服务,即新的软件
下一家万亿美元公司,会是一家披着服务公司外衣的软件公司。
The next $1T company will be a software company masquerading as a services firm.
每一个做 AI 工具的创始人都在问同一个问题:下一代 Claude 把我的产品变成一个 feature,那怎么办? 他们担心得没错。卖工具,你就是在和模型本身赛跑;但如果你卖的是工作本身,模型每进步一次,你的服务就更快、更便宜、更难被替代。一家公司可能一年花 1 万美元买 QuickBooks,再花 12 万美元雇一个会计来做账。下一个传奇公司,会直接把账做了。
Every founder building an AI tool is asking the same question: what happens when the next version of Claude makes my product a feature? They're right to worry. If you sell the tool, you're in a race against the model. But if you sell the work, every improvement in the model makes your service faster, cheaper, and harder to compete with. A company might spend $10K a year for QuickBooks and $120K on an accountant to close the books. The next legendary company will just close the books.
智能 vs 判断Intelligence vs Judgement
写代码主要靠智能。决定下一步该建什么,靠的是判断。
Writing code is mostly intelligence. Knowing what to build next is judgement.
把需求翻译成代码、测试、调试:规则可能复杂,但终究是规则。判断不一样。它需要经验和品味,是多年实践沉淀出的直觉——决定下一个该做什么 feature、要不要欠技术债、什么时候在没完全准备好的时候就发布。
Translating a spec into code, testing, debugging: the rules are complex but they are rules. Judgement is different. It requires experience and taste, instinct built on years of practice. Deciding which feature to build next, whether to take on tech debt, when to ship before it's ready.
一年前,大多数 Cursor 用户把 AI 当作自动补全。今天,被 agent 启动的任务已经多于被人启动的。软件工程占了所有职业 AI 工具使用量的一半以上,其他所有类别都还是个位数。原因是软件工程主要是智能工作。AI 已经跨过那道门槛——大部分智能工作它能自主完成,把判断留给人。软件工程是第一个到达这个节点的领域,接下来,每一个职业都会到。
A year ago, most Cursor users treated AI as autocomplete. Today, more tasks are started by agents than by humans. Software engineering accounts for over half of all AI tool usage across professions. Every other category is still in single digits. The reason is that software engineering is primarily intelligence work. AI has crossed the threshold where it can do most of the intelligence work autonomously and leave the judgement to humans. Software engineering got there first. It is coming to every single profession.
副驾驶与自动驾驶Copilots and Autopilots
Copilot 卖的是工具。Autopilot 卖的是工作本身。
A copilot sells the tool. An autopilot sells the work.
直到不久前,AI 模型的智能和判断力还在发展中,所以正确的打法是先做 copilot:把 AI 交到专业人士手里,让他们决定用它做什么。Harvey 卖给律所,Rogo 卖给投行。专业人士是客户,工具让他们更高产,产出由他们负责。
Until recently, AI models were still developing intelligence and judgement, so the right approach was to build a copilot first: put AI in the hands of a professional and let them decide what to do with it. Harvey sells to law firms. Rogo sells to investment banks. The professional is the customer, the tool makes them more productive, and they take responsibility for the output.
今天,模型已经足够聪明,在某些品类里,起步点直接做 autopilot 才是最优解。Crosby 卖给需要起草 NDA 的公司,而不是给外部律师。WithCoverage 卖给需要保险的 CFO,而不是给保险经纪人。客户买的是直接的结果。任何行业里,工作预算都远远大于工具预算,而 autopilot 从第一天起就吃到的是工作预算。
Today, the models are intelligent enough that in some categories the best place to start is as an autopilot. Crosby sells to the company that needs an NDA drafted, not to outside counsel. WithCoverage sells to the CFO who needs insurance, not to the broker. The customer is buying the outcome directly. The work budget in any profession dwarfs the tool budget, and autopilots capture the work budget from day one.
一个行业里智能工作的占比越高,autopilot 就越早赢。
The higher the intelligence ratio in any field, the sooner autopilots will win.
趋同The Convergence
今天的判断,会变成明天的智能。 当 AI 系统在自己领域内积累起"什么叫好判断"的私有数据,边界就会推进。Copilot 和 autopilot 最终会趋同。Copilot 向 autopilot 的迁移已经在几个品类里开始了。但起步点很重要——它决定了 autopilot 能在哪里立即拿下客户,开始复利积累那批最终让它也能处理判断的数据。
Today's judgement will become tomorrow's intelligence. As AI systems accumulate proprietary data about what good judgement looks like in their domain, the frontier will shift. Copilots and autopilots will converge. The copilot-to-autopilot transition has already begun in several categories. But the starting position matters because it determines where autopilots can win customers now and begin compounding the data that will eventually let them handle judgement too.
Autopilot 打法:用外包当楔子切入The Autopilot Playbook: Outsourcing as the Wedge
每花在软件上的 1 美元,对应有 6 美元花在服务上。
For every dollar spent on software, six are spent on services.
Autopilot 的 TAM 是一个品类的全部人力开销,内包加外包都算。但正确的起点,是已经存在外包的地方。
The total addressable market for autopilots is all labour spend in a category, insourced and outsourced combined. But the right place to start is where outsourcing already exists.
一个任务已经被外包,意味着三件事:(1) 公司已经接受这件事可以外部完成;(2) 有现成的预算线可以干净替换;(3) 买家已经在按"结果"采购。用 AI-native 服务商替换一份外包合同 = vendor swap。替换人头 = 组织重整。前者顺,后者难。
If a task is already outsourced, it tells you three things. One, the company has accepted that this work can be done externally. Two, there's an existing budget line that can be substituted cleanly. Three, the buyer is already purchasing an outcome. Replacing an outsourcing contract with an AI-native services provider is a vendor swap. Replacing headcount is a reorg.
