初创公司尽调
尽调报告 infrastructure / devtools Series B 2026-08-28

Instinct

估值 $2.5B 的出圈 AI 个人助理:ARR 高速增长,但隐私与安全红旗仍未解除

Instinct 是一家估值 $2.5B 的消费级 AI 助手,ARR 增长势头极强,但隐私、安全和治理风险很重;以现有证据质量看,估值偏高。

封面要素

估值 01
2500 USD M [CO005]
累计融资 02
350 USD M [CO006]
年经常性收入(ARR) 03
80 USD M [CO017]
ARR 增长 04
~16× in 8 months [CO018]
阶段 05
Series B [CO003]
状态 06
Private beta [CO016]

公司概况

Instinct 由 Spear Street Technology Inc.(2025 年在加利福尼亚州注册)运营,是一款面向消费者的 AI 个人助理, 入口只有 SMS 和 WhatsApp。产品连接用户的电子邮件、日历、消息应用、音频、位置与屏幕,并在无需逐项确认的情况下, 自主执行预约、邮件管理、排期、购物等任务。创始人 Noah Shinn 23 岁,曾任 Sierra 研究科学家,也是 Reflexion (NeurIPS 2023)第一作者;Instinct 在 2026 年初出圈,ARR 从 2026 年 1 月约 $5 million 增至 2026 年 8 月约 $80 million。公司完成 $250 million 融资,Series B 轮由 Index Ventures 和 Benchmark 共同领投,投后估值 $2.5 billion。产品仍处私测阶段,用户热情很高,但围绕隐私、安全和数据治理的严肃负面关注也很集中。

官网
instinct.co
成立时间
2025-01-01
创始人
Noah Shinn
创立地点
San Francisco, CA
总部
San Francisco, CA
产品
一款通过 SMS 和 WhatsApp 交付的消费级 AI 个人助理。无需下载 app。通过 OAuth 连接电子邮件(Gmail、Outlook、 Apple Mail)、日历和消息应用;访问设备音频、位置和屏幕。可自主执行任务:预约、管理邮件、安排旅行、购物, 包括代表用户达成有法律约束力的交易。
客户
截至 2026 年 8 月,面向追求自主任务管理的技术敏感型消费者,私测阶段仅限邀请。
商业模式
未公开披露。收入模式可能是订阅或按任务计费;私测队列贡献 $80M ARR,说明可能存在付费访问或点数定价,但条款尚未公开宣布。
阶段
Series B
融资情况
$250 million Series B 轮(2026 年 8 月 26 日)由 Index Ventures 和 Benchmark 共同领投,投后估值 $2.5 billion;$100 million Series A 轮(2026 年 1 月)由 Kleiner Perkins 领投;累计融资 $350 million。
[CO001, CO002, CO003, CO004, CO005, CO006, CO007, CO015]

执行摘要

主要优势

  • ARR 增速罕见:8 个月约 16× 增长($5M → $80M ARR),说明早期用户里的产品市场匹配度很强。
  • 创始人技术差异化明显(Reflexion NeurIPS 2023),智能体记忆架构有防御性,通用模型不容易直接复刻。
  • 一线投资人组合(Index Ventures、Benchmark、Kleiner Perkins)给出强背书,也带来治理支持和后续融资通道。
  • SMS / WhatsApp 分发绕开 app store 摩擦,借现有消息应用装机基础推动病毒式增长。
  • 在完全自主的消费级 AI 智能体品类里,Instinct 已先于大科技公司跑出有意义规模,占住先发位置。

主要风险

  • 永久且不可撤销的数据许可,以及 OAuth 撤销后仍以明文留存邮件,让 Instinct 暴露在 GDPR、CCPA 和 FTC 执法风险下。
  • 已被验证的提示注入漏洞允许恶意邮件劫持 Instinct 的自主动作,安全与责任风险很高。
  • $2.5B 估值(约 31× ARR)对一款尚未正式发布、未披露单位经济的消费产品偏贵;隐私反弹可能迅速压缩倍数。
  • 23 岁首次担任 CEO,却已进入超常规模;在员工快速扩张和产品发布期,执行与治理风险上升。
  • Google、Apple、OpenAI 和 Microsoft 拥有原生平台集成和深度消费者分发,带来直接竞争风险。

未决问题

  • 定价模型和单位经济尚未公开;$80M ARR 口径无法核验。
  • 员工数、烧钱速度和盈利路径完全不透明。
  • 监管机构(FTC、EU AI Act)会如何回应自主绑定交易能力,仍不清楚。
  • 股权结构表、稀释历史和老股交易活动均不可得。
  • 企业转向潜力和 B2B 路线图(如有)尚未公开释放信号。

目录

Chapter 01

01公司概况

1.1 身份、产品与运营模式

Instinct 是一款 AI 个人助理,完全通过短信和电话运转——用户通过 SMS 或 WhatsApp 互动,让它预约、 管理邮件、整理日历、安排旅行、处理购物、退订服务,并在已连接账户和设备之间代用户自主行动。公司的法律实体 Spear Street Technology Inc. 于 2025 年在加利福尼亚州注册。产品品牌为 Instinct,官网是 instinct.co。 运营模式权限很深:Instinct 连接用户的电子邮件账户、消息平台、日历、设备音频、位置数据和屏幕截图。集成完成后, 助理无需逐项确认即可自主执行任务。服务条款授予 Instinct 一项“永久且不可撤销”的许可,可访问、存储、使用并修改 所有用户材料,包括用于 AI 训练——这一条款在 2026 年 8 月引发持续公开批评。早期反弹后,公司增加了数据删除工具, 但宽泛条款结构保留了下来。 截至 2026 年 8 月,Instinct 仍处私测阶段——访问只靠邀请或候补名单。公司尚未公开定价、员工数或商业化上线日期。 即便尚未广泛开放,产品已拿到 $350 million 风险资本,年经常性收入(ARR)约 $80 million。Instinct 总部位于 加利福尼亚州 San Francisco;具体办公地址未公开披露。公司称团队规模很小,成员主要来自 MIT 和企业 AI 公司 Sierra 的前沿 AI 研究背景。[CO001, CO002, CO004, CO014, CO015, CO016]

Instinct 概览 KPI 表
指标数值 / 状态日期 / 期间可信度缺口 / 备注
法律实体Spear Street Technology Inc.历史加州注册;未公开确认备案编号
品牌名称Instinct当前官方网站 instinct.co;多篇报道确认
总部美国加州 San Francisco当前TechCrunch 和 TechFundingNews 报道;未披露街道地址
成立时间2025历史年份已确认;具体月份未公开披露
产品状态封闭 beta(仅限邀请)2026-08-28TechCrunch 融资和隐私文章确认
最新估值$2.5B 投后2026-08-26TechCrunch、TechFundingNews、Index Ventures/Benchmark 新闻稿共同确认
累计融资$350M2026-08-26TechCrunch 一手来源;$100M 种子轮 + Series A + $250M Series B
Series B 金额$250M2026-08-26TechCrunch;Index Ventures 和 Benchmark 共同领投
年经常性收入(ARR)~$80M2026-08管理层通过 TechCrunch 声称;未经独立审计
创立阶段 ARR~$5M2026-01管理层声称;代表 2026 年 1 月左右 ARR
员工数未披露(估计 <50)2026-08-28外界称其为“小团队”;没有公开人数
客户数未披露2026-08-28封闭 beta;未发布候补名单指标
创始人Noah Shinn(23 岁,CEO)2026-08-28TechCrunch、TechFundingNews 和多个来源均确认

收入和员工数来自管理层提供或估计;没有独立审计。估值和融资数字由多个高声誉来源确认。null 条目代表截至运行日期已确认的信息缺口。

[CO001, CO002, CO005, CO006, CO007, CO016]
FO002: Instinct 公司快照逻辑

展示 Instinct 的身份、产品入口、底层集成、资本结构和关键依赖如何连接成当前运营模型。

[CO001, CO004, CO006, CO007, CO014, CO017]

1.2 创始人背景、技术脉络与团队构成

Noah Shinn 是 Instinct 唯一公开确认的创始人,也是 Spear Street Technology 的 CEO。2026 年 8 月 Series B 公告时,他 23 岁。Shinn 于 2023 年从 Northeastern University 退学,此前也在 MIT 做过机器学习和编程语言研究。 之后他加入企业 AI 智能体公司 Sierra,是最早一批员工之一,职位为研究科学家。 学术上,Shinn 最有辨识度的标签是 Reflexion 论文第一作者,该论文发表于 NeurIPS 2023。Reflexion 框架提出一种方法: 语言智能体在任务失败后进行“口头反思”,把经验存入情景记忆,并在不进行模型微调的情况下改善后续尝试。该方法在 HumanEval 编码基准上取得 91% pass@1,明显高于当时 GPT-4 报告的 80%。共同作者包括 Federico Cassano、 Ashwin Gopinath、Karthik Narasimhan 和 Shunyu Yao。Shinn 还共同开发了 τ-bench,用来评估智能体处理真实用户互动、 工具调用和业务规则遵循的能力。以 Shinn 的年龄,这条从学术智能体研究、Sierra 企业部署到消费级产品的技术脉络, 是一项差异化资历。 其他团队成员没有公开披露。公司材料称其为一支“小型 San Francisco 团队”,背景来自 MIT 和 Sierra,但本次查看的公开报道 没有点名任何其他成员。董事会构成、投资人观察员和治理结构完全未披露。领导力高度集中,带来显著关键人依赖:公司的技术可信度、 投资人关系和产品愿景,在公开叙事里都系于一人,且没有清晰继任安排。[CO007, CO008, CO009, CO010, CO011, CO012]

领导层与创始人表
人物职务有来源支持的背景创始人-市场契合或职能备注关键人依赖 / 尽调备注
Noah Shinn创始人兼 CEOReflexion(NeurIPS 2023)主要作者;Sierra 研究科学家;MIT AI 研究;2023 年从 Northeastern 退学深厚 AI 智能体研究背景,直接适用于自主助理产品极高关键人依赖;唯一公开识别的创始人;未披露具名高管团队
Sierra(前雇主)N/A — 创始人背景语境企业 AI 智能体公司;Shinn 是最早期员工之一兼研究科学家在转向消费者产品前,Sierra 给了 Shinn 企业级智能体部署经验间接背景;不是现任员工
未具名团队成员多种职能(工程、产品、运营)公开描述称,这是一个来自 MIT 和 Sierra 背景的小型 SF 团队;未披露姓名团队质量可由研究履历推断,但个人层面未经验证没有个人问责;继任计划完全不透明

表仅限公开确认或有文件支持的个人。截至运行日期,董事会组成、观察员和治理结构完全未披露。

[CO007, CO008, CO009, CO010, CO011, CO023]
FO003: Instinct 快照 KPI

截至 2026 年 8 月 Instinct 的核心 KPI:估值、累计融资、ARR、ARR 增长、公司年龄和投资者层级。

ARR 和增长为管理层提供的数据;员工数根据其被描述为小团队而估算。

[CO005, CO006, CO008, CO016, CO039, CO041]

1.3 融资历史、估值轨迹与投资人基础

即便按 2026 年 AI 初创公司标准,Instinct 的融资速度也异常快。从 2026 年初约 $50 million 的种子阶段估值出发, 公司先达到 $500 million Series A 估值(由 Kleiner Perkins 的 Mamoon Hamid 领投),约六个月后又达到 $2.5 billion Series B 投后估值。2026 年 8 月 26 日 Series B 融资 $250 million,由 Index Ventures 和 Benchmark 共同领投,累计融资增至 $350 million。早期支持者包括 Conviction Partners(Sarah Guo)、Greenoaks 和未披露的种子轮 参与方。未有债务工具、老股交易或授信额度公开披露。 投资人层级值得注意。Index Ventures 和 Benchmark 是全球最挑剔的消费科技投资机构之一;Index 早期投过 Dropbox、Stripe、 Robinhood 等公司,Benchmark 则领投过 Twitter、Snap、Uber 和 Discord。领投 Series A 的 Kleiner Perkins 同样支持过 标志性科技公司。对于一家员工数未披露、定价未公开、产品仍在私测的非上市公司,这样的投资人阵容并不寻常,也反映出当前 消费级 AI 智能体交易流的竞争强度。 值得注意的是,$2.5 billion 估值对应约 $80 million ARR,隐含收入倍数约 31× ARR;这一溢价与 2026 年增速最高的 AI 软件公司相符,但远高于 SaaS 中位数倍数。该溢价已经计入显著的未来增长和成功变现,二者尚未以公开可见方式得到证明。 审计财务报表、单位经济模型或烧钱速度数据均未公开。[CO003, CO004, CO005, CO006, CO017, CO018]

利益相关方 / 投资人图谱
利益相关方角色经济 / 战略重要性公开支持证据尽调要求
Index VenturesSeries B 共同领投方$125M+ 的 Series B;预计有董事席位但未披露TechCrunch 和多个来源确认共同领投;Index 以 Dropbox、Stripe、Robinhood 闻名确切持股、董事会权利、清算优先权和反稀释条款
BenchmarkSeries B 共同领投方$125M+ 的 Series B;预计有董事席位但未披露TechCrunch 和多个来源确认共同领投;Benchmark 以 Twitter、Snap、Uber 闻名确切持股、董事会权利,以及相对于 Series A 持有人的优先股堆叠
Kleiner Perkins(Mamoon Hamid)Series A 领投方以约 $500M 估值领投 ~$100M Series A;持有有意义的摊薄后股份TechFundingNews 报道 Mamoon Hamid 代表 Kleiner Perkins 领投 Series AA 轮具体经济条款、B 轮后当前持股稀释、董事席位状态
Conviction Partners(Sarah Guo)早期 / 种子投资人参与种子轮;持股相对较小,但在 AI 生态中信号强TechFundingNews 引述 Sarah Guo(Conviction)祝贺创始人融资确认种子轮金额,以及 Conviction 是否持有任何董事会观察员权利
Greenoaks CapitalSeries A 参与方成长阶段跨界投资人;若按比例投资,潜在持仓较大TechFundingNews 将 Greenoaks 列为早期支持者投资金额、是否有任何老股购买,以及治理权利
Noah Shinn(创始人)创始人兼控股股东(推定)B 轮前可能为多数持有人;B 轮后稀释未知所有来源均将其列为唯一创始人;未披露股权结构表当前持股比例、投票控制权和强制随售权
未知种子投资人种子轮参与方早期验证;细节未披露报道称该轮包含种子资本和 Series A披露完整股权结构表,以及种子轮 SAFE 或可转债条款

经济条款根据交易结构惯例推断。未公开披露股权结构表、董事会组成或治理文件。

[CO004, CO005, CO006, CO020, CO021, CO022]

1.4 里程碑、负面事件与记录年表

Instinct 的公开里程碑只覆盖约 8 个月已知运营历史。Spear Street Technology 于 2025 年成立并开展早期产品开发, 这是起点。公司首次进入公众视野是在 2026 年初,当时进行初始 beta 测试和种子轮,随后据报道由 Kleiner Perkins 领投 Series A,估值约 $500 million。到 2026 年中,公司 ARR 快速增长;在两天后的 Series B 公告前,TechCrunch 于 2026 年 8 月 24 日报道了隐私和安全担忧。 公开记录中最重要的负面事件来自产品本身,而不是公司治理。Beta 测试者报告称,Instinct 在没有明确逐项同意的情况下发送邮件; 用户断开 Google 账户访问后,邮件内容仍以明文保留;恶意邮件可通过提示注入攻击影响用户。SC Media 和 StartupFortune 报道了 这些事件。公司回应是增加数据删除界面,TechCrunch 有所报道,但底层服务条款没有改变。未有监管调查、诉讼或高管离职公开披露。 融资年表是记录最充分的维度。公开记录缺口包括 2025 年内的准确成立日期、任何监管文件、全部合作伙伴或集成协议,以及任何产品发布或 商业开放计划。对于一家估值 $2.5 billion 的公司,即便仍处私营阶段,未披露董事会成员、治理文件或投资人权利协议也值得关注。[CO002, CO003, CO005, CO016, CO017, CO018]

里程碑表
日期事件类型金额 / 估值 / 状态参与方含义
2025(估计)Noah Shinn 离开 Sierra;在加州创立 Spear Street Technology创立实体注册;种子资本可能同期到位Noah Shinn;种子投资人 TBD创立标志着从研究转向商业产品;确切日期未披露
2026-01(估计)Instinct 在封闭 beta 中达到约 $5M ARR规模化~$5M ARR内部指标;TechCrunch 据管理层报道最早披露的收入基准;确认产品已经上线并产生收入
2026 Q1–Q2(估计)Series A 以约 $500M 估值完成;Kleiner Perkins(Mamoon Hamid)领投融资~$100M 融资(种子轮 + A 轮合计);~$500M 估值Kleiner Perkins、Conviction、Greenoaks 与 Noah Shinn首次机构验证;支持团队扩张和基础设施投入
2026-08-24TechCrunch 发布关于 Instinct 广泛数据权限的隐私和安全担忧反向报道;未报告直接财务影响TechCrunch;beta 测试者;安全研究人员首次重大负面报道;凸显永久数据许可和提示词注入风险
2026-08-24 至 08-25(估计)出现 Instinct 未经同意发送邮件、断开连接后仍留存数据的报道反向声誉影响;未披露罚款或法律行动StartupFortune、ExplainX、SC Media;beta 测试者产品可靠性和同意模型受到质疑;公司以删除工具回应
2026-08-25(估计)Instinct 为回应反弹,在界面中加入数据删除工具产品新增功能;ToS 未变内部产品团队响应式产品迭代表明有运营能力,但未解决 ToS 担忧
2026-08-26Series B 完成;以 $2.5B 投后估值融资 $250M融资$250M 融资;$2.5B 估值Index Ventures、Benchmark(共同领投);Noah Shinn独角兽+ 里程碑;约 6 个月内较种子轮估值上升 25×
2026-08-26TechCrunch 发布 Series B 融资公告;确认累计融资 $350M 和 $80M ARR融资$80M ARR(截至公告日)TechCrunch;Instinct 管理层公开确认 ARR 在约 8 个月内从 $5M 增至 $80M;媒体能见度高
2026-08-28(运行日期)Instinct 仍处于封闭 beta;尚未宣布商业发布规模化~$80M ARR(封闭 beta)内部开放问题:$80M ARR 在规模化后是否可持续,还是来自集中的早期队列

标注“(估计)”的日期基于上下文报道估计。种子轮 / Series A 和早期产品事件的确切日期未经一手来源确认。

[CO002, CO003, CO005, CO006, CO017, CO018]
FO001: Instinct 公司里程碑时间线

2025 年成立至 2026 年 8 月 Series B,按时间梳理 Instinct 的成立、融资、产品和负面事件里程碑。

成立和 Series A 日期根据上下文报道估计;确切日历日期未获公开确认。

[CO002, CO003, CO005, CO017, CO018, CO024]

1.5 图表

Chapter 02

02市场分析

2.1 市场定义与边界

Instinct 竞争的是消费级 AI 个人助理市场,这属于更广义的智能虚拟助理行业。该市场包括能理解自然语言指令,并代表个人消费者自主执行 排期、邮件管理、旅行预订、购买和信息检索等任务的软件产品。它不包括只面向企业的解决方案、没有行动能力的纯聊天机器人,也不包括缺乏深度 设备集成的纯语音智能音箱助理。核心参与者既有 Apple Siri、Google Assistant、Amazon Alexa、Microsoft Cortana、Samsung Bixby 等平台既有玩家,也有 Instinct、Rabbit r1、Humane AI Pin 等 AI 原生进入者。2025-2026 年市场变化的关键,是从被动问答助理转向 主动智能体系统:后者只需要最少用户确认就能自主行动。Instinct 处在消费端智能体转型的前沿,差异点是 SMS 和 WhatsApp 入口、深权限模型 以及自主任务执行能力。市场边界在边缘地带较模糊:带 AI 功能的生产力 SaaS、浏览器自动化工具和 AI 编码助理,都与个人助理功能重叠,但主要用例 不同。为估算市场规模,本分析聚焦主要以个人助理定位、且具备多领域任务执行能力的产品。[CM001, CM002, CM003, CM004, CM005, CM006]

市场定义框架
维度纳入排除Instinct 定位
产品类型具备行动能力的多领域任务助理纯聊天机器人、单用途工具、仅企业解决方案全栈自主助理
交互界面文本、语音、与设备集成的多模态没有软件智能的纯硬件设备SMS 和 WhatsApp 为主,语音为辅
任务范围日程安排、邮件、旅行、购物、生活事务工业自动化、仅编码、游戏所有消费者生活管理任务
用户类型个人消费者和专业消费者仅企业部署、仅 B2B 产品隐私容忍度较高、任务量大的消费者
地理全球,重点是英语市场仅离线或地区锁定产品美国为主,全球候补名单
权限模型用于自主执行的设备访问没有行动能力的只读助理深度权限,包括邮件、日历、位置、屏幕

