. |
| SI012 | The SaaS News | humans& Raises $480M Seed at $4.48B Valuation | The company will use the funding to advance its AI platform for human collaboration. |
| SI013 | AI Funding Tracker | Humans& Raises $480M Seed Round (2026) | enterprise incentive structures nearly always favor cost reduction and headcount elimination |
| SI014 | OpenAI | OpenAI API Pricing | GPT-5.5 Input: $5.00 / 1M tokens ... Output: $30.00 / 1M tokens |
| SI015 | OpenAI | ChatGPT Plans | Free, Go, Plus, Pro, Business, and Enterprise | Paid plans (Go, Plus, Business, and Enterprise) are priced per user per month. |
| SI016 | Anthropic | Plans & Pricing | Claude by Anthropic | $20 if billed monthly. |
| SI017 | Google Workspace | Compare Flexible Pricing Plan Options | Business Standard $14 / user per month. |
| SI018 | Microsoft | AI for Enterprise Productivity | Microsoft 365 Copilot | Microsoft 365 Copilot Chat is available at no additional cost for all Microsoft Entra account users with an eligible Microsoft 365 subscription. |
| SI019 | Slack | Slack Pricing Plans: Find the Right Fit for Your Team | $7.25 USD per user / month, when paying annually. |
| SI020 | Notion | Notion Pricing Plans: Free, Plus, Business, & Enterprise. | Each time your billing period starts ... your count of paid seats will be synchronized so it exactly matches the amount of members present in your workspace. |
| SI021 | BizProfile | Humans& Ai, Inc Redwood City, CA - filing information | Officially filed on October 27, 2025, this corporation is recognized under the document number B20250359156. |
| SI022 | California Secretary of State | Secretary of State | Secretary of State |
| SI023 | Google Cloud | Gemini for Google Cloud pricing | Gemini Code Assist Standard and Enterprise Pricing Overview |
| SI024 | Microsoft | Microsoft 365 Copilot for Business: Enterprise AI Solutions | Microsoft 365 Copilot | Copilot Chat is available at no additional cost for all Microsoft Entra account users with an eligible Microsoft 365 subscription. |
| SI025 | Notion | Meet your AI team | Notion | Infinite minds, built for teamwork. |
| SI026 | OpenAI | Our approach to advertising and expanding access | OpenAI outlined a plan to begin testing advertisements inside ChatGPT |
| SI027 | X | humans& (@humansand) / X | |
| SE001 | humans& | humans& | Today we introduce humans&, a human-centric frontier AI lab. |
| SE002 | humans& | humans& / careers and legal notices | This Site is primarily informational: we do not run advertising and we do not use analytics or tracking pixels on the Site. |
| SE003 | Eric Zelikman | Eric Zelikman | I'm CEO and co-founder of humans&. |
| SE004 | Eric Zelikman | Eric Zelikman | Publications | |
| SE005 | TechCrunch | Humans& thinks coordination is the next frontier for AI, and they're building a model to prove it | We are building a product and a model that is centered on communication and collaboration. |
| SE006 | TechCrunch | Humans&, a 'human-centric' AI startup founded by Anthropic, xAI, Google alums, raised $480M seed round | The startup aims to use software to help people collaborate with each other — think an AI version of an instant messaging app. |
| SE007 | Crunchbase News | Humans& Raises Huge $480M Seed Round At $4.48B Valuation For ‘Human-Centric AI Lab’ | Peng told Crunchbase News via email that humans& will spend the majority of the capital on compute for training models. |
| SE008 | Reworked | Humans& Bets $480M That AI Can Be Human-Centric | Humans& debuted at unicorn status with no product. |
| SE009 | arXiv | Quiet-STaR: Language Models Can Teach Themselves to Think Before Speaking | Quiet-STaR marks a step towards LMs that can learn to reason in a more general and scalable way. |
| SE010 | arXiv | STaR: Bootstrapping Reasoning With Reasoning | STaR lets a model improve itself by learning from its own generated reasoning. |
| SE011 | Google Research | STaR: Self-Taught Reasoner Bootstrapping Reasoning With Reasoning | |
| SE012 | OpenReview | Eric Zelikman | OpenReview | |