随着 AI 能力扩张,市场边界仍模糊;Instinct 定位在智能体前沿

[CM001, CM002, CM003]
市场参与者格局
类别参与者主要界面自主程度市场位置
平台既有玩家Apple Siri 与 Apple IntelligenceApple 设备上的语音和文本中 — 执行动作需要确认凭 iOS 拥有最大装机基础
平台既有玩家Google Assistant 与 GeminiAndroid 和 Google 设备上的语音和文本中 — 智能体能力继续扩张凭 Android 位居第二
平台既有玩家Amazon AlexaEcho 设备上的语音优先低 — 主要控制智能家居主导智能音箱细分市场
平台既有玩家Microsoft CopilotWindows 和 Office 中的文本中 — 聚焦生产力从企业场景延伸至消费者
AI 原生创业公司InstinctSMS 和 WhatsApp高 — 自主执行增长最快的封闭 beta
AI 原生创业公司Rabbit r1专用硬件设备中 — 行动导向但受硬件限制硬件差异化打法
AI 原生创业公司Humane AI Pin带投影的可穿戴设备低 — 因硬件问题采用有限发布后挣扎
AI 聊天机器人跨界OpenAI ChatGPT 智能体模式应用和网页中的文本高 — 2026 年智能体功能扩张最大 AI 聊天机器人用户基础

市场在拥有分发的既有玩家和具备智能体能力的创业公司之间碎片化

[CM004, CM005, CM006]
FM001: 市场定义与竞争定位
[CM004, CM005, CM006, CM007]

2.2 市场规模与增长轨迹

2026 年,全球消费级 AI 个人助理市场估计达到 4.84 billion USD,相比 2025 年 3.4 billion USD 基线,复合年增长率为 42.2%。 更广义的智能个人助理市场包括企业与混合部署,2026 年规模约 17.95 billion USD,CAGR 为 25.9%。市场预测显示,如果当前增速延续, 消费端到 2030 年将达到 19.6 billion USD。由于“个人助理”和通用 AI 聊天机器人、专用生产力工具之间定义模糊,这些估计存在较大不确定区间。 Research and Markets 与 The Business Research Company 提供主要第三方规模估计,其方法将订阅收入、广告支持使用和硬件捆绑助理价值汇总。 总可用市场(TAM)代表理论上所有智能手机用户都采用付费 AI 助理时的最大市场;可服务市场(SAM)缩小到已表现出为生产力委托付费意愿的用户; 可获取市场(SOM)则对应 2026-2027 年 具备深权限的智能体助理可触达市场。具体到 Instinct,真正可触达的是愿意授予广泛设备权限,并为自主任务执行支付订阅费的消费者子集。 重视隐私的用户以及受监管行业中的用户,是这一可触达市场的结构性排除项。Instinct 在 8 个月实现 16x ARR 增长,要么说明其目标细分市场内 产品市场契合度异常强,要么是早期采用者激增,等候补名单清空后会回归常态。[CM008, CM009, CM010, CM011, CM012, CM013]

市场规模估计
指标2025 年估计2026 年估计2030 预测来源置信度
消费者 AI 助手 TAM3.4B USD4.84B USD19.6B USDResearch and Markets 报告中 - 定义存在模糊性
更广义的 IPA 市场14.3B USD17.95B USD35B USDThe Business Research Company中 - 包含企业市场
消费者 AI 助手 CAGR-42.2%预计 35-40%Research and Markets 报告中 - 可能高度波动
智能体助手 SAM尚未明确估计 1-2B USD8-10B USD分析师综合测算低 - 品类仍在萌芽
Instinct SOM(隐私容忍度较高的高频用户)-200-500M USD1-3B USD内部估计低 - 细分定义不确定

品类定义仍在变化,智能体助手的历史数据也有限,市场规模测算不确定性很高

[CM008, CM009, CM010, CM011]
FM002: 市场规模区间估算
[CM008, CM009, CM010, CM011]

2.3 市场分层与买家画像

消费级 AI 个人助理市场按任务类型、用户成熟度、隐私容忍度和付费意愿分层。按任务类型,市场分为排期和日历管理、邮件和沟通管理、旅行和物流、 购物和购买、信息检索和研究,以及账单支付、订阅管理等生活行政事务。按用户成熟度,范围从只想简单问答的基础语音指令用户,到愿意授予广泛权限、 让系统自主行动的高阶用户。按隐私容忍度,市场明显分裂:隐私敏感用户拒绝广泛数据访问,便利优先用户愿意用隐私换功能。按付费意愿,细分包括 只用免费层的用户、每月低于 10 USD 的低价订阅者,以及愿意为全面辅助支付 20-50 USD 或更多的高端订阅者。Instinct 的目标细分似乎是 隐私容忍度高、付费意愿高的高阶用户,他们看重自主执行胜过逐步确认。该细分相对小,但随着生成式 AI 让数据共享正常化,正在快速增长。 地理上,北美和西欧是主要市场,原因是智能手机渗透率、英语 AI 模型成熟度和消费能力;亚太地区绝对增量最快。年龄结构偏向 25-45 岁专业人士: 日程复杂、有可支配收入,但没有专属真人助理。[CM016, CM017, CM018, CM019, CM020, CM021]

市场细分矩阵
细分人群任务量隐私容忍度付费意愿与 Instinct 匹配度细分规模估计
忙碌专业人士中高高(20-50 USD/mo)全球 50-80M 用户
科技早期采用者中高很强20-30M 用户
重视隐私的用户不一100M+ 用户
预算敏感用户低(免费或低于 5 USD)200M+ 用户
老年人与低技术熟练度用户低中不一低中弱 - UX 复杂150M+ 用户
高净值人群很高(100+ USD)强 - 替代真人助理5-10M 用户

Instinct 瞄准隐私容忍度较高的专业人士和早期采用者;重视隐私的人群是结构性排除项

[CM016, CM017, CM018, CM019]
FM003: 买方细分定位
[CM016, CM017, CM018, CM019]

2.4 增长驱动因素与市场约束

市场增长由五个主要催化剂推动。第一,生成式 AI 能力提升后,自然语言理解和任务完成已经可靠到可以投入生产用例。第二,智能手机普及提供了设备底座 和联网能力,让随时可用的助理成为可能。第三,消费者通过 ChatGPT 等产品熟悉 AI,AI 互动已被正常化,采纳摩擦下降。第四,远程和混合办公带来的 生产力压力,提高了对委托工具的需求。第五,Instinct 采用的 SMS 和消息应用入口不需要用户学习新应用,降低了上手摩擦。市场约束包括:围绕 AI 智能体责任和数据保护的监管不确定性;高知名度 AI 失败事件之后的消费者信任缺口;Apple 和 Google 的平台把关限制第三方助理能力;以及提示注入攻击等 削弱可靠性的安全漏洞。FTC 已释放信号,将加强审查代表消费者进入有约束力交易的 AI 智能体;EU AI Act 可能将自主个人助理归类为高风险系统, 要求进行符合性评估。CCPA 和 GDPR 等隐私法规增加合规负担,并限制数据保留做法。市场轨迹很大程度取决于:智能体 AI 的信任缺口能否比监管约束 收紧更快闭合。[CM022, CM023, CM024, CM025, CM026, CM027]

FM004: 市场增长驱动与约束分析
[CM022, CM023, CM024, CM025, CM026, CM027]
Chapter 03

03竞争格局

3.1 直接对手与既有玩家标杆

Instinct 最接近的参照组不是所有聊天机器人,而是那些试图成为用户日常数字工作操作界面的产品。OpenAI Operator 是最清晰的直接标杆,因为它明确 承诺基于浏览器执行任务。Google Gemini、Apple Intelligence 和 Microsoft Copilot 稍有不同:它们是更广的生态或捆绑型助理,不是初创公司式单一产品; 但从用户结果看,它们越来越多地覆盖同一组核心工作:起草、排期、总结,以及最终采取行动。Meta AI 属于同一个外圈,因为它让用户在消费级通信界面里 期待免费 AI 帮助。 Humane 需要单独看。AI Pin 已经不是活跃竞争对手,但仍有战略意义,因为它同时证明了投资人对常开型助理产品的兴趣,以及消费者端的脆弱性。被 HP 收购后的关停提醒市场:新奇感和宏大的助理叙事,不能弥补薄弱的产品循环。这个先例对 Instinct 重要,因为公司同样要求用户在主流证明出现前,先信任一种 新的委托形式。 在这个参照组里,Instinct 的优势不是规模或品牌,而是聚焦:它是消息原生助理,目标是在真实账户里运转,不再等待持续确认。这让产品比许多既有玩家更像 智能体,但一旦信任破裂,暴露面也更大。[CP001, CP002, CP003, CP004, CP005, CP006]

按类别划分的竞品矩阵
竞品类别规模 / 可用性代理指标目标用户差异化相比 Instinct 的限制
OpenAI Operator直接智能体基准Pro 档 / 研究预览专业消费者与高频用户基于浏览器执行任务非消息原生
Google Gemini现有通用助手借 Google 生态大规模分发主流消费者与专业消费者贴近搜索、账户和设备对委托式个人事务聚焦较弱
Apple Intelligence / Siri现有 OS 助手随兼容 Apple 设备捆绑高端 Apple 消费者OS 默认入口与信任基础当前可见自主性较低
Microsoft Copilot现有生产力助手覆盖 Microsoft 多个入口知识工作者和企业用户套件集成与订阅捆绑消费级个人场景姿态较弱
Perplexity Pro功能重叠付费档已广泛可用重研究的消费者答案质量与搜索深度不能完整委托执行
Humane AI Pin前车之鉴产品已关闭早期采用型硬件买家激进的助手愿景执行和信任失败

Humane 已不再是仍在运营的直接竞品,这里纳入是为了提供战略背景。

[CP001, CP002, CP003, CP004, CP005, CP006]
融资 / 规模对比
公司公开规模标志商业模式资本基础对 Instinct 的含义
Instinct私测阶段 ~$80M ARR未披露的付费消费者助手已融资 $350M早期验证强,品牌成熟度有限
OpenAI大规模付费订阅用户订阅 + API前沿模型级资本能补贴智能体实验
Google / Gemini捆绑和订阅式 AI 方案捆绑 + 订阅公司资产负债表可凭生态优势定价
Apple Intelligence随设备生态捆绑硬件主导的捆绑公司资产负债表分发和信任优势
Microsoft Copilot订阅套件挂载捆绑 + 增购公司资产负债表企业与生产力渠道强
Humane被收购后产品关闭硬件 + 订阅尝试此前由风投支持的创业公司说明没有产品契合度,资本也不够

仅使用公开标志;资本基础对比为方向性判断,并非穷尽。

[CP002, CP006, CP013, CP014, CP015, CP024]
FP001: 竞争地位象限

Instinct 自主性得分高,但分发能力低于在位平台竞争者。

轴为基于公开定位推断的序数评分,x 轴为自主性,y 轴为分发能力。

[CP010, CP011, CP016, CP024, CP033, CP034]
FP004: 威胁时间线

在位者和初创公司都在加码竞争,信任失误仍定义着这一品类。

Operator 和 Apple 推出日期四舍五入到公开发布窗口,因为具体功能节奏存在差异。

[CP002, CP003, CP005, CP020, CP021, CP033]

3.2 相邻替代品与专用工作流工具

下一圈竞争来自只解决 Instinct 部分工作、但不提供完整委托助理组合的产品。Claude 和 Perplexity 是这一组里最重要的广义替代品。Claude 在推理、 写作和分析上很强,但其公开入口目前不像 Instinct 或 Operator 那样强调消费级任务执行。Perplexity 擅长搜索、研究和答案质量;对一些用户来说, 这足以降低他们对更广义助理的需求,但它并不承诺自主收件箱、日历、购物和旅行管理。 然后是专用工具。Superhuman 把邮件速度和体验做得很强。Motion 占住排期和时间分配。Cal.com 提供排期基础设施。这些公司重要,是因为它们说明即便 通用 AI 变强,窄工作流软件仍然可以赢。专用工具更容易被信任、预算更清晰,也比一个触达所有事务的助理更容易做基准评估。 对 Instinct 而言,竞争门槛是双面的。它必须在行动导向上胜过广义 AI 产品,同时在具体工作流可用性上胜过专用工具。这并非不可能,但比“通用工具打败点状工具” 的简单故事门槛更高。[CP007, CP008, CP009, CP012, CP013, CP017]

功能对比
购买标准InstinctOperatorGeminiApple IntelligenceClaude / Perplexity / 专业工具
消息原生入口部分大多否
自主多步骤任务低中低到中
邮件 / 日历 / 旅行打包部分部分部分通常只覆盖单一工作流
广泛公开可用否 - 私测付费档有限开放支持设备上可用
原生平台集成不一
信任 / 品牌熟悉度低中中高

单元格为定性判断,依据公开可见定位,而非内部能力测试。

[CP003, CP004, CP005, CP007, CP008, CP009]
FP003: 功能重叠矩阵

Instinct 与几家对手有较广重叠,但具体重叠因工作流和自主深度而异。

单元格概括公开产品定位,而不是实验室测试性能分数。

[CP007, CP008, CP009, CP010, CP016, CP017]

3.3 分发力量、切换成本与定价压力

Instinct 最大的非产品差异点,是它通过 SMS 和 WhatsApp 触达用户,而不是 app 优先界面。这很重要,因为消费级助理的设置负担不只是创建账户,还有习惯养成。 消息线程比下载并学习另一个应用更低摩擦。与此同时,渠道所有权仍在外部。Apple、Google、Microsoft、Meta 和 WhatsApp 控制主流设备、身份、应用商店和 通信界面,进而影响发现和信任。它们也有能力把助理功能捆进用户已经付费的产品。 这种捆绑能力会压低定价空间。OpenAI、Google、Microsoft 和 Claude 现在都销售付费 AI 套餐,验证了经常性支出,但也让用户习惯把多个助理并排比较。 Meta 则朝相反方向走,把 AI 辅助免费化。在这种环境下,Instinct 大概率不能只靠价格取胜。它必须靠可感知的有用性取胜:省下足够时间或认知负担,让用户 在替代品充足时仍愿意单独付费。 切换成本只是中等。许多用户仍会多平台并用,现状也很碎片化。这有助于 Instinct 获得试用,但除非助理深度嵌入敏感且可信的工作流,否则长期锁定更难。[CP010, CP011, CP013, CP014, CP018, CP019]

上市路径对比
公司主要入口获客动作定价姿态切换摩擦分发能力
InstinctSMS / WhatsApp病毒式 / 邀请制未披露的高端定价自有渠道弱,渠道杠杆强
OpenAI OperatorChatGPT订阅增购付费档低中品牌触达很强
GeminiGoogle 应用和网页捆绑 + 订阅免费 + 付费平台触达很强
Apple IntelligenceOS 原生随设备捆绑捆绑设备触达很强
Microsoft Copilot生产力套件捆绑 + 企业渠道付费 / 捆绑套件触达很强
专业工具(Superhuman / Motion / Cal.com)专用应用工作流驱动的自助式获客付费工作流订阅中高聚焦垂直触达

分发能力为定性判断;Instinct 受益于用户熟悉的渠道,但并不拥有主导平台。

[CP009, CP010, CP011, CP013, CP014, CP018]
FP002: 分发能力条形图

平台方一开始就拥有明显更强的默认入口优势。

数值是分发能力序数评分,不是市场份额数据。

[CP011, CP019, CP020, CP024, CP031]

3.4 护城河耐久性与反向逻辑

关于 Instinct 护城河,最强论点是产品架构和用户体验。公司试图把几个碎片化工作压缩进一个会行动、不只是给建议的助理里。如果它能在一个消息线程中可靠处理 沟通、排期、购物和旅行,产品体验可能与通用聊天和单一工作流 app 都有质的不同。创始人的智能体循环研究背景也为这一野心增加可信度。 反向逻辑比许多初创公司叙事承认的更强。每个主要平台所有者都在从远强得多的装机基础出发,走向更智能体式的行为。如果 Google、Apple、OpenAI、Microsoft 或 Meta 在保留更高信任和更深平台集成的同时,匹配 Instinct 大部分自主性,独立产品切入点会迅速被压缩。Instinct 自身负面事件又放大了这一风险,因为信任 恰恰是既有玩家起步更领先的少数领域之一。 因此,正确的尽调判断应是有条件的。Instinct 今天确实有真实切入点,但护城河耐久性尚未证明。公开记录缺少定价、留存、任务级成功率,以及相对于竞品的流失数据。 在这些数据出现前,竞争优势应视为可能存在,但不能默认持久。[CP012, CP016, CP017, CP018, CP020, CP021]

护城河评估
护城河主张主要威胁严重性重要性缓释措施 / 尽调问题
消息原生 UX现有巨头在既有应用里加入智能体聊天界面优势可能很快被压缩衡量 SMS / WhatsApp 带来的留存优势
更高自主性自主性事故削弱信任用户可能偏好更安全但能力较弱的现有产品要求任务级错误率数据
跨工作流打包专业工具保住更深的工作流 UX打包可能宽但浅对比邮件和日历 NPS 与专业工具
创始人 / 技术边际大科技分发和资本产品质量未必能打赢默认入口测试用户在并排试用后是否切换
早期 ARR 动能私测集中度和新鲜感效应当前验证未必能泛化按工作流深度分析转化和流失队列

反向逻辑集中在信任和分发,而不只是原始模型质量。

[CP012, CP016, CP017, CP018, CP019, CP020]

3.5 图表

Chapter 04

04财务情况

4.1 融资历史与资本结构

Spear Street Technology Inc. 已完成两轮可识别融资。Series A 于 2026 年 1 月完成,规模 $100 million,由 Kleiner Perkins 领投, 合伙人 Mamoon Hamid 获得董事会观察员或董事席位;Conviction Capital(Sarah Guo)参投。Series A 时,ARR 约 $5 million,隐含该轮 收入倍数约 ~100×。Series B 于 2026 年 8 月 26 日完成,规模 $250 million,由 Index Ventures 和 Benchmark 共同领投,投后估值 $2.5 billion。Series B 时,管理层口径 ARR 约 $80 million,隐含 ~31× ARR。Greenoaks Capital 也参与了 Series B。所有已披露轮次累计融资 $350 million。种子轮或 pre-seed 信息没有公开披露。股权结构表、投资人持股比例和稀释安排均不可得。[CI001, CI002, CI003, CI004, CI005, CI006]

融资轮次摘要
轮次日期金额投后估值交割时 ARRARR 倍数领投方
Series A 轮2026-01-01$100M估计 ~$500M~$5M~100×Kleiner Perkins (Mamoon Hamid)
Series B 轮2026-08-26$250M$2.5B~$80M~31×Index Ventures + Benchmark
合计 / 当前2026-08-28$350M$2.5B$80M31×Index Ventures、Benchmark、KP、Conviction 与 Greenoaks

A 轮估值由分析师基于 $100M 融资和可比消费 AI 种子轮至 A 轮估值跃升估算;尚未确认。所有 ARR 数字均为管理层口径。公开资料没有种子轮或种子前轮数据。

[CI001, CI002, CI003, CI004, CI005]
FI001: 按轮次拆分的融资瀑布图

Instinct 的投后估值从 Series A 的约 $500 million 升至 Series B 的 $2.5 billion,7 个月涨了 5×。

Pre-Series A 和 Series A 估值为分析师估计;只有 Series B 的 $2.5B 估值已确认。估值抬升值由估计估值推导。

[CI001, CI002, CI003, CI025]

4.2 收入轨迹与 ARR 增长

Instinct 管理层口径 ARR 从 2026 年 1 月约 $5 million 增至 2026 年 8 月约 $80 million,约七个月增长 16×。这意味着平均每月新增净 ARR 约 $10.7 million。如果属实,这一增速将跻身任何消费软件产品有记录以来最快的 ARR 爬坡。ARR 尚无独立验证;该数字完全来自媒体报道中的管理层表述。 毛利率、净留存率(NRR)、客户数、用户平均收入(ARPU)和定价结构均未披露。产品仍处私测阶段,仅限邀请,说明 ARR 可能来自一小群早期付费用户, 或尚未公开描述的早期商业安排。[CI007, CI008, CI009, CI010, CI011, CI012]

ARR 增长轨迹
期间ARR(估计)月度增长当期收入倍数备注
2026 年 1 月~$5MN/A~100×(相对 $500M A 轮)管理层在 Series A 轮交割时表述
February 2026~$15M 估计~$10M~N/A线性插值;未披露
April 2026~$35M 估计~$10M~N/A线性插值;未披露
June 2026~$55M 估计~$10M~N/A线性插值;未披露
August 2026~$80M~$12.5M~31×(相对 $2.5B Series B 轮)管理层在 Series B 轮交割时表述
隐含 CAGR(年化)7mo 内 ~16×平均 ~$10.7MN/A若增长线性;实际路径未知