| SE013 | GitHub | ezelikman (Eric Zelikman) · GitHub | quiet-star quiet-star Public |
| SE014 | GitHub | GitHub - ezelikman/quiet-star: Code for Quiet-STaR | Code for Quiet-STaR |
| SE015 | Taylor Sorensen | On joining humans& | Humans& is pretty unique because basically everything we do exists between these spaces. |
| SE016 | Taylor Sorensen | Taylor Sorensen - Taylor Sorensen’s Website | Hi! I’m Taylor Sorensen, a researcher at humans&. |
| SE017 | Taylor Sorensen | Taylor Sorensen CV | |
| SE018 | Saurabh Shah | Saurabh Shah - Member of Technical Staff at humans& | Hello! I work at humans&. We're training AI systems to work with people, not replace them |
| SE019 | Alexis Ross | Blog - Alexis Ross | Why I joined humans& and some (belated) reflections on pursuing AI research with meaning |
| SE020 | Alexis Ross | Alexis Ross | Hi, I'm Alexis! I am a PhD student at MIT CSAIL with Jacob Andreas and a founding researcher at humans& working on human-AI collaboration. |
| SE021 | Niloofar Mireshghallah | Niloofar Mireshghallah | I'm a Member of Technical Staff at humans&. |
| SE022 | Niloofar Mireshghallah | Blog – Niloofar Mireshghallah | |
| SE023 | GitHub | GitHub - abhi-arya1/parallel: The lab notebook where AI agents are collaborators, not assistants. (https://humansand.ai/ hackathon project) | Built solo in 8 hours at the Humans& Product Hackathon. |
| SE024 | GitHub | ezelikman (Eric Zelikman) / Repositories · GitHub | Updated Jan 23, 2026 |
| SE025 | Ashby | humans& Jobs | humans& Jobs |
| SU001 | humans& | humans& | Announcing humans& |
| SU002 | Ashby | humans& Jobs | humans& Jobs |
| SU003 | TechCrunch | Humans&, a 'human-centric' AI startup founded by Anthropic, xAI, Google alums, raised $480M seed round | The startup aims to use software to help people collaborate with each other — think an AI version of an instant messaging app. |
| SU004 | TechCrunch | Humans& thinks coordination is the next frontier for AI, and they're building a model to prove it | Humans& still doesn’t have a product, nor has it been clear about what exactly it might be. |
| SU005 | Crunchbase News | Humans& Raises Huge $480M Seed Round At $4.48B Valuation For ‘Human-Centric AI Lab’ | So far, not much is known about the company. |
| SU006 | CNBC | 'I hate customer-service chatbots': The consumer-AI refund relationship is off to a rocky start | Nearly one in five consumers who have used AI for customer service saw no benefit from the experience. |
| SU007 | PR Newswire | AI Backlash Grows Across US, UK, and Canada: More Customers Reject Bots for Human Support in 2026 | 85% would rather speak to a real person than AI when contacting a business. |
| SU008 | Deloitte | The State of AI in the Enterprise - 2026 AI report | Worker access to AI rose by 50% in 2025, and expectations for scale are high: the number of companies with ≥40% projects in production is set to double in six months. |
| SU009 | Deloitte | Four futures of generative AI in the enterprise: Scenario planning for strategic resilience and adaptability | Organizations have struggled to identify and/or scale clear, high-value use cases that align with critical business goals. |
| SU010 | Deloitte | Getting human and machine relationships right | Organizations are twice as likely to exceed their return on investment expectations for AI when they prioritize work design. |
| SU011 | PwC | Want ROI from AI? Go for growth | 20% of the 1,217 companies we surveyed capture 74% of the AI-driven returns. |
| SU012 | Deloitte Global | AI ROI: The paradox of rising investment and elusive returns | All organizations have one or more working implementations of AI in daily use. |
| SU013 | Microsoft | 2026 Work Trend Index report: Agents, human agency, and the opportunity for every organization | 49% of all conversations support cognitive work. |
| SU014 | World Economic Forum | Invest in the workforce for the AI age: A blueprint for scale, skills and responsible growth | Humans focus on judgment, relationships and trade-offs; areas where context, accountability and trust matter. |