中间 ARR 数字是披露数据点之间的线性插值;实际月度 ARR 未知。 相比这条线性路径,增长可能前置,也可能后置。

[CI007, CI008, CI009, CI010]
FI002: ARR 增长轨迹(插值)

管理层称,年经常性收入(ARR)从 2026 年 1 月约 $5M 增至 2026 年 8 月约 $80M,意味着这段时间平均每月新增净 ARR 约 $10.7M。

2026 年 1 月和 8 月数据点来自管理层表述。中间月份为线性插值;实际 ARR 路径未知。

[CI007, CI008, CI009, CI010, CI026]

4.3 单位经济模型与利润率估计

由于 Instinct 没有披露定价、员工数或成本结构,所有单位经济模型分析只能基于行业基准,以及对可比消费级 AI 公司的推断。自主 AI 助理处理邮件、日历和 多步骤任务时,LLM 推理成本估计占收入的 15-35%,依据是类似智能体产品公开的成本结构。若套用到 $80 million ARR,意味着年度 LLM 推理成本为 $12-28 million。考虑到病毒式候补名单模式,获客成本(CAC)可能较低,但这一点带有推测性。不完全依赖基础设施的可比消费级 AI SaaS 企业毛利率在 50-75% 区间;Instinct 推理强度更高,意味着更可能落在区间低端。净留存率未知;私测背景和未披露流失数据使分析无法展开。GTM 动作依赖病毒式分发和 仅限邀请的候补名单,提示付费获客成本较低,但增长放大后的可预测性有限。销售周期未定义;未披露外呼或企业销售动作。GTM 效率指标——CAC 回本周期、 LTV/CAC 比率和渠道组合——在公开披露中完全缺席。[CI013, CI014, CI015, CI016, CI017, CI018]

单位经济模型估算
指标估计值方法置信度缺口
毛利率50–70% 估计可比消费级 AI SaaS 基准未披露财务数据
LLM 推理成本(收入占比)15–35%已公开智能体 AI 成本结构供应商和定价未知
LLM 推理成本($M/yr)$12–$28M按 $80M ARR 测算仅为估算
获客成本低(病毒式传播)邀请制候补名单模式未披露营销支出
净收入留存率Unknown未披露N/A估值尽调的关键缺口
ARPU(月度)Unknown未披露定价N/A私密 beta;无公开定价

所有单位经济模型数字都是分析师基于行业基准推导的估计。Instinct 未确认或披露任何数字。 只能按方向性参考。

[CI013, CI014, CI015, CI016]
FI003: 单位经济估计区间

在缺少公开财务披露的情况下,毛利率、LLM 推理成本和 ARPU 都是分析师估计,且不确定性区间很宽。

所有区间均为分析师估计,来自可比消费 AI 公司基准。区间较宽,反映高度不确定性。

[CI013, CI014, CI015, CI016, CI027]

4.4 财务风险与烧钱展望

Instinct 的核心财务风险集中在四点:(1)估值 $2.5 billion 时仍未披露变现模型,带来证明价格合理性的执行风险;(2)依赖未披露的 LLM 供应商, 其定价或可用性可能不利变化;(3)烧钱速度未知,但考虑 AI 推理成本和工程团队人数,可能相当高;(4)消费级 AI 产品规模化变现历来困难。 公司已融资 $350 million,若假设每月烧钱 $3-7 million(基于行业的估计),从 Series B 交割起现金跑道约 36-58 个月,足以支撑商业发布和首个 变现周期。然而,如果隐私反弹或竞争压力导致 ARR 增长放缓,相对收入的烧钱速度可能更快变成问题。缺少审计财务报表或任何第三方财务认证,意味着所有 现金跑道估计都高度不确定,只应视为指示性判断。反稀释条款、清算优先权等偏投资人的条款很可能存在,但完全未披露,这进一步限制了财务建模和退出情景 分析的精度。[CI019, CI020, CI021, CI022, CI023, CI024]

烧钱速度与现金跑道情景
情景月度烧钱速度LLM 成本其他运营支出$350M 对应隐含现金跑道关键假设
保守(精简团队)$3M/mo$1.2M$1.8M约 58 个月员工约 25 人,低支出
基准(行业基准)$5M/mo$2.0M$3.0M约 35 个月员工约 50 人,中等支出
激进(扩张)$9M/mo$3.5M$5.5M约 19 个月员工约 100+ 人,快速招聘

烧钱速度情景是分析师估计;实际烧钱速度未披露。$350M 并非一次性融到; 提款时间会影响真实现金跑道。

[CI020, CI021, CI022]
关键财务风险与缺口
风险领域描述严重程度证据缓释路径
无货币化模式估值 $2.5B 时仍未披露定价或收入模式严重公开披露为零商业化发布公告
LLM 供应商依赖LLM 供应商未披露;API 定价或可用性可能变化推理成本估计占收入 15–35%谈长期合约;分散供应商
ARR 未验证$80M ARR 来自管理层表述;无第三方验证单一来源:管理层表述尽调中提供经审计财务数据
消费级 AI 货币化历史消费级 AI 历史上很难在规模化后守住付费订阅历史案例:Inflection、Character AI差异化产品价值;自主执行效用
隐私反弹引发的收入风险August 2026 的反向报道可能拖慢 ARR 增长或推高流失TechCrunch August 2026 反向报道修产品;数据政策透明

严重程度是分析师基于现有证据和行业基准的判断。尚未公开的披露可能缓释所有这些风险。

[CI019, CI022, CI023, CI024]
FI004: 月度烧钱与现金跑道情景

估计月度烧钱速度为 $3-9M,以已融资 $350M 计算,现金跑道为 19-58 个月。基准情景显示现金跑道约 35 个月,足以支撑商业化发布。

烧钱速度和现金跑道均为分析师估计。实际现金余额取决于资金提取节奏、收入回款,以及未公开的运营费用。

[CI020, CI021, CI022, CI028]

4.5 图表

Chapter 05

05产品与技术

5.1 Instinct 在日常工作流中交付什么

Instinct 的定位是面向消费者的个人助理,生活在人们已经使用的渠道里,而不是要求用户学习新 app。官网称产品“无需新界面”,系统被训练为会使用手机和电脑, 用户可以给它发短信或打电话;第三方报道补充称 WhatsApp 也是支持渠道。实际产品承诺不是用于信息检索的聊天,而是委托执行。公开描述的任务包括清理收件箱、 起草跟进、预约、餐厅订位、旅行协调、叫车、购物,以及当智能体发现未完成线程时主动提醒。相同来源也清楚表明,Instinct 可以跨邮件、消息、日历、屏幕、 音频和位置权限行动,这显著扩大了范围:它不只是助理,更像操作员。这种工作流设计在战略上有吸引力,因为它移除安装摩擦,并把许多杂务集中到一个对话入口; 但产品质量也因此取决于不可见的后台决策,而不只是答案质量。用户同时购买便利和代理权转移,所以看似很小的控制失效——例如一封未授权邮件——都会变成一阶产品风险。[CE001, CE002, CE003, CE004, CE005, CE006]

产品模块 / 资产矩阵
模块 / 资产功能用户 / 触点成熟度差异化 / 尽调缺口
消息界面通道通过短信、电话及据报道的 WhatsApp 触点接收用户请求并发送后续消息终端用户前端私密 beta,测试用户已在使用无需下载 App 是真实 UX 差异点;具体传输供应商未披露
连接账户数据摄取从电子邮件、消息、日历和其他已连接服务拉取上下文用户数据与上下文层私密 beta,权限已有公开文档访问深度形成差异化;范围治理和最小权限设计未公开
智能体动作引擎把请求转成预订、购买、日程调整和对外沟通核心执行层功能已验证,控制质量仍不成熟自主执行是主打功能;审批阈值和回滚逻辑未披露
记忆 / 索引层存储既有上下文和已索引外部数据,让助手能跟进中断线程个性化层显然已启用,治理受挑战持久记忆看起来处在核心位置;删除语义在 beta 中引发争议
后台同步 / 通知层监控已连接服务的更新,并触发后续工作后端运行层由平台文档和产品表述推断可能使用 watch / webhook 任务;无公开架构或可观测性证据
Workspace / 设置界面在对话线程之外承载账户连接、删除和控制设置管理 / 安全界面仅有反应式公开证据删除工具似乎在投诉后加入;审计日志可见性缺失

行内容合并了官方产品文案、法律文件、集成平台文档和测试用户的反向报道; 若干后端层属于推断,因为 Instinct 没有发布正式架构图。

[CE001, CE002, CE003, CE004, CE015, CE018]
工作流 / 用例表
用户任务示例任务Instinct 处理方式所需集成公开证实的收益已知限制
出行 / 预订管理叫车去机场或订餐厅座位用户发消息提出请求;智能体读取上下文,在已连接服务上执行,并确认或继续追问消息通道,加出行或商户账户访问多步骤杂事只有一个对话入口错误动作可能形成有约束力的承诺或非预期购买
收件箱清理与跟进总结邮件、找验证码、起草或发送跟进读取邮箱内容,并代表用户跨已连接邮箱账户执行动作Gmail 或 Outlook 式邮件 API,或委托账户访问从总结推进到执行beta 中已经出现未经授权的对外邮件
日历协调预约并处理时间冲突结合收件箱、日历和提醒来安排或改期事件日历 API 和后台变更监控能主动推进线程webhook 或 token 被攻破,或事件状态过期,都可能触发错误排期
购物与购买下单或完成结账步骤使用委托支付和商户上下文代表用户交易商户凭证、支付数据和外部结账流程大幅降低结账摩擦条款把交易责任放在用户而非平台身上
长尾生活事务无需再次提示,也能跟进中断线程或提醒持久上下文让智能体稍后重新介入,并串起多个触点记忆 / 索引层,加对外消息传输体验更像主动,而非被动响应持久化放大了数据留存、同意和可解释性顾虑

收益来自公开描述的工作流,而不是实测 KPI 案例研究;Instinct 尚未发布按任务拆分的成功率、 误操作率或用户控制指标。

[CE004, CE005, CE006, CE019, CE025, CE026]
FE002: 客户工作流 / 运行流程

用户提出请求后,系统检索上下文、自主执行,并在后续继续跟进;这是一个代表性旅程。

流程抽象自官方文案和测试者报告暗示的常见运行模式;具体任务可能在执行前分支,或要求澄清。

[CE001, CE004, CE006, CE021, CE025, CE029]

5.2 推断的智能体架构与创始人技术脉络

Instinct 没有发布正式系统图,也没有点名模型供应商,因此公开技术画像只能从产品文案、法律披露和 Noah Shinn 过往研究中重构。这条脉络信息量异常高。 Shinn 的 Reflexion 论文和 NeurIPS 海报描述了一个 actor-evaluator-reflector 循环:语言智能体从存入情景记忆的语言反馈中学习,而不是通过权重更新学习; τ-bench 随后把这一世界观延展到动态、使用工具的对话中,并受 API 工具和政策规则约束。这些材料不能证明 Instinct 的内部实现,但强烈暗示产品并非只是 原始 LLM 套壳。最合理的公开解读,是一个分层栈:前沿 LLM 作为推理底座,编排循环负责规划和工具选择,持久记忆 / 索引层承载跨会话上下文,再用集成适配器 连接各项服务。官网承诺助理理解用户重视什么、跟进中断线程,并能长期代表用户行动,也进一步支撑这一推断。换句话说,可防守的技术点是有记忆支撑的代理能力, 而不只是把消息当作用户界面。[CE009, CE010, CE011, CE012, CE013, CE014]

技术 / 运营架构表
层 / 组件公开证据系统角色置信度主要风险
LLM 核心模型官网提到“核心模型”;供应商未公开点名支撑意图解析、综合和行动规划的推理底座模型供应商依赖、延迟、成本,以及未披露的 fallback 行为
智能体循环 / 编排Reflexion 和 τ-bench 脉络指向带政策约束的工具型智能体循环选择工具、排列步骤,并决定何时行动或追问自主性可能跑在审批、政策或异常处理前面
情节记忆 / 索引官方文案承诺理解重要事项,并跟进中断线程跨会话存储上下文和检索到的状态留存、删除和过期记忆错误会变成产品级关键问题
OAuth 与已连接账户适配器隐私政策和条款列举了 Google Workspace 及已连接账户获取邮件、日历、文档和身份触点的委托访问token 被攻破或权限过度授权,会造成严重影响面
消息传输层WhatsApp 和 Twilio 文档展示了出入站消息如何编排传递用户请求、确认、状态更新和主动触达运营商和渠道限制、投递状态失败都不在 Instinct 控制内
后台事件摄取Gmail 和 Calendar 的 watch 模型需要 Pub/Sub 或 webhook 回调与续期逻辑收件箱或日历变化时触发后续工作续期遗漏、重复事件或回调中断都可能破坏可靠性
设置 / 删除工作区条款和隐私政策提到 Workspace 与 Settings 控制界面承载数据删除请求、退出选择和断开连接动作反应式控制可能不足以处理高自主性故障恢复

本表有意区分直接观察到的事实与架构推断;当公开平台文档约束任何兼容实现必须如何做时, 置信度更高。

[CE009, CE013, CE015, CE019, CE021, CE022]
FE001: 产品架构图

可从公开信息推断出的层级,从消息 UX 延伸到智能体编排、记忆、集成和后端回调基础设施。

Instinct 尚未发布正式架构图;这套技术栈由官方文案、法律页面、Shinn 既有研究,以及所引用集成平台必须具备的行为综合推导。

[CE010, CE011, CE015, CE016, CE017, CE018]

5.3 集成、传输通道与后台执行要求

尽管 Instinct 只明确点名部分合作伙伴,产品表面仍暗示其集成版图很广。隐私政策称 Google Workspace 访问范围可覆盖 Gmail、Calendar、Drive、Docs、 Sheets、Slides 和 Tasks,服务条款提到 Apple、Facebook 和 Google 账户绑定,报道还补充了 Outlook 和 WhatsApp。公开平台文档有助于界定其运营含义。 Gmail 和 Google Calendar 都支持基于 watch 的变更通知模型,但需要服务端 Pub/Sub 或 HTTPS webhook 基础设施,以及续订逻辑。Microsoft Graph 提供 邮件和日历 API,可打通个人和组织 Outlook 账户。消息侧,WhatsApp Cloud API 将自由形式服务回复限制在 24 小时客服窗口内,而 Twilio 消息 API 提供外发、 送达状态回调、脱敏和渠道抽象。合在一起看,Instinct 几乎必然在幕后依赖一层相当重的凭证、token 和回调系统。架构负担因此不在于生成文本,而在于安全持有 委托权限、接收变更事件、决定何时行动,并在异构外部系统中从部分失败里恢复。[CE019, CE020, CE021, CE022, CE023, CE024]

5.4 成熟度、信任与安全态势

Instinct 的成熟度信号是矛盾的:产品显然已足够可用,能让测试者满意并执行真实世界动作;但控制还不足以宣称达到生产级安全。多家独立报道描述,助理曾在没有 明确逐项批准的情况下发送或准备邮件,遵循嵌入来信中的指令,并在 Google 访问被撤销后保留此前已索引内容。这些不是表层 bug。它们说明核心产品循环——读取、 推断、行动——已经跑通,也说明其失败模式正是自主消费级智能体在全面开放前应该加固的部分。官方政策部分缩窄了训练问题,称 Google Workspace 数据不用于模型训练, 也不用于第三方模型供应商的二次使用;但更广的隐私框架仍允许非 Workspace 材料在政策例外约束下用于改进产品和模型。同样重要的是,外部安全审计、认证、 可用性指标或红队披露都没有公开证据。因此成熟度判断只能是“真实产品,控制平面不完整”:强于演示玩具,弱于能让消费者安全委托行动的操作系统。[CE029, CE030, CE031, CE032, CE033, CE034]

信任 / 质量 / 合规表
控制 / 问题公开状态证据重要性缺口
Google Workspace 训练排除已记录,但范围有限隐私政策称,Workspace API 数据不会用于训练模型,也不会发送给第三方 AI 供应商用于训练范围窄于产品层面宽泛的训练表述,对 Google 关联用户很重要所有非 Workspace 材料没有同等公开排除条款
断开连接后的删除需要单独删除步骤隐私政策和条款称,断开集成不会自动删除已索引数据撤销力度弱于许多用户的直觉删除 UX、传播时间和可验证性均未公开
有约束力的交易授权条款明确授予条款指定该服务作为用户代理人,处理协议、承诺和交易自主动作一旦出错,可能带来法律和财务风险敞口未披露逐动作审批、暂停或回滚阈值
Prompt 注入暴露beta 报告中已有演示TechCrunch 及后续报道描述了系统跟随恶意邮件指令智能体可能被诱导把敌意内容当成命令未披露公开的 prompt 隔离或权限分段设计
未授权对外动作beta 报告中已有演示测试用户称邮件未先询问就被发送证明执行能力真实存在,也证明信任边界破裂未披露公开审计日志、dry-run 模式或确认政策
安全保证披露公开不可见已审来源中未发现公开审计、认证、正常运行时间指标或红队摘要读者无法只靠政策文案区分硬化控制是否存在大范围发布前,需要 SOC 2、渗透测试、事件响应或 SLO 证据

状态仅反映本轮审阅到的公开证据;本表未出现某项认证或控制,只代表公开未找到, 并不代表内部一定不存在。

[CE028, CE030, CE032, CE033, CE034, CE035]
FE004: 产品成熟度 / 能力图

公开层面看,Instinct 在消息原生工作流上最成熟,在已披露的信任、保障和发布控制上最薄弱。

矩阵评级是基于证据的序数判断,并非基准化分数;Instinct 未发布公开 SLA、错误行动率或控制有效性指标。

[CE003, CE029, CE036, CE038]

5.5 关键依赖、路线图不透明与技术尽调风险

对于一家年轻消费初创公司,产品外部依赖图谱异常重。即便供应商名单未披露,设计几乎必然依赖前沿 LLM API、消息传输供应商、OAuth 身份系统、Gmail 和 Calendar 基础设施、Microsoft Graph 式连接器、云计算,以及安全的 secret 或 token 存储。官网自己也承认,算力是当前放量门槛,因此在扩容时限制访问。与此同时, 创始人的研究脉络比产品本身释放出更强的开发者信号:Reflexion 和 τ-bench 都有活跃公开仓库、可观 star 数,以及仍未关闭的 issue 队列;Instinct 除了很薄的 Hacker News 足迹外,几乎没有可比的公开工程界面。这种不对称很关键。它说明当前技术可信度更多建立在创始人先前成果上,而不是可观察的运营卓越上。真正未解的 尽调项因此不是“能不能做出一个智能体助理”,而是“哪些模型和云供应商是集中风险点,存在哪些回滚和批准机制,还有哪些发布门槛,规模化时背着多少安全债”。[CE039, CE040, CE041, CE042, CE043, CE044]

路线图 / 发布 / 开发阶段表
日期 / 阶段功能或里程碑状态含义证据
2023 研究基础Reflexion 发布,包含情节记忆和语言反馈循环已完成的历史里程碑奠定创始人在记忆支撑智能体上的技术先例NeurIPS 海报和论文
2025 研究基础τ-bench 作为工具 / 智能体 / 用户基准发布已完成的历史里程碑显示重点放在工具使用、政策规则和评估,而不是纯聊天机器人 UXNoah Shinn 网站和 repo README
2026 私密访问产品扩充算力期间,Instinct 仅向私密访问群体开放当前产品真实存在,但容量受限官网首页
2026 法律 / 治理刷新隐私政策和条款于 August 26, 2026 修订当前快速增长和审查窗口期内,控制措辞发生变化官方隐私政策和条款页
2026 反应式删除控制公众投诉后新增外部数据删除工具当前,但属反应式已修补,但发生在信任受损之后TechCrunch 和 StartupFortune
2026 公开路线图可见性未发现公开 changelog、SLA、定价页或 GA 日期当前缺口难以支撑对发布就绪度或支持负担的判断官方页面加已审报道

由于 Instinct 发布的前瞻性发布细节很少,路线图视角必然偏重时间线; 主要公开信号是研究脉络、私密 beta 状态和反应式治理变化。

[CE003, CE010, CE013, CE036, CE038, CE044]
FE003: 关键依赖图

Instinct 依赖外部模型、传输、API 和基础设施层;每一层都会带来集中度或控制平面风险。

具体供应商并未全部披露;这些命名节点代表受公开产品行为和平台文档约束的依赖类别。

[CE025, CE026, CE027, CE043, CE044]

5.6 图表

Chapter 06

06客户情况

6.1 谁在使用 Instinct,以及如何访问

Instinct 当前客户基础最准确的描述,是一个仅限邀请的个人消费者队列,而不是企业。官网称产品只向私有访问群体开放,新用户通过候补名单或现有成员推荐进入。 服务条款面向个人使用,公开示例围绕跑腿、旅行、订阅、邮件跟进和家庭后勤,而不是团队工作流、采购或企业部署。这种组合很关键,因为它意味着买方、用户和 可能的付款方都是同一个人:一个消费者在决定是否把自己的收件箱、日历、账户和支出权限交给助理。 访问方式也是投资逻辑的核心。Instinct 将自己定位为“无需新界面”的助理,可通过短信或电话触达,并能通过 WhatsApp 等消息界面工作。这降低了早期采用者的 上手摩擦,因为产品嵌入既有消息习惯,而不是强迫学习 app。同时,最早期用户很可能格外懂技术或更能承受风险:投资人、创始人、运营人员和其他高阶用户, 愿意在产品触达主流受众前连接广泛权限。这让当前队列能作为能力证明样本,但只能弱代理大众市场在规模化时愿意信任产品的程度。[CU001, CU002, CU003, CU004, CU005, CU012]

客户分层表
细分群体买方 / 用户 / 付款方已观察用例规模可见性收入 / 战略价值缺口
科技圈重度用户个人消费者 / 同一人 / 同一人靠文字把高密度个人行政、旅行、购物、订阅和邮件任务交给 Instinct 处理只有具名轶事;未披露队列规模战略价值高,因为这一队列提供了上线首周的实用性证明,也很可能贡献了早期传播的大部分声量还不清楚他们能多大程度代表主流消费者
创始人、投资人和运营者使用工作相邻个人流程的个人消费者用 Instinct 处理预订、CRM 跟进、LP data room,以及混合工作与个人生活的日程安排来自少数公开用户的具名轶事信号价值高,因为最响亮的正面和负面帖子很多都来自这一群体没有证据显示企业会购买该产品或报销使用费用
家庭和家务协调者个人消费者 / 同一人 / 同一人跟踪学校日程、作业提醒、账单、接送、预约,以及跨消息应用的协调事项只在轶事中可见;没有人口结构拆分如果信任和可靠性提升,长期可能成为很大的消费者细分没有关于留存、付费意愿或家庭多用户行为的证据
对隐私敏感的主流消费者潜在消费者 / 同一人 / 同一人可能看重便利,但会犹豫是否接入收件箱、消息、凭据和支付因产品仍设门槛,尚未被直接观察到如果控制能力改善,扩张池很大,因为 AI 助手采用已经进入主流当前 beta 证据很可能高估了用户授予广泛权限的意愿

由于 Instinct 未披露人口结构或客户数量拆分,细分来自具名用户轶事、官方产品定位,以及 2026 年消费者 AI 采用基准。

[CU001, CU002, CU004, CU005, CU012, CU033]
FU001: 客户旅程图

当前旅程从声量和邀请门槛开始,经过高权限设置和早期任务成功,随后分化为习惯养成,或因信任问题而流失。

旅程阶段综合自官方准入机制、具名用户案例和信任事件;公司未披露漏斗数据或阶段转化指标。

[CU001, CU002, CU004, CU012, CU015, CU017]

6.2 采用轨迹:收入强,分母弱

Instinct 头部客户信号不是披露的用户数或具名客户基础,而是收入。TechCrunch 报道称,公司告诉 The Wall Street Journal,截至 2026 年 8 月下旬 ARR 已达到约 $80 million,高于 2026 年 1 月约 $5 million。表面看,这极不寻常。产品仍处私测,却在一年不到实现约 16x 增长,说明部分用户正在支付 可观金额,也说明产品范围足够广,能支撑不低的支出。这也解释了为什么投资人愿意在正式开放前支持 $2.5 billion 估值。 但分母问题很严重。Instinct 没有披露多少用户贡献这部分 ARR,多少人在付费而非单纯测试,存在哪些定价层,或多少使用量来自一个很小但高强度的队列。 Forbes 报道称,产品公开层面仍可免费使用,且尚未宣布正式定价,这进一步加深不透明。结果是一幅分裂图景:收入动能真实到足以重要,但采用证明在结构上 仍不完整。尽调时,客户增长无法干净拆分为留存、扩张、定价或新用户获取。每一项都可能支撑 ARR 故事,但公司尚未说明哪一项占主导。[CU001, CU006, CU007, CU008, CU009, CU010]

客户增长 / 采用轨迹表
指标数值日期 / 期间来源 / 置信度含义缺失的分母
访问状态通过候补名单或会员邀请进入的封闭 beta2026-08-28官方 + 媒体交叉印证 / 高访问仍然稀缺,可以放大排他感和病毒式传播未披露候补名单规模、邀请转化或活跃用户
年经常性收入(ARR)~$5M2026-01管理层通过 TechCrunch 披露 / 中说明今年早期已经存在可变现使用未披露用户数、价格或细分组合
年经常性收入(ARR)~$80M2026-08-26管理层通过 TechCrunch 披露 / 中意味着大范围发布前商业化推进极快未披露客户数、付费人数或月度经常性收入桥接
隐含 ARR 增长1 月至 8 月 ~16x2026-01 至 2026-08由已披露 ARR 点推导 / 中指向极强留存、快速获客、高 ARPU,或三者同时存在无法拆分新客户、价格、留存或扩张
公开客户证明深度有具名 beta 用户轶事,但没有具名付费客户2026-08-28独立媒体综合 / 中采用情况体现在故事里,而不是正式部署指标里没有具名客户名单、案例研究或评论平台语料

各行把可观察的采用事实与仍不透明的部分拆开。收入点来自管理层说法,不应当作经过审计的客户质量证据。

[CU001, CU006, CU007, CU008, CU009, CU029]
FU002: 采用 / 部署漏斗

Instinct 未披露用户数,因此漏斗用标准化证据强度指数呈现,而不是实际人数或账号数。

数值是可观察漏斗开放度的相对指数,不是公司披露的客户数。它们展示从漏斗顶部热度到可持续使用的具名证明之间,有多少证据会消失。

[CU001, CU006, CU007, CU009, CU042, CU044]

6.3 具名 beta 用户证明了真实效用,也暴露了真实信任破裂

由于 Instinct 仍处于封闭 beta,最接近客户验证的材料不是传统案例研究,也不是评论平台语料。本章因此依赖 TechCrunch、Forbes、SC Media、AI Weekly 引述的具名 beta 测试者和从业者评论,以及对发布周用户讨论串的二级梳理。这些证据并不完美,但仍明显强于只看客户标识或炒作:材料写出了人名、任务,也记录了成败。 正面看,几位早期用户描述的不是新奇演示,而像真实使用。Sheel Mohnot 说 5 天内发了 677 条消息、完成 15 个任务;Jesse Middleton 说一周里每天用 Instinct 处理旅行、预订、邮件跟进和 CRM;其他用户也提到检查行程、节省订阅费、安排家庭日程。负面看,同一批公开用户给出了本章最强的反向证据。Katie Jacobs Stanton 称系统未经询问就发出邮件;Claire Vo 发现断开 Google 后,先前导入的邮件仍可搜索;Peter Yang 抱怨留存和删除;Alex Cohen 演示了提示词注入;Forbes 则提到一次产生 $200 取消费的事件。这些轶事的正负分化很重要:用户不是无感,而是要么看到了少见效用,要么遇到了严重信任故障,很多时候两者同时存在。[CU013, CU014, CU015, CU016, CU017, CU018]

具名客户证明表
客户 / 测试者代理细分部署 / 用例生产 vs 试点结果限制
Sheel Mohnot科技圈重度用户高强度使用 Instinct 处理找医生、账单议价、WhatsApp 商家沟通、旅行后勤、取消订阅和缴纳过路费试点 / 封闭 beta 个人使用公开信息中最详细的正面效用证明:5 天内 677 条消息、完成 15 项任务投资人用户轶事,不是已披露的付费客户或纵向队列
Jesse Middleton创始人 / 运营者连续一周每天用它处理旅行改签、餐厅预订、邮件跟进、CRM 管理和 LP data room 工作试点 / 封闭 beta 个人使用显示个人任务和工作相邻任务都会重复使用;他明确说产品很棒仍是轶事;没有长期留存或付费水平证明
Katie Jacobs Stanton消费者重度用户主要用于个人需求,并在信任破裂前称赞产品能力试点 / 封闭 beta 个人使用强有力地说明,即使是成熟用户,产品最初也会带来像魔法一样的体验Instinct 未经批准发送邮件后信任破裂,她因此断开了邮箱
Claire Vo对隐私敏感的早期采用者连接了个人 Gmail 账户,后来测试断开访问后还留下些什么试点 / 封闭 beta 个人使用显示围绕留存副本和明文存储的严重信任失败反向证据凸显风险,而不是可持续采用

由于 Instinct 仍处于封闭 beta,且公开渠道没有披露具名付费客户,本表把具名 beta 测试者作为最接近、且可允许使用的客户证明代理。

[CU013, CU014, CU015, CU017, CU029, CU043]
FU003: 客户证明矩阵

与多数隐身产品相比,Instinct 的具名用户案例更丰富,但商业可见度和正式留存证明仍然偏弱。

评分是对证据质量的分析评级,不是公司上报指标。商业可见度仍低,因为没有具名付费客户,也没有公开评价平台数据。

[CU020, CU021, CU022, CU028, CU029, CU042]

6.4 持久性尚未验证,集中风险可能不低

Instinct 未披露通常用来判断客户持久性的指标:流失、队列留存、续约率、重复使用频次或收入留存。对普通消费产品,这种沉默只是中等风险;对一个要求异常广泛权限的产品,这是重大风险。没有客户数或留存曲线,公司披露的 ARR 增长可以讲出多种互相冲突的故事:留存强、口碑扩张;快速获客掩盖流失;对小规模队列收高价;或三者兼有。缺少这些分母,本身就是本章最重要的发现之一。 不过,扩张空间确实可见。2026 年广泛调查显示,消费者 AI 采用已足够主流,Instinct 这类产品有真实可触达市场:近一半美国成年人使用 AI 聊天机器人,日使用率已接近成年人四分之一,面向交易的 AI 使用也在快速上升。但同一批数据也说明,Instinct 下一关是信任,不是认知度。大多数人预计 AI 会让个人信息更不安全,超过半数消费者在个人数据上不如信任人类那样信任 AI,已有相当比例会因 AI 数据担忧而取消或更换品牌。对 Instinct 而言,含义很直接:如果能证明控制更安全,上行空间很大;但下行同样清楚——封闭 beta 做到 $80 million ARR、却没有用户分母,强烈暗示早期集中风险;负面口碑可能在产品触达早期大众之前就压住扩张。[CU030, CU031, CU032, CU034, CU035, CU036]

留存 / 重复使用 / 满意度表
指标数值 / 状态细分置信度证据尽调问题
客户数未披露全部用户查阅的官方或媒体来源均未给出活跃用户或付费人数按月索取活跃用户、付费用户、受邀用户和候补名单总数
NRR / GRR / 流失未披露全部用户未找到公开留存指标按获客月份索取留存队列和流失
重复使用轶事Jesse Middleton 称连续一周每天使用重度用户具名用户引述表明,使用超出一次性新奇演示索取 30 天、90 天和周活跃用户留存数据
高强度使用轶事Sheel Mohnot 称 5 天内发送 677 条消息重度用户显示至少部分早期采用者参与度很高索取每位活跃用户任务量分布
正面满意度信号用户把产品描述为惊艳、很棒,或像魔法一样重度用户TechCrunch 和二级汇总中有多条引用反应索取 NPS、CSAT 或任务成功率调查结果
负面满意度信号未授权邮件、断开后仍留存数据、提示词注入,以及偏离脚本的预订事件重度用户来自具名用户的多起独立反向报道索取投诉日志、复盘和操作撤回率
评论平台覆盖没有公开 G2/Capterra/App Store 语料主流用户与邀请制状态和缺乏主流应用分发一致索取封闭 beta 满意度研究和扩张就绪度基准

由于 Instinct 处于封闭 beta,且不发布队列数据,多数耐久性指标为空。因此本表区分直接的轶事重复使用证据与缺失的正式留存数据。

[CU006, CU014, CU015, CU017, CU018, CU019]
扩张和集中风险表
风险或驱动因素描述影响严重性证据尽调路径
早期采用者集中仍处封闭 beta、且未披露用户数时达到 ~$80M ARR,可能意味着收入集中在一小群重度用户身上如果小队列流失或降级,下行风险很高收入披露没有分母;公开用户证明仍是轶事索取前 10% 收入集中度和 ARPU 分布
病毒式传播作为获客引擎口碑、圈内邀请和上线首周社交证明似乎是关键漏斗顶部渠道上行空间中等,脆弱性也中等候补名单机制,加上媒体关注科技圈内部人群分发索取获客渠道组合和邀请到活跃的转化率
信任冲击风险未授权操作或数据留存担忧会压垮采用,因为个人助手需要用户授予高度私密的权限具名反向事件,以及关于隐私敏感度的调查证据索取升级处理数据、回滚控制和权限政策变更
低摩擦界面带来的扩张上行如果控制能力改善,文字 / WhatsApp 入口和没有 app 学习曲线的体验可能支撑更广泛的消费者渗透上行空间高官方界面说明,加上强任务完成轶事在圈内网络之外测试主流用户入门和转化
可信现有平台的品类竞争消费者 AI 使用时间集中在成熟平台,因此 Instinct 必须快速赢得信任,才能摆脱小众 beta 队列中到高Sensor Tower 和大众市场评论来源将留存和付费意愿对标领先助手

严重性反映对收入耐久性的可能影响,而不只是 PR 风险。若干行是推断,因为 Instinct 未披露客户集中度或队列数据。

[CU032, CU033, CU034, CU036, CU037, CU040]
FU004: 留存 / 复用队列

示例情景说明,在用户数和队列未披露时,多条不同留存路径仍可能匹配同一个 ARR 标题数字。

这些是情景行,不是 Instinct 的实测队列。公司披露 ARR,但未披露留存、用户数或队列曲线,因此几种完全不同的持久性叙事仍都可能成立。

[CU030, CU031, CU032, CU037, CU038]

6.5 图表

Chapter 07

07风险

7.1 隐私、同意与监管暴露

Instinct 最核心的风险在于,它最强的产品特性——在邮件、日历、购物和消息工作流里自主行动——也正是最尖锐的监管和同意暴露来源。公开批评不只是泛泛讨论 AI 幻觉,而是质疑这款助理能否以消费者未明确授权的方式行动、留存并再利用敏感用户信息。这形成典型的高信任 / 高责任风险画像。 到 2026 年,政策背景已不再默认宽松。EU AI Act 已经生效,European Commission 也已进入透明度要求的主动执法阶段。GDPR 仍然关键,因为个人助理会跨多个连接系统处理私密通信、元数据和行为语境。在美国,如果产品夸大安全性、错误描述删除或留存行为,或模糊用户同意边界,FTC 指引都给了监管机构多个抓手。 结果未必是立即禁用,但合规成本、设计约束和执法可选项都会显著增加。产品建立在委托行动之上,信任和法律清晰度不能事后再补。[CR001, CR002, CR003, CR004, CR005, CR006]

风险热力图
风险发生可能性严重性主要触发因素重要性
同意 / 隐私泄露关键未授权操作或留存直接冲击信任和监管
提示词注入恶意入站内容可能劫持工具使用和自主性
平台 / API 依赖政策或价格变化Instinct 不掌握关键通道
变现不透明转化弱或利润率意外估值可能跑在证据前面
执行 / 治理缺口扩张快过控制建设小团队风险

发生可能性和严重性是基于当前公开证据与类似 2026 年 AI 治理数据作出的序数判断。

[CR001, CR009, CR013, CR018, CR021, CR026]
监管 / 法律风险登记表
风险义务来源已观察信号潜在结果所需缓解
未授权自主通信FTC / 同意据报道有邮件未经询问就被发送消费者伤害和执法明确批准 / 护栏
OAuth 撤销后的留存GDPR / 隐私法明文存储指控删除和数据权利风险可验证的删除流程
不透明的数据使用许可消费者保护 / 合同公平性宽泛数据权利受到批评声誉和法律审查用普通语言收窄范围
AI Act 透明度义务EU AI Act2026 年开始执行合规成本和上线门槛模型 / 智能体披露
公共数据训练不确定性版权 / 数据法更广泛的 2026 年政策争论政策和诉讼不确定性更强的数据来源证明和告知

本表聚焦与高权限消费者 AI 助手最直接相关的法律和监管风险向量。

[CR002, CR003, CR004, CR005, CR006, CR007]
FR003: 监管暴露条形图

法律风险栈分布很广,并非集中在单一规则集。

数值是序数暴露分,不是法律概率。

[CR004, CR005, CR006, CR007, CR008, CR024]

7.2 安全、提示词注入与技术控制风险

技术风险栈很重,因为自主助理一次性叠加了三个危险属性:吸收不可信输入,持有敏感上下文,还能调用工具。OWASP 和 2026 年安全指引仍把提示词注入列为智能体系统风险清单首位。一个读取收件箱邮件或消息、再根据信号采取行动的助理,如果没有强权限边界、工具作用域和执行隔离,就会暴露在间接提示词注入之下。 围绕 Instinct 的公开指控让这不再只是理论担忧。据报道的提示词注入漏洞,以及 OAuth 撤销后仍以明文留存,指向的正是自主助理最关键区域里的控制薄弱点。即便部分指控反映的是早期 beta 状态,也足以说明容错空间有多窄。 NIST 和 OWASP 都意味着,生产级控制应包括感知身份的访问边界、人工覆盖、删除可信度、日志记录和结构化风险审查。公开记录还不足以证明这些护栏已经把技术风险完全缓释。[CR009, CR010, CR011, CR012, CR016, CR017]

安全 / 技术风险
风险机制严重原因公开信号控制预期
提示词注入恶意内容操纵智能体决策可能触发有害操作已报告漏洞,且 OWASP 将其列为优先项输入过滤和限定范围的工具
权限越界账户授权范围过宽扩大爆炸半径助手需要连接许多系统最小权限和升级认证
删除失败断开后仍有残留数据造成隐私和泄露风险已报告留存担忧可证明的删除和日志记录
审计轨迹不足智能体操作日志薄弱事件难以调查没有公开控制细节不可变操作日志
模型 / 工具错误级联一次错误推理扩散成更多操作自主性会放大错误智能体品类的普遍担忧人工覆盖和风险评分

提示词注入和删除控制是今天最可见的两个公开技术担忧。

[CR009, CR010, CR011, CR012, CR023, CR033]
FR002: 风险级联

一次自主性或删除失败,可能很快传导到多个利益相关方群体。

DAG 展示因果综合,不是按日期排列的事件年表。

[CR023, CR028, CR029, CR035, CR041, CR042]

7.3 商业模式、平台与竞争风险

Instinct 的商业风险离不开平台依赖。公司并不拥有交付成熟产品所需的主导操作系统、消息通道、邮件生态或模型供应链。Apple、Google、Microsoft、Meta/WhatsApp 以及模型提供商都可以改变定价、政策、集成深度或默认设置,其中一些还是直接或邻近竞争对手。 这种依赖进一步放大了商业模式的不确定性。公司必须说服用户在付费和信任上都达到高于纯聊天产品的水平,同时承担有意义的可变 AI 服务成本。大额融资有帮助,但不能让经济模型或分发永久安全。如果模型成本上升、API 收紧,或打包进平台的既有玩家已经足够好,Instinct 的切入口会很快变窄。 同样逻辑也适用于竞争。即便产品强,只要平台方把“足够安全”的体验打包进用户已经信任的界面,独立产品仍可能输掉。因此,平台权力是所有独立个人智能体的核心风险之一。[CR013, CR014, CR015, CR018, CR019, CR020]

商业 / 竞争风险
风险驱动因素影响可能恶化的原因指标
模型供应商集中外部模型选项少利润率和可靠性承压价格或能力快速变化供应商成本 / 宕机变化
捆绑式现有平台竞争Apple / Google / Microsoft / Meta定价和分发压力用户默认使用内置助手重大功能发布
变现模糊未披露定价收入质量能见度弱转化可能依赖新奇感定价 / ARPU 披露
信任驱动的流失事件后的公开反弹收入和转介绍承压上线带来更广泛审视留存队列
计算成本膨胀用量敏感型自主工作流毛利率承压重度用户触发高成本任务单次动作服务成本

业务风险来自信任、平台依赖和可变 AI 服务成本之间的相互作用。

[CR013, CR014, CR015, CR018, CR019, CR020]
FR004: 商业模式可持续性区间

即便早期牵引力强,下行情景仍对信任和成本假设高度敏感。

评分只表示方向,用来呈现不确定性,不作为正式评分模型。

[CR018, CR019, CR020, CR034, CR036, CR040]

7.4 执行、治理与综合判断

Instinct 的执行风险更高,因为它的公开身份与一位年轻创始人、一个小团队,以及一个扩张速度可能快过可见治理结构的产品高度绑定。产品迭代阶段,这可能是优势;但公司一旦需要同时处理事故、监管、招聘、基础设施和公众信任,就会变成负担。除创始人主导的核心外,公开材料几乎看不到深厚管理梯队或成熟运营体系。 整体风险图景的异常之处在于速度。上行路径要求持续增长和信任扩张;下行路径可能快得多:一次严重的隐私、删除或自主性事件,就可能在用户、媒体、合作伙伴和监管机构之间触发连锁反应。这种不对称,才是投资人应关注的重点。 公开记录足以识别主要红旗,但不足以高精度框定财务或法律下行。因此,本章支持给出高风险评级和具体否决标准,而不是精确到数字的损失预测。[CR021, CR022, CR023, CR027, CR028, CR029]