| SU015 | S&P Global | AI impact on employment 2026: Labor market data and outlook | There is a direct relationship between such a discipline and the success rates of AI projects, their return on investment, and the recognition of AI as an integrated, organization-wide capability. |
| SU016 | BCG | AI Adoption in 2024: 74% of Companies Struggle to Achieve and Scale Value | Seventy-four percent of companies have yet to show tangible value from their use of AI. |
| SU017 | BCG | AI Will Reshape More Jobs Than It Replaces | When AI systems can reliably handle these repeatable inquiries end to end, fewer representatives are required. |
| SU018 | Druid AI | Druid's 2026 AI Adoption Benchmark: What production AI usage reveals across four industries | Most published State of AI content captures executive sentiment, budget intent, and pilot plans. Druid's benchmark adds a different signal: production behavior. |
| SU019 | G2 | Best Artificial Intelligence Software: User Reviews from June 2026 | These software solutions are ranked using an algorithm that calculates customer satisfaction and market presence based on reviews from our user community. |
| SU020 | G2 | Best Project Collaboration Software: User Reviews from January 2026 | Project collaboration software aims to increase the productivity of employees involved in project management by streamlining communications, collaboration, and remote work. |
| SU021 | Capterra | Best Artificial Intelligence Software 2026 | AI-powered chatbots and virtual assistants can respond to natural language queries, providing customer support and personalized recommendations. |
| SU022 | Capterra | Best Collaboration Software 2026 | Collaboration software offers many benefits to an organization that results in a streamlined workflow and effective completion of tasks and goals. |
| SU023 | Gartner | Gartner Forecasts Worldwide AI Spending to Grow 47% in 2026 | Enterprises will expand their use of both the GenAI models embedded in existing software applications and the new AI agents within multiple workflows. |
| SU024 | Bain & Company | Your AI Budget Is Growing. Your Returns Aren't. Here's Why. | Only 7% of companies are running fully autonomous agents in production today. |
| SU025 | RAND | Why AI Projects Fail and How They Can Succeed | By some estimates, more than 80 percent of AI projects fail. |
| SU026 | Microsoft AI Economy Institute | Global AI Adoption in 2025 | Global adoption of artificial intelligence continued to rise in the second half of 2025. |
| SR001 | humans& | humans& | |
| SR002 | humans& | humans& Terms of Use | humans& reserves the right, in its sole discretion, to restrict, suspend, or terminate this Agreement and your access to all or any part of the Web Site or the Content at any time and for any reason without prior notice or liability. |
| SR003 | humans& | humans& Privacy Notice | We may share information in the following limited circumstances: ... if we are involved in a merger, acquisition, financing, or sale of assets, information may be transferred as part of that transaction. |
| SR004 | TechCrunch | Humans&, a 'human-centric' AI startup founded by Anthropic, xAI, Google alums, raised $480M seed round | The company’s 20-odd employees also come from OpenAI, Meta, Reflection, AI2, and MIT, according to the company. |
| SR005 | Reworked | Humans& Bets $480M That AI Can Be Human-Centric | Humans& debuted at unicorn status with no product. |
| SR006 | Tech Funding News | Ex-OpenAI researchers’ Humans& raises $480M: Can it beat Thinking Machines Lab? | |
| SR007 | Forbes | xAI Researcher In Talks To Raise $1 Billion For New Frontier Lab Humans& | Former xAI researcher Eric Zelikman is raising $1 billion for a new startup called Humans&, that will train AI models to be better at collaborating with humans. |
| SR008 | Crunchbase News | Humans& Raises Huge $480M Seed Round At $4.48B Valuation For Human-Centric AI Lab | Peng told Crunchbase News via email that humans& will spend the majority of the capital on compute for training models. |