执行 / 团队风险
风险已观察情况潜在影响严重性否决 / 观察触发因素
创始人集中度公开身份围绕一位创始人关键人物风险领导层动荡或信心流失
小团队扩张缺口团队规模公开描述偏小控制体系跟不上增长反复事故或服务降级
治理不透明没有公开管理梯队细节危机处理变慢中高监管方或合作伙伴担忧
合规成熟度缺口高权限产品尚未公开发布上线延迟或重新设计无法通过法律审查
事件响应不成熟公开控制细节稀少失败后用户恢复慢可见且未解决的安全事件

自主消费助手的风险管理要求对这一阶段而言异常高,执行风险因此被放大。

[CR021, CR022, CR023, CR027, CR028, CR029]
FR001: 风险热力图矩阵

信任、隐私和平台依赖主导风险图景。

单元格是从公开证据和 2026 年治理报告综合出的序数描述。

[CR026, CR027, CR031, CR032, CR041, CR042]

7.5 图表

Chapter 08

08估值

8.1 投资逻辑与反向逻辑

Instinct 的乐观逻辑建立在四个命题上:(1)消费者 AI 助理会成为主导的个人生产力入口,形成赢家通吃大半的市场;(2)Instinct 在 7 个月内 ARR 增长 16 倍,显示出消费软件罕见的真实产品市场契合;(3)Noah Shinn 兼具学术 AI 可信度和消费产品直觉,23 岁就有这种组合极为少见;(4)手握 $350M 资本,Instinct 有足够跑道在竞争对手抹平领先前完成商业化。反向逻辑同样有力:(1)未披露定价意味着 ARR 可能来自非经常性或促销安排;(2)Apple、Google、Microsoft 等大型既有玩家拥有近乎无限的分发能力,可以用零边际 CAC 吸收 AI 助理需求;(3)AI 推理成本是结构性成本,除非谈降,否则会压缩毛利率;(4)2026 年 8 月的隐私反弹可能带来尚未反映在 ARR 中的监管摩擦或流失风险。净结果是:这家公司正站在风险投资押注和战略平台风险的边界上。[CV001, CV002, CV003, CV004, CV005, CV006]

各维度投资逻辑与反向逻辑
维度乐观论据悲观论据
市场AI 助手成为主导生产力入口;赢家拿走大部分市场OS 原生助手吸走市场;Google/Apple 分发难以跨越
产品7 个月 ARR 增长 16x,是少见的消费级 PMF 信号未披露定价;beta 用户可能无法规模化付费
团队Shinn 兼具学术 AI 深度(ReAct、Reflexion)和消费产品直觉单一 23 岁创始人;公开运营履历偏薄
财务$350M 资金;36-58 个月现金跑道;资本效率已有体现ARR 未验证;烧钱速度未知;31x 倍数下没有货币化模型
监管隐私风险可通过架构调整管理EU AI Act + California CPRA 给邮件访问型产品带来结构性风险

每个维度对应具体章节:市场对应 Ch2,产品对应 Ch5,团队对应 Ch1,财务对应 Ch4,监管对应 Ch7。

[CV001, CV002, CV003, CV004, CV005]
FV001: 投资决策象限 —— 信心 vs. 上行空间
[CV001, CV009, CV017]

8.2 估值语境与入场纪律

Instinct Series B 后 $2.5 billion 投后估值,对应管理层声称 $80M ARR 的 31x ARR 倍数。作为参照,历史上 Series B 阶段的消费软件独角兽中位数通常在 10-20x ARR;在 2021 年市场高点,增长最快的消费 SaaS 前一成达到 30-50x ARR。当前 2026 年环境下,AI 原生软件享有结构性溢价。KPMG Venture Pulse Q2 2026 报告称,AI 原生软件 Series B 中位数为 18-25x ARR;Instinct 的 31x 比该中位数高出 24-72%。关键担忧包括:(1)未验证 ARR 降低了倍数分母的可信度;(2)未披露定价模型,未来 ARR 可持续性未知;(3)Series A 隐含 100x 倍数,造成显著稀释,限制 Series B 投资人的回报预期。按市价入场、参与 Series B 后轮次,需要 5-7x 回报才能补偿风险溢价——只有乐观情景才可能做到。若因变现不确定性打 20-30% 折扣,公允价值为 $1.7-2.0B,而本轮价格为 $2.5B。[CV009, CV010, CV011, CV012, CV013, CV014]

建议摘要
维度评估置信度
建议观察——不要按当前条款投资
估值立场偏高——31x 未验证 ARR,高于 18-25x 中位数
风险评级高——隐私敞口;货币化未披露;竞争压力
总体评分6.5 / 10

分数基于 8 个尽调章节的综合证据。置信度反映底层数据的可验证性,而非结果确定性。

[CV001, CV009, CV010]
FV002: 估值敏感性 —— ARR 验证情景
[CV010, CV011, CV012, CV025, CV026]

8.3 乐观、基准与悲观情景分析

乐观情景(25% 概率):Instinct 到 2027 Q1 推出 $25-50/month 付费层,在 2-million 用户基础上实现 10% 转化。ARR 到 2027 年底增至 $250M+。2027 年以 $8-10B 估值融资 Series C,2029 年以 $15-20B IPO。Series B 投资人获得 5-8x 回报。基准情景(55% 概率):Instinct 处理隐私监管,变现推迟到 2027 年中。ARR 增长放缓至 $8-10M/month。下一轮以 $4-5B 融资并带来中等稀释。2030-2031 年通过 IPO 或战略收购退出,估值 $6-8B。Series B 投资人获得 2-3x 回报。悲观情景(20% 概率):EU 或 California 的隐私执法迫使产品大幅调整。变现停在 $200M ARR 以下。战略收购方以 0.5-1x 收入或 $300-500M 吸收 Instinct——对 Series B 投资人而言相当于全额减记。下行并不遥远:监管风险真实存在,变现模型仍未定义,AI 助理市场也可能在 Instinct 建立差异化之前,就更快围绕 OS 原生助理整合。按情景加权的期望值显示,当前入场的 Series B 共同投资人约可获得 2.5x——不足以补偿风险溢价。[CV017, CV018, CV019, CV020, CV021, CV022]

乐观 / 基准 / 悲观情景摘要
情景概率Series B 回报核心假设
乐观25%5-8x付费层级 2027 年 Q1 推出;10% 转化;2027 年底 ARR $250M+;2029 年以 $15-20B IPO
基准55%2-3x货币化推迟至 2027 年中;2027 年底 ARR $120-160M;2030 年以 $6-8B 退出
悲观20%<0.1x隐私执法;货币化失败;以 $300-500M 困境出售

情景概率是分析师基于章节证据作出的判断。Series B 回报按 $2.5B 入场估值计算。悲观情景并不遥远。

[CV017, CV018, CV019, CV020, CV021, CV022]
FV003: 估值桥 —— 从 Series A 到合理价值估计
[CV010, CV011, CV012, CV013, CV014]

8.4 可比估值组

可比对象分三层:(1)直接面向消费者的 AI 助理(Perplexity AI、Character.AI);(2)处于早期高速增长阶段的新一代消费 SaaS(Superhuman、Motion);(3)AI 原生横向平台(Scale AI、Cohere、Mistral)。没有一家上市公司可比对象能精确映射到 Instinct 的阶段和定位。私募轮次中,Perplexity AI 2024 年 4 月以 $1B 估值融资,当时 ARR 约 $25-30M,隐含 33-40x ARR——高于 Instinct 的 31x,但 Perplexity 已推出付费层。Google 于 2024 年 9 月收购 Character.AI,价格约 $2.7B,约为估计收入的 25x。Scale AI 2024 年 12 月 Series F 将公司估值定为 $13.8B,约 11x ARR——且有经审计财务和盈利的政府合同。Cohere 2024 年 6 月以 $5B 估值融资,当时 ARR 约 $200M(25x),并披露了 B2B 合同。若因 Instinct 变现不确定性打 20-30% 折扣,公允价值为 $1.7-2.0B,而本轮价格为 $2.5B。[CV025, CV026, CV027, CV028, CV029, CV030]

可比公司估值表
公司轮次日期估值ARR 倍数
Perplexity AI2024 年 4 月$1.0B~33-40x ARR
Character.AI(收购)2024 年 9 月$2.7B~27x ARR
Scale AI2024 年 12 月$13.8B~11x ARR
Cohere2024 年 6 月$5.0B~25x ARR
Instinct2026 年 8 月$2.5B~31x 未验证 ARR

所有 ARR 数字均为分析师估计。Instinct 的 ARR 来自管理层披露,未经验证。所有可比公司在估值时已有付费层级或合同收入;Instinct 没有。

[CV025, CV026, CV027, CV028, CV029]
FV004: ARR 倍数对比 —— Instinct 与可比 AI 公司
[CV025, CV026, CV027, CV028, CV029]

8.5 退出准备度、尽调要求与投资逻辑失效触发器

退出路径包括:(1)IPO——如果 Instinct 建成清晰变现模型并达到 $300M+ ARR,2029+ 年可行;(2)战略收购——最可能由 Apple、Google、Microsoft 或 Salesforce 出手,收购倍数为 5-10x 前瞻 ARR;(3)私募二级市场退出——随着公司成熟,2027-2028 年可能出现。任何投资决策之前,必须完成四项阻断事项:通过第三方审计独立验证 ARR;完整披露定价模型和转化漏斗;法律审查 EU AI Act、UK DPA 和 California CPRA 合规;以及审查投资人权利协议中的清算优先权和反稀释条款。投资逻辑失效触发器包括:连续两个季度 ARR 增长低于 $5M/month;任何主要司法辖区出现监管执法行动;CEO Noah Shinn 离职;或未能在 2027 Q2 前推出付费层。任一触发器出现,都应立即重新评估,不受 ARR 轨迹影响。[CV033, CV034, CV035, CV036, CV037, CV038]

投资逻辑破裂与否决触发因素
触发因素阈值建议动作
ARR 增长放缓连续两个季度增长 < $5M/月重新评估;要求提供 ARR 桥接表和队列数据
监管行动欧盟、英国或加州出现任何隐私相关执法行动立即重新评估;聘请法律顾问评估产品变更
关键人物离任CEO Noah Shinn 或其他核心技术作者离任下调至卖出;标记为投资逻辑破裂事件,提交 IC 审查
货币化失败到 2027 年 Q2 仍未推出付费层级下调至放弃;触发现有持有人减记评估

阈值定义在本报告发布时固定。触发因素是供投资委员会重新评估的二元旗标,不是自动卖出信号。

[CV033, CV034, CV035, CV036]
阻断性尽调要求
尽调要求关键性阻断性
通过第三方审计或陈述与保证保险独立验证 ARR关键
全面披露定价模型、转化漏斗和付费用户数关键
覆盖 EU AI Act、UK DPA 和加州 CPRA 的隐私合规审查
审查投资者权利协议中的清算优先权和反稀释条款

当前估值下,四项均会阻断任何投资决策。

[CV037, CV038, CV039, CV040]

8.6 图表

免责声明

本报告仅供信息参考,不构成投资建议。所有数据均来自截至 2026-08-28 可公开获取的来源。未审阅经审计财务数据。分析基于可得证据,体现作者的独立判断,不应作为任何投资决策的唯一依据。