| SR009 | Built In | New AI Startup Humans& Raises $480M Seed at $4.8B Valuation | |
| SR010 | SiliconANGLE | Newly launched AI startup Humans& raises $480M round backed by Nvidia, GV | |
| SR011 | TechStartups | AI startup Humans& raises $480M seed at $4.48B valuation as former OpenAI and Google researchers launch frontier AI lab | |
| SR012 | The Economic Times | AI startup Humans& raises $480 million at $4.5 billion valuation in seed round | |
| SR013 | The SaaS News | humans& Raises $480M Seed at $4.48B Valuation | |
| SR014 | Reuters | In the AI age, firms chase growth but with fewer workers | |
| SR015 | CNBC | Behind the AI talent war: Why tech giants are paying millions to top hires | As long as companies will have to spend billions of dollars to build the model, they will spend tens of millions, or hundreds of millions, to hire engineers to build those models. |
| SR016 | Forbes | The AI Talent Wars Have Hit Data Labeling | You need to understand what you’re talking about and it’s a very complex service so there is a shortage of talent that understands how to work on these pipelines. |
| SR017 | National Institute of Standards and Technology | AI Risk Management Framework | |
| SR018 | National Institute of Standards and Technology | NIST AI 100-1: AI RMF 1.0 | |
| SR019 | National Institute of Standards and Technology | NIST AI 600-1: Generative Artificial Intelligence Profile | |
| SR020 | U.S. Copyright Office | Copyright and Artificial Intelligence | Part 3: Generative AI Training (Pre-publication) |
| SR021 | U.S. Copyright Office | Copyright and Artificial Intelligence, Part 3: Generative AI Training (Pre-Publication Version) | |
| SR022 | European Commission | AI Act | |
| SR023 | European Commission | Guidelines for providers of general-purpose AI models | From 2 August 2026, the Commission’s enforcement powers enter into application. |
| SR024 | International AI Safety Report | International AI Safety Report 2026 | |
| SR025 | OpenAI | Safety & responsibility | |
| SR026 | OpenReview | Eric Zelikman profile | |
| SR027 | Reuters / U.S. News | AI Startup Humans& Raises $480 Million at $4.5 Billion Valuation in Seed Round | Humans& said it was working on human-centric AI tools for communication and collaboration, and expects to launch a product early this year. |
| SR028 | Bizprofile | Humans& Ai, Inc Redwood City, CA - filing information | |
| SR029 | Tracxn | humans& | |
| SR030 | Ashby | humans& Jobs | |
| SV001 | humans& | humans& | Today we introduce humans&, a human-centric frontier AI lab. |
| SV002 | TechCrunch | Humans&, a 'human-centric' AI startup founded by Anthropic, xAI, Google alums, raised $480M seed round | Humans&, a startup with a philosophy that AI should empower people rather than replace them, has raised $480 million in seed funding at a $4.48 billion valuation. |
| SV003 | Reuters / U.S. News | AI Startup Humans& Raises $480 Million at $4.5 Billion Valuation in Seed Round | AI startup Humans&, founded by former OpenAI, Alphabet and xAI researchers, has raised $480 million in a seed financing round, which values the company at $4.48 billion. |
| SV004 | Crunchbase News | Humans& Raises Huge $480M Seed Round At $4.48B Valuation For Human-Centric AI Lab | Peng told Crunchbase News via email that humans& will spend the majority of the capital on compute for training models. |
| SV005 | Reworked | Humans& Bets $480M That AI Can Be Human-Centric | Humans& debuted at unicorn status with no product. |
| SV006 | Tracxn | humans& | humans& has raised $480M in funding. |
| SV007 | Forbes | xAI Researcher In Talks To Raise $1 Billion For New Frontier Lab Humans& | xAI Researcher In Talks To Raise $1 Billion For New Frontier Lab Humans& |
| SV008 | Tech Funding News | Ex-OpenAI researchers’ Humans& raises $480M: Can it beat Thinking Machines Lab? | Ex-OpenAI researchers' Humans& raises $480M: Can it beat Thinking Machines Lab? |