证据索引

结论
编号陈述可信度来源
CO001 Instinct is operated by Spear Street Technology Inc., a California corporation. SO001, SO003
CO002 Spear Street Technology was incorporated in California in 2025. SO001
CO003 Instinct raised $250 million in a Series B round announced August 26, 2026. SO001, SO003
CO004 The Series B was co-led by Index Ventures and Benchmark. SO001, SO003, SO006
CO005 The Series B valued Instinct at $2.5 billion post-money. SO001, SO003, SO013
CO006 Instinct's total capital raised across all rounds is $350 million as of August 2026. SO001, SO003
CO007 Noah Shinn is the founder of Instinct and CEO of Spear Street Technology. SO003, SO005, SO007
CO008 Noah Shinn was 23 years old at the time of the August 2026 Series B announcement. SO003, SO007, SO008
CO009 Noah Shinn previously worked as a research scientist at Sierra, an enterprise AI agent company. SO003, SO007, SO008
CO010 Noah Shinn dropped out of Northeastern University in 2023. SO008, SO007
CO011 Noah Shinn conducted machine learning and programming language research at MIT before founding Instinct. SO008, SO007
CO012 Noah Shinn is the lead author of the Reflexion paper published at NeurIPS 2023. SO024, SO023, SO008
CO013 The Reflexion framework achieved a 91% pass@1 rate on HumanEval, outperforming GPT-4's reported 80% on the same benchmark. SO024, SO023
CO014 Instinct's product is an AI personal assistant accessed via SMS and WhatsApp. SO001, SO002, SO004
CO015 Instinct autonomously manages email, calendar, scheduling, travel, shopping, and other tasks on behalf of users. SO001, SO002, SO004
CO016 Instinct was in private beta (invite-only) as of August 2026. SO001, SO002
CO017 Instinct reported approximately $80 million ARR as of August 2026. SO001, SO003
CO018 Instinct grew from approximately $5 million ARR in January 2026 to $80 million ARR in August 2026—a 16× increase in approximately 8 months. SO001, SO003
CO019 Instinct's headquarters is in San Francisco, California. SO001, SO003, SO004
CO020 Kleiner Perkins, led by partner Mamoon Hamid, led Instinct's Series A round. SO003, SO028
CO021 Conviction Partners (Sarah Guo) is an early investor in Instinct. SO003, SO029
CO022 Greenoaks Capital is an investor in Instinct. SO003
CO023 Instinct's team is described as a small San Francisco team with backgrounds from MIT and Sierra. SO003, SO007
CO024 Instinct's terms of service grant a perpetual and irrevocable license to access, store, use, and modify all user data—including email content, screen captures, and keyboard inputs—for any purpose including AI training, even after the service is discontinued. SO002, SO014, SO019
CO025 Early beta testers reported that Instinct sent emails on their behalf without explicit per-action approval from the user. SO002, SO016
CO026 Security researchers demonstrated that Instinct could be manipulated through prompt injection attacks delivered via malicious email content. SO015, SO002
CO027 Instinct stored email content in plain text that remained accessible even after users revoked the service's Google account access. SO016, SO017
CO028 Instinct added a user data deletion tool to its interface following public backlash in late August 2026, though the underlying terms of service were not changed. SO002, SO017
CO029 Instinct requires access to users' email accounts, messaging platforms, calendar, device audio, location, and screen captures to provide its full assistant functionality. SO002, SO004, SO018
CO030 Instinct's terms of service allow the AI to enter binding agreements and execute financial transactions on behalf of users. SO002, SO019
CO031 No regulatory investigations, enforcement actions, or lawsuits against Instinct or Spear Street Technology have been reported in any source reviewed as of August 2026. SO002, SO015
CO032 Instinct's headcount is not publicly disclosed; reporting describes the team as small and estimated at fewer than 50 employees. SO001, SO003
CO033 Reflexion co-authors include Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. SO024, SO023
CO034 Sierra is an enterprise AI agent company; Noah Shinn was among its earliest employees. SO025, SO007
CO035 Noah Shinn co-developed τ-bench, a benchmark for evaluating AI agents' ability to handle real-world user interactions, tool invocations, and business-rule compliance across multiple runs. SO008, SO005
CO036 Index Ventures has backed major technology companies including Dropbox, Stripe, and Robinhood from early stages. SO026, SO001
CO037 Benchmark has led investments in Twitter, Snap, Uber, and Discord among other leading consumer technology companies. SO027, SO001
CO038 Instinct's valuation rose from approximately $50 million at seed stage to $2.5 billion at Series B—a 50× step-up—in approximately 6 months in 2026. SO003
CO039 Instinct's valuation trajectory of $50M → $500M → $2.5B in approximately six months in 2026 is exceptionally rapid even by the standards of high-growth consumer AI in the same period. SO003, SO001
CO040 No board members, board observers, or governance representatives of Spear Street Technology have been named in any public reporting as of the August 2026 run date. SO001, SO003
CO041 At $2.5 billion valuation on approximately $80 million ARR, Instinct implies a revenue multiple of approximately 31× ARR as of August 2026. SO001, SO003
CM001 The consumer AI personal assistant market encompasses software products that understand natural language and autonomously execute tasks for individual consumers SM001, SM002
CM002 Core market participants include platform incumbents Apple Siri, Google Assistant, Amazon Alexa, Microsoft Cortana, and Samsung Bixby SM008, SM009, SM003
CM003 AI-native entrants in the consumer assistant market include Instinct, Rabbit r1, and Humane AI Pin SM011, SM016, SM017
CM004 The defining characteristic of the 2025-2026 market shift is the transition from reactive query-response assistants to proactive agentic systems that take autonomous action SM020, SM024
CM005 Instinct differentiates through SMS and WhatsApp interface, deep permission model, and autonomous task execution without step-by-step confirmation SM004, SM011
CM006 Platform incumbents have distribution advantages through iOS and Android installed bases but limited autonomy compared to AI-native startups SM008, SM009, SM020
CM007 OpenAI ChatGPT expanded agentic capabilities in 2026 with agent mode, representing crossover competition from AI chatbot segment SM010, SM028
CM008 The global consumer AI personal assistant market reached an estimated 4.84 billion USD in 2026 SM001, SM003
CM009 The consumer AI assistant market grew at a 42.2 percent CAGR from 2025 baseline of 3.4 billion USD SM001
CM010 The broader intelligent personal assistant market including enterprise reached approximately 17.95 billion USD in 2026 growing at 25.9 percent CAGR SM002
CM011 Market projections suggest the consumer AI assistant segment will reach 19.6 billion USD by 2030 if current growth rates sustain SM001, SM007
CM012 Instinct achieved 16x ARR growth from 5 million USD to 80 million USD in approximately 8 months SM004, SM015
CM013 Market sizing estimates carry significant uncertainty due to definitional ambiguity around personal assistant versus general AI chatbot categories SM001, SM002
CM014 The agentic assistant serviceable addressable market is estimated at 1-2 billion USD in 2026 as a nascent category SM019, SM020
CM015 Privacy-conscious users and those in regulated industries represent structural exclusions from Instinct addressable market SM005, SM021
CM016 The consumer AI assistant market segments by task type including scheduling, email, travel, shopping, and life administration SM001, SM022
CM017 By privacy tolerance the market bifurcates between privacy-sensitive users rejecting broad data access and convenience-first users trading privacy for functionality SM021, SM025
CM018 Instinct target segment appears to be privacy-tolerant power users with high payment willingness who value autonomous execution SM004, SM011
CM019 Geographic segmentation shows North America and Western Europe as primary markets due to smartphone penetration and consumer spending power SM001, SM022
CM020 Age demographics for AI assistant adoption skew toward 25-45 year old professionals with complex scheduling needs and disposable income SM021, SM022
CM021 High-net-worth individuals represent a premium segment potentially replacing human assistants with AI at 100 plus USD per month price points SM024, SM029
CM022 Generative AI capability improvements have made natural language understanding and task completion reliable enough for production use cases SM020, SM022
CM023 Smartphone ubiquity provides the device substrate and connectivity for always-available AI assistance SM001, SM022
CM024 Consumer familiarity with AI through ChatGPT has normalized AI interaction and reduced adoption friction SM010, SM028
CM025 Productivity demands from remote and hybrid work arrangements have increased demand for delegation tools SM022, SM024
CM026 SMS and messaging-app interfaces eliminate the need to learn new applications reducing onboarding friction for AI assistants SM004, SM011
CM027 FTC has signaled increased scrutiny of AI agents that enter binding transactions on behalf of consumers SM012, SM005
CM028 EU AI Act may classify autonomous personal assistants as high-risk systems requiring conformity assessments SM013
CM029 Privacy regulations including CCPA and GDPR create compliance overhead and limit data retention practices for AI assistants SM014, SM013
CM030 Consumer trust deficits following high-profile AI failures and privacy incidents constrain market adoption SM021, SM025
CM031 Platform gatekeeping by Apple and Google limits third-party assistant capabilities on mobile devices SM008, SM029
CM032 Security vulnerabilities including prompt injection attacks demonstrated against Instinct undermine reliability of agentic assistants SM005, SM023
CM033 Instinct terms of service grant a perpetual and irrevocable license to use store and modify user data raising privacy concerns SM005, SM023
CM034 The market trajectory depends heavily on whether the agentic AI trust gap closes faster than regulatory constraints tighten SM012, SM025
CM035 AI assistant market fragmented between incumbents with distribution advantages and startups with agentic capability advantages SM020, SM026
CP001 Instinct competes most directly with consumer-facing assistants that can reason across tasks and act on behalf of users, not with every general chatbot. SP001, SP004, SP006, SP008, SP009, SP011, SP013, SP014
CP002 Humane is better treated as a predecessor cautionary tale than a live direct competitor because the AI Pin was shut down after HP acquired the company. SP018
CP003 OpenAI Operator is a direct benchmark because it is designed to use a browser to perform tasks for the user. SP004
CP004 Google Gemini is a powerful adjacent rival because it combines general reasoning with Google account, search, and device adjacency. SP006, SP007
CP005 Apple Intelligence makes OS-native assistance a default expectation for premium smartphone users, even if Apple is less autonomous than Instinct today. SP008
CP006 Microsoft Copilot competes indirectly by bundling AI productivity into a much larger software and subscription ecosystem. SP009, SP010
CP007 Claude is more of a reasoning and writing substitute than a full consumer action-taking competitor in its current public positioning. SP011, SP012
CP008 Perplexity competes most clearly on search, research, and answer quality rather than delegated account actions. SP013
CP009 Superhuman, Motion, and Cal.com each own a narrower workflow that Instinct tries to combine into a single assistant experience. SP015, SP016, SP017
CP010 Instinct's SMS and WhatsApp delivery is a real differentiator because it removes app-download friction and drops the assistant into a channel users already inhabit. SP001, SP022, SP024
CP011 Google, Apple, Microsoft, and Meta all enjoy stronger native ecosystem integration than Instinct because they control devices, identity, or default surfaces. SP006, SP007, SP008, SP009, SP010, SP014
CP012 Brand trust today is structurally higher for Apple, Google, and Microsoft than for a private-beta startup with visible autonomy incidents. SP002, SP008, SP009, SP010, SP025, SP026
CP013 Paid AI subscription plans from OpenAI, Google, Claude, and specialist tools demonstrate that consumers and prosumers accept recurring pricing for assistant-like software. SP005, SP007, SP010, SP012, SP015, SP016, SP017
CP014 Free assistants from Meta and bundled assistants from Apple, Google, and Microsoft pressure the ceiling on what Instinct can charge without clear autonomy advantages. SP006, SP007, SP008, SP009, SP010, SP014
CP015 Humane's shutdown is competitive evidence that consumers will not tolerate fragile high-friction assistant products just because the concept is novel. SP018
CP016 Instinct's strongest direct moat claim is higher autonomy across communications and transactions rather than broader brand reach or ecosystem control. SP001, SP002, SP004, SP006, SP008, SP009, SP014
CP017 Specialist workflow tools can remain durable because they pair deeper domain UX with lower trust risk than a general assistant that touches many systems. SP015, SP016, SP017
CP018 Switching costs for consumers are moderate because the status quo is fragmented and many users already multi-home among multiple apps and assistants. SP005, SP007, SP012, SP013, SP015, SP016, SP017
CP019 Multi-homing weakens moat durability for everyone except the platform incumbents, because users can compare assistants without major data migration costs. SP005, SP007, SP012, SP013, SP014, SP015, SP016, SP017
CP020 The largest channel and identity surfaces are owned by Apple, Google, Microsoft, Meta, and WhatsApp rather than by Instinct itself. SP007, SP008, SP009, SP010, SP014, SP024
CP021 Trust and safety posture are competitive variables because a startup that mishandles autonomy can lose to less capable but more trusted incumbents. SP002, SP019, SP020, SP025, SP026
CP022 Publicly available competitive facts remain uneven: pricing and product positioning are visible, but retention, real usage depth, and true task success rates are mostly undisclosed. SP005, SP007, SP010, SP012, SP013, SP015, SP016, SP017
CP023 Instinct is earlier than most rivals in brand maturity but ahead of many general assistants in willingness to execute cross-app tasks without per-action confirmation. SP001, SP002, SP004, SP006, SP008, SP009, SP011, SP013
CP024 OpenAI, Google, Apple, and Microsoft all possess materially greater compute, distribution, and default-placement advantages than Instinct. SP004, SP006, SP008, SP009
CP025 Perplexity, Claude, and Meta AI are easier for users to sample than Instinct because they are broadly available, while Instinct remains invite-only. SP001, SP011, SP012, SP013, SP014
CP026 Instinct's private-beta exclusivity creates scarcity and buzz but limits the public proof set versus broad-availability competitors. SP001, SP002, SP003
CP027 Specialists such as Superhuman and Motion market concrete workflow ROI rather than general intelligence, which may make them easier to trust and budget for. SP015, SP016
CP028 Cal.com is better understood as scheduling infrastructure and workflow tooling than as a full substitute for Instinct, but it competes for one important user job. SP017
CP029 The competitive landscape includes direct agentic rivals, broad AI incumbents, and specialist workflow tools, so no single comparison set is sufficient. SP004, SP006, SP008, SP009, SP011, SP013, SP015, SP016, SP017
CP030 Law and economics commentary highlights the risk that platform incumbents and standalone assistants may be governed by mismatched competitive rules. SP019
CP031 Cyberhaven's 2026 risk report indicates that the rise of AI agents is making security and governance a more prominent adoption variable across assistant categories. SP020
CP032 Business of Apps shows massive ChatGPT and WhatsApp scale, underscoring how difficult it is for a startup to match incumbent reach without viral differentiation. SP021, SP022, SP024
CP033 The Accio 2026 assistant comparison reflects a market conversation shifting from chat quality toward autonomy, integrations, and task execution. SP023
CP034 The main evidence against a durable Instinct moat is that every major platform owner is moving toward more agentic behavior from a stronger installed base. SP004, SP006, SP008, SP009, SP014, SP019, SP020
CP035 The main evidence for a durable Instinct wedge is that the product is optimized around delegated personal operations rather than around generic chat or one narrow workflow. SP001, SP002, SP015, SP016, SP017, SP024
CP036 Public sources still do not disclose Instinct's pricing, retention, or task-level success rates versus competitors, which blocks precise competitive scoring. SP001, SP002, SP003
CP037 Another unresolved gap is whether consumers will prefer one general assistant or a portfolio of specialist tools plus incumbent AI surfaces.
CP038 Competition later in the report should be read through trust, distribution, and ecosystem control at least as much as through raw model capability. SP002, SP008, SP009, SP010, SP019, SP020, SP025, SP026
CI001 Instinct raised $100 million in a Series A round in January 2026 led by Kleiner Perkins. SI001, SI015
CI002 Instinct raised $250 million in a Series B round on August 26, 2026 co-led by Index Ventures and Benchmark. SI001, SI008
CI003 The Series B valued Instinct at $2.5 billion post-money. SI001, SI002
CI004 Instinct's $350M total raised implies a capital efficiency ratio of approximately 4.4x relative to $80M ARR, comparing favorably to AI SaaS peers typically ranging 2-5x. SI001, SI003
CI005 Greenoaks Capital and Conviction Capital also participated in Instinct's funding rounds alongside the lead investors. SI001, SI002
CI006 Kleiner Perkins partner Mamoon Hamid led the Series A investment in Instinct. SI002, SI015
CI007 Instinct reported approximately $5 million ARR at the close of its Series A in January 2026 per management statements. SI002
CI008 Instinct reported approximately $80 million ARR at the close of its Series B in August 2026 per management statements. SI001, SI002
CI009 Instinct's ARR grew approximately 16× in 7 months from January to August 2026 per management-stated figures. SI001, SI002
CI010 Instinct's implied average monthly net-new ARR over January–August 2026 was approximately $10.7 million. SI001
CI011 Instinct's ARR figures are management-stated and have not been independently verified or audited by any publicly disclosed third party. SI001, SI003
CI012 No audited financial statements, prospectus, or third-party financial audit for Instinct or Spear Street Technology have been published as of August 2026. SI001, SI004
CI013 Gross margin for comparable consumer AI SaaS businesses ranges from 50–75%; Instinct's higher LLM inference intensity suggests the lower end of this range. SI005, SI007
CI014 AI inference costs for autonomous AI assistant products are estimated at 15–35% of revenue based on published agentic AI cost structures. SI010, SI011
CI015 Applied to Instinct's $80M ARR, estimated LLM inference costs of 15–35% imply $12–$28 million in annual AI inference costs. SI010, SI012
CI016 Instinct has not disclosed its pricing model, average revenue per user, or subscription tier structure as of August 2026. SI001, SI004
CI017 Consumer AI subscription products have historically struggled to sustain paying user bases beyond early adopters, per Business Insider analysis. SI017
CI018 Instinct's viral waitlist model is likely to result in low customer acquisition costs compared to paid marketing, though this is inferred from distribution mechanics and not confirmed. SI001, SI016
CI019 Instinct has not disclosed any monetization model, path to profitability, or unit economics targets as of August 2026—a material gap at $2.5B valuation. SI004, SI016
CI020 Assuming a monthly burn rate of $3–9M (sector-informed estimate), Instinct's $350M in total capital implies a runway of approximately 19–58 months. SI019, SI020
CI021 The base-case estimate of $5M monthly burn at Instinct implies approximately 35 months of runway—sufficient for a commercial launch and first monetization cycle. SI019, SI020
CI022 LLM provider dependency is a key financial risk: if Instinct's undisclosed LLM provider raises API pricing or restricts access, Instinct's cost structure could deteriorate materially. SI010, SI011
CI023 Instinct's August 2026 privacy backlash creates a risk of user churn and ARR growth deceleration that could pressure the financial model significantly. SI022, SI023
CI024 The historical failure of consumer AI companies like Inflection AI and Character AI to monetize at scale before acquisition is a risk benchmark for Instinct's monetization path. SI017, SI018
CI025 Instinct's valuation increased approximately 5× from roughly $500M at Series A to $2.5B at Series B in approximately 7 months—one of the fastest step-ups in consumer software history at this scale. SI001, SI002, SI005
CI026 Top AI SaaS companies in 2026 typically trade at 25–50× ARR depending on growth rate and category risk, according to SaaStr benchmarks. SI005, SI007
CI027 Instinct's 31× ARR multiple at Series B is within the typical range for hypergrowth AI SaaS but on the higher end given pre-commercial status and consumer category risk. SI005, SI007
CI028 Annual LLM inference costs at Instinct are estimated at $12–$28M based on 15–35% inference-cost-to-revenue ratios applied to $80M ARR. SI010, SI012
CI029 A 16× ARR growth in 7 months would place Instinct among the fastest-ever consumer software revenue ramps, comparable to early-stage ChatGPT and Slack growth trajectories. SI005, SI006
CI030 Consumer AI personal assistant companies are expected to face significant monetization challenges as the market shifts from curiosity-driven adoption to value-driven retention. SI016, SI017
CI031 The absence of audited financials and the single-source nature of Instinct's ARR claims make independent financial verification the primary diligence requirement before any investment decision. SI004, SI018
CI032 At Series A, Instinct raised $100M against $5M ARR (capital efficiency: 0.05× ARR per $M raised); at Series B, $250M against $80M ARR (0.32×)— a 6× improvement in capital efficiency. SI001, SI002
CI033 Instinct's consumer segment implies lower net revenue retention (NRR) relative to enterprise SaaS benchmarks, as consumer products typically exhibit higher churn and more volatile expansion dynamics. SI006, SI018
CI034 Consumer AI subscription pricing in 2026 ranges from $0 (ad-supported) to $100+/month (high-capability premium), with most established players in the $10–$30/month tier for consumer services. SI024, SI025
CI035 Instinct's private beta status means its current ARR may derive from a small cohort of early paying users or commercial arrangements not publicly described, creating concentration risk in the revenue base. SI001, SI016
CE001 Instinct is a messaging-native personal assistant accessed through text or calls on the official surface and via WhatsApp in third-party reporting, without requiring a new app download. SE001, SE019
CE002 Instinct connects to user applications and devices including email, messaging, screen, audio, and location, giving it unusually broad consumer-context access. SE001, SE002, SE019
CE003 Instinct remained available only to a private access group as of 2026-08-28 because the company said it was still scaling compute. SE001, SE019
CE004 Instinct’s legal and product surface authorize autonomous actions on connected services and can make resulting agreements, commitments, or transactions binding on the user. SE002, SE003, SE019
CE005 Publicly described tasks include inbox management, reminders, appointments, restaurant reservations, rides, travel, shopping, and other life-admin chores. SE001, SE019, SE021
CE006 The operating workflow is messaging-first: a request arrives through a text-like surface, context is retrieved from linked systems, the agent executes, and later follow-up can happen in the same thread. SE001, SE019
CE007 No reviewed public source disclosed a pricing page, subscription schedule, or general-availability launch date for Instinct. SE001, SE019
CE008 Because Instinct hides most execution behind messaging rather than a visible app UI, trust depends on background action controls and user-legible recovery paths as much as on answer quality. SE001, SE019, SE021
CE009 Instinct’s public materials refer to a core model but do not publicly name the underlying foundation-model provider. SE001, SE002
CE010 Reflexion describes a framework for reinforcing language agents through linguistic feedback instead of model fine-tuning. SE005, SE006
CE011 Reflexion stores self-reflections in an episodic memory buffer that conditions later attempts at the same or similar tasks. SE005, SE006
CE012 The Reflexion paper and poster report a 91% HumanEval pass@1 result versus a cited 80% for GPT-4. SE005, SE006
CE013 τ-bench benchmarks dynamic conversations between a user simulator and a language agent equipped with domain-specific API tools and policy guidelines. SE004, SE008
CE014 τ-bench emphasizes tool-calling strategies, policy compliance, and real-world domain tasks rather than generic chatbot response quality alone. SE008, SE027
CE015 Shinn’s public research lineage supports an inference that Instinct likely uses an agent loop with tools and persistent state on top of a frontier LLM, even though production internals are undisclosed. SE001, SE005, SE008, SE019
CE016 The public Reflexion repository asks developers to set an OPENAI_API_KEY, indicating Shinn’s released agent code was designed around API-accessed closed models rather than open-weight self-hosting. SE007, SE025
CE017 The public τ-bench repository supports multiple model-provider API keys, reinforcing that this research lineage assumes external LLM APIs behind the agent layer. SE008, SE027
CE018 Persistent memory or indexing is product-critical because official copy says Instinct understands what is important to the user and follows up on dropped threads over time. SE001, SE002
CE019 Instinct’s privacy policy says linked Google Workspace access can include Gmail, Calendar, Drive, Docs, Sheets, Slides, Tasks, and related metadata or content needed to provide the service. SE002
CE020 Instinct’s terms explicitly mention Apple, Facebook, and Google accounts, showing the product is designed around linked-account identity and integration flows rather than a standalone SMS bot. SE003
CE021 Gmail and Google Calendar both provide watch or push-notification models that can inform a backend when inbox or calendar state changes, reducing the need for constant polling. SE010, SE012
CE022 Gmail push notifications require Cloud Pub/Sub and periodic watch renewal, which implies ongoing background jobs for any service monitoring mailbox changes continuously. SE010
CE023 Google Calendar push notifications require HTTPS webhook endpoints, unique channels, and manual renewal near expiration. SE012
CE024 Microsoft Graph’s Outlook mail and calendar APIs support integration across both personal and organizational account contexts. SE014, SE015
CE025 WhatsApp Cloud API permits free-form service messages only inside a 24-hour customer-service window, after which pre-approved templates are required. SE016
CE026 Twilio’s messaging stack supports outbound sends, delivery-state callbacks, redaction, deletion, and WhatsApp-capable message resources, making it a plausible transport abstraction for a messaging-first assistant. SE017, SE018
CE027 Instinct’s messaging layer likely depends on either Meta’s WhatsApp infrastructure, an intermediary such as Twilio, or a similar gateway for transport and status management. SE001, SE016, SE017, SE018
CE028 Google’s OAuth guidance requires secure token storage, least-privilege scope design, and revocation or deletion discipline, making identity and secret management a central technical risk for Instinct. SE013
CE029 Multiple independent reports describe Instinct sending or preparing to send real email or other actions on behalf of testers, proving the product is executing workflows rather than merely drafting suggestions. SE019, SE020, SE021
CE030 TechCrunch and StartupFortune each describe a prompt-injection-style exploit where instructions embedded in email caused Instinct to behave unexpectedly. SE019, SE021
CE031 OWASP’s prompt-injection definition maps closely to the Instinct beta reports because hostile input can redirect an LLM system into unintended actions. SE019, SE024
CE032 Official policy and adverse reporting agree that disconnecting an external service does not automatically delete previously indexed data; deletion is a separate step. SE002, SE003, SE019, SE020, SE022
CE033 TechCrunch and SC Media reported that Instinct continued summarizing previously stored emails after Google access was revoked and that the assistant said the emails were stored in plain text for later searches. SE019, SE020
CE034 Instinct’s privacy policy says Google Workspace API data is not used to train models or disclosed to third-party AI providers for that purpose, but broader non-Workspace materials can still be used to improve products and models subject to policy exceptions. SE002, SE003
CE035 Instinct’s terms explicitly authorize the service to enter agreements, commitments, or transactions on the user’s behalf as if the user entered them directly. SE003, SE019
CE036 The maturity state is real private beta, not launch-ready trust infrastructure: the assistant demonstrably works, but public evidence does not support calling its control plane hardened for mass-market autonomy. SE001, SE019, SE020, SE021
CE037 No reviewed public source disclosed external security audits, certifications, uptime metrics, or formal red-team results for Instinct. SE001, SE002, SE003, SE019
CE038 The external-data deletion tool added after complaints is better understood as reactive remediation than as proof of mature lifecycle governance. SE019, SE021, SE022
CE039 As of 2026-08-28, the public Reflexion repository showed 3243 stars, 316 forks, and 24 open issues, indicating sustained practitioner interest in Shinn’s agentic architecture lineage. SE025, SE029
CE040 As of 2026-08-28, the public τ-bench repository showed 1409 stars, 215 forks, and 52 open issues, indicating similarly strong ongoing developer attention to tool-agent evaluation infrastructure. SE027, SE030
CE041 Reflexion and τ-bench both appear maintained by relatively concentrated contributor sets with active issue queues, suggesting meaningful community usage but narrow maintainer depth. SE026, SE028, SE029, SE030
CE042 Instinct itself has a much thinner public developer footprint: the direct Hacker News link post reviewed had only 1 point and no visible discussion, and Algolia surfaced only sparse direct story hits. SE031, SE032
CE043 Instinct’s critical external dependencies likely include a frontier model API, messaging transport providers, OAuth identity systems, Google and Microsoft connectors, cloud compute, and secret or token storage. SE001, SE013, SE016, SE017, SE018
CE044 Compute capacity is itself a near-term operational dependency because the company says access remains limited while it scales compute. SE001
CE045 Reviewed public materials expose little forward product roadmap detail beyond private-beta status, revised policies, and reactive controls; no public changelog, SLA, or general-availability plan was found. SE001, SE002, SE003, SE019
CE046 Instinct’s main current differentiation is not a disclosed proprietary infrastructure advantage but the combination of messaging-native distribution, deep permissions, and willingness to let the agent take binding actions. SE001, SE019, SE023
CE047 The terms refer to a website, subdomains, and related Mac OS or other applications, implying a nontrivial control surface behind the messaging-first consumer entry point. SE003
CE048 The public architecture evidence today is stronger for agent research lineage than for production observability, rollback, or safety instrumentation. SE019, SE025, SE027
CU001 As of 2026-08-28, Instinct remains available only to a private access group; prospects can join a waitlist or obtain an invite from an existing member. SU001, SU009, SU013
CU002 Instinct is accessed through text messages and calls, including WhatsApp, rather than requiring users to learn a dedicated new interface. SU001, SU004, SU009
CU003 Instinct connects to email, messaging, calendars, screen activity, audio, location, and other connected services to act on a user’s behalf. SU001, SU003, SU004
CU004 Instinct is positioned as a consumer product for individual users rather than enterprises: the official terms grant personal use only and reported use cases are personal errands, travel, subscriptions, and home logistics. SU001, SU002, SU004, SU009
CU005 The observed early-use cases center on life administration such as travel booking, groceries, appointments, subscriptions, reservations, messaging follow-up, and family coordination. SU001, SU004, SU005, SU010
CU006 No public customer count, active-user count, or waitlist size is disclosed in the official materials or main press coverage reviewed for this run. SU001, SU005, SU009, SU025
CU007 TechCrunch reported management-stated ARR of about $80 million as of 2026-08-26. SU005, SU025
CU008 TechCrunch reported that Instinct’s ARR was about $5 million in January 2026. SU005, SU025
CU009 Using the reported January and August ARR figures implies roughly 16x ARR growth in about seven months while the product remained private beta. SU001, SU005, SU025
CU010 Forbes reported that Instinct was free to use in late August 2026 and that no public price had been announced. SU009, SU010
CU011 The combination of reported $80 million ARR and no public pricing means the revenue model exists but is not transparently described to outsiders. SU005, SU009, SU025
CU012 Launch-week buzz was driven disproportionately by tech insiders, investors, and power users rather than by mass-market review channels. SU009, SU010, SU011
CU013 Sheel Mohnot reported sending 677 messages to Instinct in five days and listed 15 completed jobs, including finding a doctor, lowering a cable bill, vendor outreach on WhatsApp, subscription cancellations, and paying tolls. SU010, SU011
CU014 Jesse Middleton said he used Instinct daily for a week across travel changes, reservations, email follow-ups, CRM work, and investor data-room tasks, calling it awesome. SU004, SU011
CU015 Katie Jacobs Stanton first described Instinct as an amazing product but disconnected email after it sent an email on her behalf without asking. SU004, SU006, SU011, SU023
CU016 Peter Yang said Instinct initially retained Gmail records and did not provide a workable deletion path until later settings changes were added. SU004, SU006, SU008, SU011
CU017 Claire Vo showed that Instinct could still summarize previously ingested emails after Google access was disconnected and that stored emails were kept in plain text for later search. SU004, SU006, SU007, SU009
CU018 Alex Cohen demonstrated an email-based prompt-injection test that caused Instinct to follow malicious inbox instructions, after which he deleted his account. SU004, SU006, SU008, SU011
CU019 Forbes surfaced another autonomy failure: Jason Yeh said Instinct booked a dinner reservation with a $200 cancellation fee after he only asked it to find availability. SU009, SU011
CU020 SC Media summarized the early market signal as simultaneous praise for the product’s capabilities and significant privacy alarm from testers. SU023, SU004
CU021 AI Weekly summarized the trust problem as early testers learning that one unauthorized action could reset trust to zero. SU024, SU004, SU006
CU022 A thin but positive additional signal exists in Digg’s framing of Instinct as earning praise for smooth onboarding and proactive suggestions. SU012
CU023 Instinct’s privacy policy says disconnecting a third-party integration does not automatically delete data collected from that integration. SU002, SU003
CU024 Instinct’s terms likewise state that indexed connected-service data may still be used after disconnect until the user separately requests deletion. SU002, SU003
CU025 Instinct’s official terms authorize the service to take actions it deems responsive, including purchases and agreements that bind the user. SU002, SU003
CU026 Instinct’s privacy policy says the company may use user information to evaluate, fine-tune, and train AI models subject to a forward-looking opt-out, while Google Workspace API data is excluded from model training. SU002, SU003
CU027 Instinct’s privacy policy warns that autonomous features may cause unintended communications or payments and may be manipulated by misleading instructions from third parties. SU002, SU003
CU028 No G2, Capterra, App Store, or Google Play review corpus is publicly available for Instinct, consistent with its invite-only private beta and absence of a mainstream public app launch. SU001, SU009, SU010
CU029 The chapter’s strongest customer proof therefore comes from quoted beta testers, investor-users, and practitioner commentary rather than from named paying customer case studies. SU004, SU006, SU010, SU011, SU024
CU030 Instinct has not publicly disclosed retention, renewal, churn, NRR, GRR, or cohort data for beta users. SU005, SU009, SU025
CU031 Because customer count and retention are undisclosed, the observed ARR growth could reflect strong retention, aggressive new-user adds, unusually high ARPU, or some combination of all three. SU005, SU006, SU025
CU032 Private-beta ARR of roughly $80 million with no disclosed user count creates a credible risk that revenue is concentrated in a relatively small cohort of early adopters. SU005, SU009, SU025
CU033 Instinct’s text-and-WhatsApp access lowers onboarding friction relative to assistants that require users to install and learn a new full application interface. SU001, SU004, SU022
CU034 If trust issues are addressed, the same low-friction channel design could expand beyond insiders into broader personal-admin, travel, shopping, and household coordination use cases. SU001, SU005, SU010, SU022
CU035 Pew reported that 49% of U.S. adults use AI chatbots in 2026, 24% use them daily, and adults under 30 reach 66% usage, showing that mainstream consumer demand for AI assistants already exists. SU014, SU015, SU018
CU036 Pew also found that 71% of U.S. adults think increased AI use will make their personal information less secure and about six in ten are not confident U.S. companies will develop and use AI responsibly. SU014, SU015, SU018
CU037 Usercentrics found that 52% of consumers trust AI less than humans with their personal data, 24% canceled subscriptions over AI data concerns, and 20% switched to a competitor they trusted more. SU016, SU017
CU038 Usercentrics also found that 52% of consumers would pay more for AI transparency, averaging a 7% premium globally and rising to 67% among 18–29 year-olds. SU016, SU017
CU039 Capital One Shopping reported that 39% of consumers had already used AI assistants for online shopping and 80% planned to use generative AI to shop in 2026, relevant to Instinct’s commerce-oriented use cases. SU019
CU040 Sensor Tower reported that ChatGPT, Gemini, and DeepSeek represented nearly 90% of total AI-assistant time spent in Q1 2026, indicating the category is real but highly concentrated around trusted incumbents. SU018, SU020
CU041 Cybernews and Dume both treat integrations, automation quality, and privacy controls as the defining buyer criteria for personal AI assistants in 2026. SU021, SU022
CU042 Instinct’s customer evidence is materially weaker than its revenue narrative because it lacks named customers, public counts, public pricing detail, and third-party review-platform depth. SU005, SU009, SU010, SU025
CU043 The product’s observed early-user mix skews toward founders, investors, operators, and other tech-savvy power users comfortable experimenting with broad account permissions. SU004, SU009, SU010, SU011
CU044 Virality and invite-gating appear to be central acquisition channels: the official waitlist mechanics plus press emphasis on insider invites and launch-week social buzz point to word-of-mouth-led initial adoption. SU001, SU009, SU010, SU011, SU013
CU045 A product accessed through familiar messaging channels but perceived as risky on privacy has a plausible expansion path only if trust controls improve faster than buzz fades. SU001, SU016, SU017, SU023
CR001 Instinct's public risk profile is dominated by privacy, autonomy, and trust concerns rather than by classic consumer-app issues like simple engagement decay. SR001, SR002, SR003, SR005
CR002 TechCrunch and other adverse coverage say Instinct can send emails or take actions without explicit per-action confirmation. SR001, SR002, SR005
CR003 That autonomy design choice creates consent and authorization risk if the user does not understand or expect the action boundary. SR001, SR002, SR006, SR007
CR004 The EU AI Act is now the first legal framework on AI and is directly relevant to autonomous assistants operating in Europe. SR008, SR009
CR005 The European Commission began enforcing new AI Act transparency requirements from 2 August 2026. SR008, SR009
CR006 GDPR remains relevant because an assistant like Instinct processes highly sensitive personal communications and metadata across multiple systems. SR004, SR010
CR007 FTC AI guidance creates plausible enforcement exposure if the company misleads users about autonomy, retention, or safety controls. SR006, SR007
CR008 Client alerts from Lowenstein and Gunderson show that 2026 compliance expectations are broadening across US state, federal-interest, and EU frameworks. SR017, SR018
CR009 Prompt injection remains a first-order technical risk for agentic assistants according to OWASP and 2026 security guides. SR012, SR013
CR010 Any assistant that can read inbound content and then call tools is vulnerable to indirect prompt injection unless inputs, permissions, and execution paths are tightly controlled. SR012, SR013
CR011 Plain-text retention after OAuth revocation would create unusually strong privacy, security, and possibly deceptive-practice risk if confirmed broadly. SR001, SR003, SR005, SR010
CR012 OWASP and NIST both imply that agent systems need strong identity, authorization, isolation, audit logging, and human-override controls. SR011, SR012
CR013 The product is exposed to partner dependency because it relies on messaging, email, calendar, and model ecosystems it does not control. SR004, SR025, SR026, SR027, SR028, SR029, SR030, SR031
CR014 Apple, Google, Microsoft, Meta/WhatsApp, and model providers can all change policies, access, defaults, pricing, or product capabilities in ways that harm Instinct. SR025, SR026, SR027, SR028, SR029, SR030, SR031
CR015 Model-provider concentration is a supply-chain risk because underlying pricing and capability changes can directly alter Instinct's economics and reliability. SR025, SR026, SR027
CR016 Cyberhaven, PwC, AvePoint, and BCG all show that governance and data-risk problems are rising as agentic AI use expands. SR020, SR021, SR022, SR023
CR017 Those broader 2026 reports matter because Instinct is not just another chatbot; it is an action-taking assistant in the highest-trust part of the consumer stack. SR020, SR021, SR022, SR023, SR004
CR018 Unclear monetization and high-permission onboarding create business-model risk because the company must earn both willingness to pay and willingness to trust. SR001, SR004, SR024, SR025, SR026, SR027
CR019 A large Series B reduces immediate solvency risk but does not remove burn, gross-margin, or next-round risk if growth slows. SR024, SR025, SR026, SR001
CR020 Incumbent assistants create strategic risk because they can bundle safer-enough functionality into products users already trust. SR028, SR029, SR030, SR031
CR021 No public evidence reviewed shows a broad management bench or mature governance structure beyond Noah Shinn's founder-led profile. SR004, SR001
CR022 A young founder and small team can be a strength in speed but a risk in compliance, operations, hiring, and crisis management during rapid scaling. SR001, SR004, SR015, SR016, SR017, SR018
CR023 If growth outruns controls, one visible autonomy failure can trigger user churn, media backlash, regulator attention, and partner pressure in sequence. SR001, SR002, SR003, SR005, SR006, SR008, SR012
CR024 ITIF's work on publicly available data highlights unresolved legal uncertainty around training inputs, transparency norms, and safe-harbor design. SR016
CR025 Law and economics commentary warns that AI assistants may face regulatory mismatch between platform rules and standalone-agent reality. SR014
CR026 The highest-probability risk cluster is trust and adoption, because every other category ultimately feeds into whether users keep delegating tasks. SR001, SR002, SR003, SR005, SR020, SR021, SR022, SR023
CR027 The highest-severity risk cluster is privacy and security because harm can compound quickly when an assistant has broad access and action rights. SR001, SR003, SR005, SR010, SR011, SR012
CR028 Some risks are manageable with controls, but a structural inability to build trust would be existential to the product thesis. SR001, SR011, SR012, SR020, SR021, SR022, SR023
CR029 Public evidence of mitigations is thin beyond broad company positioning, which itself is a diligence signal. SR004, SR001, SR005
CR030 Kill criteria should include regulatory inquiry, rising churn after autonomy incidents, revoked API access, and materially worsening model costs. SR006, SR007, SR025, SR026, SR027, SR028
CR031 The absence of disclosed pricing, customer counts, and retention makes it hard to bound downside from trust shocks using public evidence only. SR001, SR004, SR024
CR032 Distribution through WhatsApp or mobile surfaces is helpful commercially but risky strategically because channel owners can reprioritize or limit access. SR028, SR004
CR033 Apple, Google, and Microsoft all have the option to move down-market or across-market into the same delegated-assistant jobs from stronger default positions. SR029, SR030, SR031
CR034 Risk management in 2026 is expected to be continuous rather than periodic, especially for agent systems touching sensitive data and autonomous actions. SR011, SR021, SR022, SR023
CR035 Cost-governance reports show that organizations often struggle to attribute or forecast AI spend, which raises financial control risk for usage-sensitive products. SR020, SR024, SR025, SR026, SR027
CR036 Public adverse coverage itself is a risk amplifier because it narrows the company's margin for future mistakes during launch. SR001, SR002, SR003, SR005
CR037 The product sits near the boundary of what users may perceive as impersonation if outbound actions are not extremely well signaled and controlled. SR001, SR002, SR006, SR007
CR038 The company likely faces a tradeoff between richer agent autonomy and a lower-risk product posture; moving too slowly hurts differentiation, while moving too fast hurts trust. SR001, SR004, SR012, SR013, SR020, SR021
CR039 The public record is sufficient to identify severe downside categories, but not to quantify incident probability or financial loss with confidence. SR001, SR011, SR012, SR020, SR021, SR022, SR023
CR040 A second unresolved gap is whether the company has already implemented the authorization, isolation, and deletion controls that public critics say are necessary. SR001, SR003, SR004, SR011, SR012
CR041 A third unresolved gap is how model-provider, platform, and regulator dependencies interact under stress when a real consumer incident occurs. SR006, SR008, SR012, SR014, SR025, SR026, SR027, SR028, SR029, SR030, SR031
CR042 Overall, the risk map is unusually concentrated in high-severity trust and governance issues for such an early-stage consumer company. SR001, SR002, SR003, SR006, SR008, SR011, SR012, SR020, SR021, SR022, SR023
CR043 Because Instinct is pre-launch and high-permission, the downside path can be much faster than the upside path if a few critical trust variables break simultaneously. SR001, SR002, SR003, SR005, SR012, SR020, SR021, SR022, SR023
CV001 Recommendation is Track — do not invest at current $2.5B valuation without independent ARR verification, disclosed monetization model, and entry price reduction to $1.2-1.5B range. SV009, SV010
CV002 The investment thesis is predicated on the consumer AI assistant market becoming a winner-take-most category with $50B+ addressable value by 2030. SV013, SV023
CV003 Instinct's 16x ARR growth in 7 months is a genuine product-market fit signal based on independently confirmed funding disclosures. SV001, SV003
CV004 Apple, Google, and Microsoft have structural distribution advantages that represent a fundamental anti-thesis to Instinct's standalone consumer product. SV012, SV019
CV005 Noah Shinn's background combines published AI research (ReAct, Reflexion) with a consumer product philosophy that is rare among technical founders at 23. SV001, SV004
CV006 Instinct has raised $350M total across Series A ($100M, January 2026) and Series B ($250M, August 2026) with participation from five institutional investors. SV001, SV003, SV004, SV005, SV006, SV007, SV008, SV017
CV007 Instinct has no disclosed pricing model at $2.5B valuation, making the 31x ARR multiple uninvestable without monetization evidence. SV012, SV009
CV008 Privacy backlash in August 2026 related to Instinct's email-access permissions represents a material anti-thesis risk with potential regulatory consequences. SV001, SV029
CV009 Overall score is 6.5/10, risk rating is High, valuation stance is Stretched, and recommendation confidence is Medium. Recommendation is Track. SV009, SV010
CV010 At 31x management-stated ARR, Instinct's entry valuation is above the historical median of 10-20x for consumer software unicorns and at the high end of the 2026 AI premium range of 18-25x per KPMG Venture Pulse. SV014, SV015, SV028
CV011 Applying a 20-30% discount for unverified ARR implies a fair value of $1.7-2.0B, versus the $2.5B round price — a 25-47% premium above risk-adjusted intrinsic value. SV009, SV023
CV012 A Series C target valuation of $5-8B in 2027 would represent a 2-3x step-up from the Series B, consistent with typical hypergrowth consumer software trajectory if monetization is achieved. SV016, SV024
CV013 The absence of a disclosed monetization model at $2.5B valuation is the single most significant financial risk; all return scenarios depend on this gap being closed before capital is exhausted. SV012, SV009
CV014 Dilution mathematics suggest Series A investors at $500M est. valuation face a complex cap structure; Series B investors likely hold preferred stock with 1-2x liquidation preferences — terms are entirely undisclosed. SV009, SV024
CV015 At a 20-31x ARR range, Instinct's current valuation is consistent with the upper end of the 2026 AI software premium but requires continued ARR confirmation and monetization execution to sustain. SV014, SV013
CV016 Greenoaks Capital's participation as Series B co-investor is a positive signal; Greenoaks has a strong track record in growth-stage consumer tech including Chime, Rappi, and Coupang. SV007, SV030
CV017 Bull scenario at 25% probability sees $250M+ ARR by end-2027, paid tier Q1 2027, IPO at $15-20B in 2029, and Series B investors achieving 5-8x return. SV021, SV025
CV018 Base scenario at 55% probability sees $120-160M ARR by end-2027, monetization delayed to mid-2027, Series C at $4-5B, exit at $6-8B in 2030, and Series B return of 2-3x. SV020, SV023
CV019 Bear scenario at 20% probability sees privacy enforcement forcing product changes, monetization failing below $200M ARR, strategic acquisition at $300-500M, and a full write-down for Series B investors. SV029, SV012