| SV009 | TechStartups | AI startup Humans& raises $480M seed at $4.48B valuation as former OpenAI and Google researchers launch frontier AI lab | The company just raised a staggering $480 million seed round, pushing its valuation to $4.48 billion — a rare figure for a startup that has yet to ship its first product. |
| SV010 | Thinking Machines Lab | Thinking Machines Lab | Thinking Machines Lab is an artificial intelligence research and product company. |
| SV011 | TechCrunch | Mira Murati’s Thinking Machines Lab closes on $2B at $10B valuation | Thinking Machines Lab, the secretive AI startup founded by OpenAI’s former chief technology officer Mira Murati, has closed a $2 billion seed round. |
| SV012 | TechCrunch | Mira Murati's Thinking Machines Lab is worth $12B in seed round | The deal, which includes participation from Nvidia, Accel, ServiceNow, CISCO, AMD, and Jane Street, values the startup at $12 billion. |
| SV013 | TechCrunch | Thinking Machines wants to build an AI that actually listens while it talks | Still, this is a research preview, not a product. |
| SV014 | Safe Superintelligence Inc. | Safe Superintelligence Inc. | We have started the world’s first straight-shot SSI lab, with one goal and one product: a safe superintelligence. |
| SV015 | Built In San Francisco | AI Innovator Safe Superintelligence Raises $2B at $32B Valuation | The company has yet to commercially launch its product. |
| SV016 | The Economic Times | Alphabet, Nvidia invest in OpenAI cofounder Sutskever's SSI | SSI, which sources say was recently valued at $32 billion in a round led by Greenoaks, is one of the highest-profile startups working on AI model research. |
| SV017 | Computing | Alphabet and Nvidia back Ilya Sutskever's Safe Superintelligence | The investment comes amid a fresh $2 billion funding round that catapults SSI's valuation to $32 billion. |
| SV018 | Tech Funding News | $2B raise at $32B valuation: 5 facts OpenAI co-founder’s Safe Superintelligence | $2B raise at $32B valuation: 5 facts OpenAI co-founder’s Safe Superintelligence. |
| SV019 | Crunchbase News | Q1 2026 Shatters Venture Funding Records As AI Boom Pushes Startup Investment To $300B | Four of the five largest venture rounds ever recorded were closed in Q1 2026. |
| SV020 | Crunchbase News | Sector Snapshot: Venture Funding To Foundational AI Startups In Q1 Was Double All Of 2025 | Funding to foundational AI startups, also known as generative AI companies or frontier labs, has doubled in the first quarter of 2026 so far compared to all of 2025. |
| SV021 | AI Funding Tracker | AI Funding Tracker | AI Startup Investment Roundups 2026 | The biggest private AI rounds in 2026 are concentrating in a small set of frontier labs and infrastructure companies. |
| SV022 | FeedTheAI | Biggest AI Funding Rounds of 2026 (So Far) | AI funding in 2026 is concentrating. |
| SV023 | State of AI | State of AI Report 2025 | OpenAI retains a narrow lead at the frontier, but competition has intensified. |
| SV024 | J.P. Morgan Asset Management | Is AI running out of compute? | A single user could demand 10-100x more compute with the latest tools and use cases. |
| SV025 | Colliers | 2026 Data Center Marketplace Report | 90%+ of new capacity pre-leased prior to delivery. |
| SV026 | CompaniesMarketCap | CoreWeave (CRWV) - Market capitalization | As of June 2026 CoreWeave has a market cap of $64.35 Billion USD. |
| SV027 | Securities and Exchange Commission | CoreWeave, Inc. Form 10-Q for quarter ended March 31, 2026 | For the quarterly period ended March 31, 2026. |
| SV028 | C3 AI | C3 AI Announces Fiscal Fourth Quarter and Full Fiscal Year 2026 Results | Full Fiscal Year 2026 Financial Highlights: Total Revenue was $250.3 million. |
| SV029 | CompaniesMarketCap | C3 AI (AI) - Market capitalization | As of June 2026 C3 AI has a market cap of $1.49 Billion USD. |
| SV030 | AI Funding Tracker | Humans& Raises $480M Seed Round (2026) | On January 20, 2026, a three-month-old AI startup with no product announced a $480 million seed round at a $4.48 billion valuation. |