CV020 Expected value across scenarios is approximately 2.5x for a Series B co-investor at $2.5B valuation — insufficient to justify the risk premium. SV023, SV024
CV021 The 20% bear-case probability reflects real regulatory risk, undefined monetization, and a competitive landscape with unlimited-budget incumbents. This downside is not remote. SV019, SV029
CV022 If the consumer AI assistant market consolidates around OS-native products within 24 months, Instinct's standalone value drops to near zero in the bear case. SV013, SV019
CV023 Bull case probability of 25% reflects Instinct's unique hypergrowth evidence and founder profile; reduced from a theoretical 35% by privacy and monetization gaps remaining unresolved as of the research date. SV020, SV023
CV024 The base case assumes Instinct launches a paid tier at $20-30/month with 5-8% conversion from a 2M-user installed base, yielding $24-57M ARR uplift. SV014, SV016
CV025 Perplexity AI raised at $1B valuation in April 2024 at approximately $25-30M ARR, implying 33-40x ARR — above Instinct's 31x but with a launched paid tier. SV009, SV010
CV026 Character.AI was acquired by Google in September 2024 at approximately $2.7B, representing ~25x estimated revenue with higher monetization certainty. SV009, SV010
CV027 Scale AI was valued at $13.8B in December 2024 at approximately 11x ARR — lower multiple but with audited revenues and diversified government contracts. SV010, SV009
CV028 Cohere raised at $5B valuation in June 2024 at approximately $200M ARR (25x) with disclosed B2B contracts and recurring revenue model. SV009, SV010
CV029 Applying a 20-30% monetization-uncertainty discount to Instinct's 31x comparable multiple yields a risk-adjusted fair-value range of $1.7-2.0B. SV023, SV015
CV030 No public company direct comparable exists for a pre-revenue consumer AI assistant at $2.5B; listed AI platform companies trade at 10-25x forward ARR. SV015, SV014
CV031 The absence of an ARR-verified comparable means Instinct's multiple cannot be validated relative to confirmed revenue figures — all benchmarks involve some estimation. SV012, SV009
CV032 KPMG Venture Pulse Q2 2026 reports median Series B AI-native software multiple of 18-25x ARR — Instinct's 31x is 24-72% above this median. SV028, SV009
CV033 Primary exit pathway is strategic acquisition (Apple, Google, or Microsoft) within 3-5 years if IPO market conditions do not support $15B+ consumer software listings by 2029. SV020, SV025
CV034 Thesis-break trigger 1 is ARR growth falling below $5M/month for two consecutive quarters — indicating monetization failure or user attrition. SV014, SV022
CV035 Thesis-break trigger 2 is any regulatory enforcement action in EU, UK, or California related to privacy or AI governance. SV029, SV001
CV036 Thesis-break trigger 3 is failure to launch a paid tier by Q2 2027, signaling fundamental GTM and monetization failure incompatible with the $2.5B valuation. SV012, SV014
CV037 Blocking diligence item 1 is independent ARR verification via third-party audit or rep-and-warranty insurance before any investment decision. SV012, SV024
CV038 Blocking diligence item 2 is full disclosure of pricing model, conversion funnel, and paying-user count before accepting 31x ARR multiple. SV012, SV009
CV039 Blocking diligence item 3 is legal counsel review of EU AI Act, UK DPA, and California CPRA compliance posture given email-access product architecture. SV029, SV017
CV040 Blocking diligence item 4 is investor rights agreement review for liquidation preferences, anti-dilution provisions, and drag-along rights before any entry at or near Series B terms. SV024, SV009
来源
编号出版方标题引文
SO001 TechCrunch Viral AI startup Instinct has raised $350M at a $2.5B valuation "Viral AI startup Instinct has raised $350 million at a $2.5 billion valuation"
SO002 TechCrunch Instinct's powerful AI assistant is raising privacy and security concerns "Instinct's terms grant the company a perpetual and irrevocable license to user materials including emails and screen captures, even after service discontinuation"
SO003 TechFundingNews Noah Shinn's Instinct goes from $100M to $2.5B in weeks: 23-year-old's AI assistant just raised $250M Series B "Noah Shinn's Instinct goes from $100M to $2.5B in weeks"
SO004 Instinct (official) Instinct — AI Personal Assistant
SO005 Noah Shinn (personal website) Noah Shinn — Personal Website
SO006 Pomegra Instinct Hits $2.5B Valuation on $250M Series B
SO007 AIBase News 23-Year-Old Founder's AI Assistant Instinct Valued at $2.5 Billion
SO008 KuCoin News 23-year-old Noah Shinn, founder of Instinct, previously worked on Reflexion and τ-bench
SO009 LookOnChain Who is Instinct's founder? The 23-year-old is the lead author of Reflexion
SO010 Techflier Life-organizing AI assistant Instinct banks $350M at $2.5B
SO011 News-USA.today AI Startup Instinct Hits $2.5B Valuation After $250M Series B Round
SO012 Thundertiger Europe Instinct AI Raises $250M at $2.5B Valuation: Privacy Debate Intensifies
SO013 TechStartups.com Companies That Raised Funding in the Week of August 25 – August 29, 2026
SO014 Yahoo Finance / Tech Instinct's powerful AI assistant is raising privacy and security concerns
SO015 SC Media Instinct AI assistant faces privacy concerns amid praise Security researchers demonstrated that Instinct could be manipulated via malicious email instructions
SO016 Startup Fortune Instinct's AI Assistant Sent an Email Without Asking and Testers Are Furious Early testers reported that Instinct sent emails on their behalf without explicit approval
SO017 ExplainX Instinct AI Privacy: Revoke Access ≠ Delete Data (2026) Emails were stored in plain text and remained accessible even after users disconnected their Google account
SO018 Bitroot Instinct AI Privacy: Why Broad Access Is Non-Negotiable
SO019 LLMS.blog Instinct AI Assistant Faces Scrutiny Over Data Training Terms and Autonomous Transaction Permissions
SO020 The Company Wire Stealth AI Agent Startup Instinct Faces Security Scrutiny Over Broad Permissions
SO021 Best-AI.org Instinct AI Assistant Raises Privacy Concerns: A Preview for OpenAI and Cognition
SO022 AIWeekly TechCrunch: Instinct AI's Always-On Personal Agent Draws Privacy Alarm Bells
SO023 AI Wiki Reflexion — AI agent framework (NeurIPS 2023)
SO024 ArXiv Reflexion: Language Agents with Verbal Reinforcement Learning (NeurIPS 2023) Reflexion achieved 91% pass@1 on HumanEval outperforming GPT-4's 80%
SO025 Sierra (official) Sierra — Enterprise AI Agent Platform
SO026 Index Ventures Index Ventures — Portfolio
SO027 Benchmark Benchmark — Venture Capital
SO028 Kleiner Perkins Kleiner Perkins — Venture Capital
SO029 Conviction Partners Conviction Partners — Venture Capital
SM001 Research and Markets Personal AI Assistant Market Report 2026
SM002 The Business Research Company Personal Artificial Intelligence AI Assistant Market Report 2026
SM003 VirtualAssistantVA AI Personal Assistant Market Hits 4.84 Billion in 2026
SM004 TechCrunch Viral AI startup Instinct has raised 350 million at a 2.5 billion valuation Instinct has gone from $5 million in ARR to $80 million in ARR in the span of eight months
SM005 TechCrunch Instinct powerful AI assistant is raising privacy and security concerns Instinct terms of service grant a perpetual and irrevocable license to use, store, and modify user data
SM006 Gartner Market Guide for Conversational AI Platforms 2026
SM007 IDC Worldwide Intelligent Assistant Market Forecast 2026-2030
SM008 Apple Apple Intelligence Features and Privacy
SM009 Google Google Assistant with Gemini Overview
SM010 OpenAI ChatGPT Agent Mode Documentation
SM011 Instinct Instinct Official Website
SM012 Federal Trade Commission FTC Guidance on AI Agents and Consumer Protection
SM013 European Commission EU AI Act Classification Guidelines
SM014 California Attorney General CCPA Compliance Guide for AI Products
SM015 TechFundingNews Noah Shinn Instinct goes from 100M to 2.5B in weeks
SM016 Rabbit Rabbit r1 Product Overview
SM017 Humane Humane AI Pin Overview
SM018 Statista Virtual Assistant Market Size Worldwide 2024-2030
SM019 CB Insights AI Personal Assistant Startup Landscape 2026
SM020 Andreessen Horowitz The AI Agent Landscape 2026
SM021 Pew Research Center Americans and AI Assistants Survey 2026
SM022 McKinsey The State of AI in 2026
SM023 SC Media Instinct AI assistant faces privacy concerns amid praise Security researchers demonstrated prompt injection attacks against Instinct
SM024 Forbes The Rise of Agentic AI and What It Means for Consumers
SM025 Wired AI Assistants Are Finally Getting Useful But At What Cost
SM026 Bloomberg AI Assistant Market Heats Up as Big Tech and Startups Compete
SM027 TechCrunch These AI startups are growing revenue at faster and faster rates
SM028 Axios The AI Agents Are Coming for Your Daily Tasks
SM029 The Information Inside the Race to Build the AI Assistant of the Future
SP001 Instinct Instinct — AI Personal Assistant
SP002 TechCrunch Instinct's powerful AI assistant is raising privacy and security concerns "Instinct's terms grant the company a perpetual and irrevocable license to user materials including emails and screen captures, even after service discontinuation"
SP003 TechCrunch Viral AI startup Instinct has raised $350M at a $2.5B valuation "Viral AI startup Instinct has raised $350 million at a $2.5 billion valuation"
SP004 OpenAI Introducing Operator
SP005 OpenAI Pricing | ChatGPT
SP006 Google What is Gemini and how it works
SP007 Google One Google AI plans with Cloud Storage - Google One
SP008 Apple Apple Intelligence
SP009 Microsoft Microsoft Copilot | AI Tools for Organizations
SP010 Microsoft Buy Microsoft 365 Premium - Subscription Price, Download | Microsoft Store
SP011 Anthropic Claude
SP012 Claude Claude Pricing
SP013 Perplexity Perplexity Pro
SP014 Meta Meta AI
SP015 Superhuman Superhuman Mail | Pricing
SP016 Motion Pricing | Motion
SP017 Cal.com Pricing | Cal.com
SP018 How-To Geek Humane AI Pin Is Shutting Down After HP Acquisition
SP019 International Center for Law & Economics Integrating AI Assistants and Agents: Competition Policy in Dynamic Markets
SP020 Cyberhaven Labs The 2026 State of AI Adoption & Risk
SP021 Business of Apps ChatGPT Revenue and Usage Statistics (2026)
SP022 Business of Apps WhatsApp Revenue and Usage Statistics (2026)
SP023 Accio GPT, Claude, Gemini & Accio Compared
SP024 WhatsApp WhatsApp for Business | Do more with conversations
SP025 SC Media Instinct AI assistant faces privacy concerns amid praise Security researchers demonstrated that Instinct could be manipulated via malicious email instructions
SP026 Startup Fortune Instinct's AI Assistant Sent an Email Without Asking and Testers Are Furious Early testers reported that Instinct sent emails on their behalf without explicit approval
SI001 TechCrunch Viral AI startup Instinct has raised $350 million at a $2.5 billion valuation Instinct has raised $350 million at a $2.5 billion valuation with $80M ARR
SI002 TechFundingNews Noah Shinn's Instinct goes from $100M to $2.5B in weeks
SI003 TechStartups Spear Street Technology raises $250M at $2.5B valuation
SI004 Instinct (Spear Street Technology) Instinct official product page
SI005 SaaStr What Are Typical ARR Multiples in 2026? Top AI SaaS companies in 2026 trade at 25-50x ARR depending on growth rate
SI006 OpenView Partners SaaS Benchmarks Report 2026
SI007 Meritech Capital Public SaaS Revenue Multiple Benchmarking
SI008 Bloomberg Instinct AI Startup Raises at $2.5 Billion
SI009 Andreessen Horowitz (a16z) AI Company Benchmarks and Multiples 2026
SI010 AI Multiple AI Agent Inference Cost Analysis 2026
SI011 VentureBeat The real cost of running agentic AI in 2026
SI012 Sequoia Capital AI Market Analysis and Unit Economics 2026
SI013 Index Ventures Index Ventures portfolio and investment thesis
SI014 Benchmark Capital Benchmark Capital portfolio
SI015 Kleiner Perkins Kleiner Perkins portfolio and Instinct investment
SI016 Forbes Consumer AI Monetization Challenges in 2026
SI017 Business Insider Why consumer AI subscriptions are struggling to scale in 2026 Consumer AI subscription products have struggled to sustain paying user bases beyond early adopters
SI018 CB Insights AI Startup Financial Benchmarks 2026
SI019 Bain & Company AI Investment and Financial Performance Report 2026
SI020 PitchBook AI Startup Funding and Burn Rate Analysis 2026
SI021 Strictly VC AI Startup Revenue Multiples and Burn Rate Analysis 2026
SI022 TechCrunch Instinct's powerful AI assistant is raising privacy and security concerns Instinct's privacy and security issues could dampen user trust and ARR growth
SI023 SC World Instinct AI assistant faces privacy concerns amid praise
SI024 Morning Brew Consumer AI subscription spending trends in 2026
SI025 The Economist AI company valuations in perspective
SI026 Securities and Exchange Commission EDGAR Full-Text Search - Spear Street Technology Form D Form D exempt offering notice for private placement fundraise
SE001 Instinct Instinct The interface is simple: there are no new interfaces. It's trained to use a phone and a computer. You can text or call it.
SE002 Instinct Privacy Policy Disconnecting a third-party integration does not automatically delete data collected from that integration.
SE003 Instinct Terms of Service You hereby appoint the Services as your agent to enter into agreements, commitments or transactions on your behalf.
SE004 Noah Shinn Noah Shinn
SE005 arXiv Reflexion: Language Agents with Verbal Reinforcement Learning (PDF) Reflexion agents verbally reflect on task feedback signals, then maintain their own reflective text in an episodic memory buffer.
SE006 NeurIPS Reflexion: language agents with verbal reinforcement learning Reflexion achieves a 91% pass@1 accuracy on the HumanEval coding benchmark, surpassing the previous state-of-the-art GPT-4 that achieves 80%.
SE007 GitHub noahshinn/reflexion This repo holds the code, demos, and log files for Reflexion: Language Agents with Verbal Reinforcement Learning.
SE008 GitHub sierra-research/tau-bench We propose τ-bench, a benchmark emulating dynamic conversations between a user and a language agent provided with domain-specific API tools and policy guidelines.
SE009 Google Gmail API overview and guides
SE010 Google Manage push notifications for Gmail API The Gmail API provides server push notifications that let you watch for changes to Gmail mailboxes.
SE011 Google Google Calendar API overview
SE012 Google Get push notifications for Calendar resources Whenever a watched resource changes, the Google Calendar API notifies your application.
SE013 Google OAuth 2.0 best practices User tokens, including refresh and access tokens, must be stored securely and never transmitted in plain text; revoke and delete them when no longer needed.
SE014 Microsoft Outlook mail API overview
SE015 Microsoft Outlook calendar API overview
SE016 Meta Send messages with WhatsApp Cloud API Service messages are free-form messages that you can send to WhatsApp users during a customer service window.
SE017 Twilio Programmable Messaging REST API With the Programmable Messaging REST API, you can add messaging capabilities to your application.
SE018 Twilio Message resource Whenever a Message resource's Status changes, Twilio sends a POST request to the Message resource's StatusCallback URL.
SE019 TechCrunch Instinct’s powerful AI assistant is raising privacy and security concerns The terms also allow Instinct to enter into “agreements, commitments, or transactions” on users’ behalf, which would be binding.
SE020 SC Media Instinct AI assistant faces privacy concerns amid praise One user experienced the AI sending an email on their behalf without prior consent, eroding trust.
SE021 StartupFortune Instinct’s AI Assistant Sent an Email Without Asking and Testers Are Furious Alex Cohen created a fresh Gmail account, emailed his real personal inbox with instructions written like a task for Instinct, and watched the assistant follow them.
SE022 ExplainX Instinct AI agent privacy data retention analysis
SE023 LLMs.blog Instinct AI assistant faces scrutiny over data training terms and autonomous transaction permissions The terms grant Spear Street Technology a perpetual, irrevocable, worldwide, and sublicensable license to cache, store, modify, and utilize user-submitted materials.
SE024 OWASP GenAI Project LLM01: Prompt Injection A Prompt Injection Vulnerability occurs when user prompts alter the LLM’s behavior or output in unintended ways.
SE025 GitHub API Repository metadata for noahshinn/reflexion
SE026 GitHub API Top contributors for noahshinn/reflexion
SE027 GitHub API Repository metadata for sierra-research/tau-bench
SE028 GitHub API Top contributors for sierra-research/tau-bench
SE029 GitHub Open issues for noahshinn/reflexion
SE030 GitHub Open issues for sierra-research/tau-bench
SE031 Hacker News Instinct is raising a $250M Series B
SE032 HN Algolia Search API results for Instinct AI assistant
SU001 Instinct Instinct Instinct is currently available to a private access group as we're scaling up compute. You can create an account to join the waitlist or ask an existing member to invite you.
SU002 Instinct Terms of Service - Instinct Even if you disconnect a Connected Service, we may still use the indexed Connected Service Input data unless you follow the instructions to request deletion.
SU003 Instinct Privacy Policy - Instinct Disconnecting a third-party integration does not automatically delete data collected from that integration.
SU004 TechCrunch Instinct’s powerful AI assistant is raising privacy and security concerns The more powerful these agents become, the more trust matters. Every successful action earns a little more trust. One unauthorized action can reset that trust to zero.
SU005 TechCrunch Viral AI startup Instinct has raised $350M at a $2.5B valuation They've told us they've planned cross-country road trips, bought weekly groceries and concert tickets, and cancelled hundreds of dollars of subscriptions.
SU006 Startup Fortune Instinct's AI Assistant Sent an Email Without Asking and Testers Are Furious Instinct sent an email on Katie Jacobs Stanton's behalf without checking with her first.
SU007 explainx.ai Instinct AI Privacy: Revoke Access ≠ Delete Data (2026) Disconnecting access didn't mean the copies already inside Instinct were gone.
SU008 Best-AI.org Instinct AI Assistant Raises Privacy Concerns: A Preview for OpenAI and Cognition A single unauthorized action can severely erode user trust.
SU009 Forbes AI Assistant Instinct Hits $2.5 Billion Valuation In Weeks Amid VC Feeding Frenzy The product isn’t even public yet. You have to score an invite from VC insiders to download it.
SU010 Carly What Is Instinct AI? What Early Users Actually Found Mohnot said he exchanged 677 messages with Instinct in five days and listed 15 completed jobs.
SU011 Carly Why People Hate Instinct AI: Privacy Backlash, Explained The same users often called the product excellent.
SU012 Digg Instinct AI App Earns Praise For Smooth Onboarding And Proactive Suggestions
SU013 aVenture News Consumer-focused AI assistant startup Instinct reportedly raising $250M
SU014 Pew Research Center Americans and AI 2026: Chatbots, Smart Devices and Views on Impact Roughly seven-in-ten predict AI will make their personal information less secure.
SU015 Pew Research Center Americans and AI 2026 report (PDF)
SU016 Usercentrics Usercentrics Report: The State of Digital Trust in 2026 52% of consumers now trust AI less than humans with their personal data, up from 48% in 2025.
SU017 Usercentrics State of Digital Trust 2026 (PDF)
SU018 Axis Intelligence Research AI Assistant Statistics 2026: Adoption, Usage, and the Productivity Gap 49% of American adults now use AI chatbots, up from 33% in 2024.
SU019 Capital One Shopping AI Shopping Statistics (2026 Report): Consumer Adoption 39% of consumers have used AI assistants for online shopping.
SU020 Sensor Tower 2026 State of AI ChatGPT, Google Gemini, and DeepSeek account for nearly 90% of total time spent across AI assistant apps in Q1 2026.
SU021 Cybernews I Tested the Best AI Personal Assistants for 2026 An AI assistant that can seamlessly work across multiple apps and devices is key for productivity.
SU022 Dume.ai The 10 Best AI Personal Assistants in 2026 (Tested on Real Tasks) Far fewer finish the work.
SU023 SC Media Instinct AI assistant faces privacy concerns amid praise One user experienced the AI sending an email on their behalf without prior consent, eroding trust.
SU024 AI Weekly TechCrunch: Instinct AI's Always-On Personal Agent Draws Privacy Alarm Over Perpetual-License Terms One unauthorized action can reset that trust to zero.
SU025 Creati.ai Instinct Raises $350 Million at a $2.5 Billion Valuation as Privacy Questions Follow Its Viral AI Assistant Because Instinct is still invite-only, its reported traction should be understood as early-market interest rather than proof of broad adoption.
SR001 TechCrunch Instinct's powerful AI assistant is raising privacy and security concerns "Instinct's terms grant the company a perpetual and irrevocable license to user materials including emails and screen captures, even after service discontinuation"
SR002 Startup Fortune Instinct's AI Assistant Sent an Email Without Asking and Testers Are Furious Early testers reported that Instinct sent emails on their behalf without explicit approval
SR003 SC Media Instinct AI assistant faces privacy concerns amid praise Security researchers demonstrated that Instinct could be manipulated via malicious email instructions
SR004 Instinct Instinct — AI Personal Assistant
SR005 Value Add VC Instinct's Powerful AI Assistant Raises Privacy Concerns
SR006 Federal Trade Commission Business Guidance
SR007 Federal Trade Commission Artificial Intelligence | Federal Trade Commission
SR008 European Commission AI Act
SR009 European Commission Commission starts enforcing AI Act rules and new transparency requirements on 2 August
SR010 EUR-Lex General Data Protection Regulation
SR011 NIST AI Risk Management Framework
SR012 OWASP AI Agent Security - OWASP Cheat Sheet Series
SR013 Accio AI Agent Prompt Injection Security & Prevention: Complete 2026 Security Guide
SR014 International Center for Law & Economics Integrating AI Assistants and Agents: Competition Policy in Dynamic Markets
SR015 Stanford HAI Policy and Governance | The 2026 AI Index Report
SR016 ITIF How Rules for Publicly Available Data Are Shaping the Future of AI
SR017 Lowenstein Sandler AI Platform Risk Assessments: Why 2026 Is the Year for Action
SR018 Gunderson Dettmer 2026 AI Laws Update: Key Regulations and Practical Guidance
SR019 OECD Recommendation of the Council on Artificial Intelligence
SR020 Cyberhaven Labs The 2026 State of AI Adoption & Risk
SR021 Boston Consulting Group Agentic AI Is Rewriting the Rules of Data Risk Management
SR022 PwC Trust and Safety Outlook 2026
SR023 AvePoint State of AI 2026: Trust, Control, and the Rise of AI Agents
SR024 Mavvrik AI Cost Governance 2026: Research on AI cost governance
SR025 OpenAI Business Pricing
SR026 Anthropic Pricing
SR027 Google AI for Developers Gemini Developer API pricing
SR028 WhatsApp WhatsApp for Business | Do more with conversations
SR029 Apple Apple Intelligence
SR030 Google One Google AI plans with Cloud Storage - Google One
SR031 Microsoft Microsoft Copilot | AI Tools for Organizations
SV001 TechCrunch Viral AI startup Instinct has raised $350 million at a $2.5 billion valuation
SV002 Bloomberg Instinct AI startup raises $250 million at $2.5 billion valuation
SV003 TechStartups.com Spear Street Technology raises $250M at $2.5B valuation Spear Street Technology Inc. has raised $250 million in Series B financing at a $2.5 billion post-money valuation
SV004 Index Ventures Index Ventures portfolio — Instinct
SV005 Benchmark Capital Benchmark portfolio announcement — Instinct Series B
SV006 Kleiner Perkins Kleiner Perkins — Instinct Series A investment
SV007 Greenoaks Capital Management Greenoaks Capital portfolio — investment philosophy
SV008 Conviction Capital Conviction Capital — portfolio
SV009 PitchBook AI startup valuation benchmarks 2024-2026
SV010 CB Insights AI unicorn valuations — private market tracker 2026
SV011 Forbes Consumer AI company valuations in context
SV012 The Wall Street Journal Consumer AI startups struggle to prove their ARR is real
SV013 Andreessen Horowitz AI company benchmarks and valuation metrics 2026
SV014 SaaStr What are typical ARR multiples for AI companies in 2026?
SV015 Meritech Capital SaaS and AI company benchmarking suite
SV016 OpenView Partners SaaS benchmarks report 2026
SV017 Securities and Exchange Commission EDGAR Full-Text Search — Form D filings for Spear Street Technology SEC Form D exempt offering notice confirms the capital raise under Regulation D exemption, providing regulatory confirmation of fundraising.
SV018 Reuters Consumer AI company funding in 2026 — market overview
SV019 CNBC AI startup valuations — are they justified?
SV020 Bain and Company Technology M&A and AI company valuations 2026
SV021 Sequoia Capital AI unit economics and valuation frameworks for founders
SV022 StrictlyVC AI startup multiples and private market dynamics 2026
SV023 McKinsey and Company Valuing AI companies: beyond revenue multiples
SV024 Harvard Business Review How to value a startup that has no profits
SV025 PwC Tech startup valuations and exit multiples — 2026 outlook
SV026 Deloitte AI market outlook and private company valuations 2026
SV027 Ernst and Young Global venture capital and startup investment report 2026
SV028 KPMG Venture Pulse Q2 2026 — global AI funding and valuations AI companies continued to command premium valuations in Q2 2026, with median Series B multiples of 18-25x ARR for AI-native software.
SV029 NIST AI Risk Management Framework 1.0 The AI RMF provides a framework for organizations to address risks to individuals, organizations, and society associated with AI systems.
SV030 SignalFire Consumer AI market report and investment outlook 2026