打法: 从外包的、智能密集的任务起步。打透分销。随着 AI 复利,逐步向内包的、判断密集的工作扩张。外包任务是楔子,内包工作是长期 TAM。
The playbook: companies should start with the outsourced, intelligence-heavy task. Nail distribution. Expand toward the insourced, judgement-heavy work as the AI compounds. The outsourced task is the wedge. The insourced work is the long-term TAM.
Crosby 从 NDA 切入:任务定义清楚,主要是智能工作,大多数公司已经在外包给外部律师。预算存在、scope 清楚、ROI 立刻见效、替换没有摩擦。
Crosby started with NDAs: a well-defined task, primarily intelligence, that most companies already outsource to external counsel. The budget exists, the scope is clear, the ROI is immediate, and the substitution is frictionless.
机会地图Opportunity Map
把每个服务垂直行业画在"智能-判断"和"外包-内包"两个轴上,得到一张带人力 TAM 的优先级地图。下面这份清单是示意性的。
Plotting every services vertical on an intelligence-to-judgement spectrum and outsourced-to-insourced ratio produces a priority map with labour TAM in brackets. The list is illustrative.
保险经纪 · $1400-2000 亿
清单里最大的市场。标准化的商业险种高度结构化:经纪人的增值本质就是在多家保险公司之间比价 + 填表,纯智能工作。分销层极度碎片化,数万家小经纪商各自跑同一套流程,没有任何巨头掌握客户关系。WithCoverage 和 Harper 是值得关注的新玩家。
The largest dollar market on this list. Standard commercial lines are highly standardised: the broker's value-add is essentially shopping across carriers and filling forms, pure intelligence work. The distribution layer is incredibly fragmented, tens of thousands of small brokers each running the same process, so no single incumbent controls the customer relationship. WithCoverage and Harper are interesting newcomers.
会计与审计 · 美国外包 $500-800 亿
美国五年内会计师减少约 34 万人,需求却在增长。75% 的 CPA 接近退休,执照路径漫长,起薪又落后于科技和金融行业。这种结构性短缺正逼着事务所比几乎任何其他职业都快地接受 AI。Rillet 在做 AI-native ERP,目标是自动做账;Basis 起步是会计副驾驶。
The US has lost roughly 340,000 accountants over five years while demand has grown. 75% of CPAs are nearing retirement, the licensing path is long, and starting salaries lag tech and finance. That structural shortage is pushing firms to accept AI faster than almost any other profession. Rillet is building the AI-native ERP that will close the books. Basis started as a copilot for accountants.
医疗收费周期 · 美国外包 $500-800 亿
听到"医疗"大家以为是判断密集型,但收费层几乎是纯智能工作。医疗编码就是把临床记录翻译成约 7 万个标准化 ICD-10 编码。规则复杂,但终究是规则。这一层的外包已经成熟,且按结果计费。Autopilot 只要用更低成本做同样的事就行。Anterior 走得最快。
People hear "healthcare" and assume it's judgement-heavy, but the billing layer is almost pure intelligence. Medical coding is translating clinical notes into ~70,000 standardised ICD-10 codes. The rules are complex but they are rules. The outsourcing is already mature and outcome-based. An autopilot just has to do the same thing at lower cost. Anterior is the furthest along.
保险理赔 · 含 TPA $500-800 亿
保单的另一面。标准险种的理赔流程是按保单条款对损失清单 + 用精算表设定准备金。理赔员队伍在老化,而且没人接班。市场已经大规模外包给独立理赔人和 TPA(Crawford、Sedgwick 这些)。同一个行业,至少有两个独立的 autopilot 机会。Pace 在做理赔 autopilot,Strala 在做 AI-native TPA。
On the other side of the insurance policy, claims adjusting is a separate autopilot surface. Standard-line claims are settled by interpreting policy language against damage schedules and setting reserves using actuarial tables. The adjuster workforce is aging out and nobody's replacing them. The market is massively outsourced to independents and TPAs like Crawford and Sedgwick. One industry, at least two distinct autopilot opportunities. Pace is building the autopilot for claims handling. Strala is building an AI-native TPA.
税务咨询 · $300-350 亿
CPA 执照构成监管护城河,但底层 80-90% 的工作是智能工作。Autopilot 每多覆盖一个司法辖区,数据护城河就深一层。多辖区复杂度恰恰就是 SMB 选择外包的原因——没有任何一个内部会计能全覆盖。TaxGPT 是早期玩家,欧洲有 Skalar 和 Ravical。
CPA licensing creates a regulatory moat, but 80-90% of the underlying work is intelligence. Every additional jurisdiction a tax autopilot handles deepens its data moat. Multi-jurisdiction complexity is exactly what SMBs outsource because no single in-house accountant can cover it. TaxGPT is an early mover alongside Skalar and Ravical in Europe.
法务交易类工作 · $200-250 亿
合同起草、NDA、合规备案:高智能、常规外包。产出标准化到质量可验证的程度,买家不需要法律专业就能信任 AI 的产出。Harvey 是新晋领头羊,正快速向 autopilot 转型;Crosby 和 Lawhive 是 autopilot-native 的新玩家。
Contract drafting, NDAs, regulatory filings: high intelligence, routinely outsourced. The work product is standardised enough that quality is verifiable, so the buyer can trust AI output without deep legal expertise. Harvey is the emerging leader and is moving quickly to autopilot; Crosby and Lawhive are the autopilot-native newcomers.
IT 托管服务 · $1000 亿+
每一家 SMB 都把 IT 外包出去。打补丁、监控、账号开通、告警分流:在数千个完全相同的环境上重复运行的智能工作。现有软件层(ConnectWise、Datto)是卖工具给 MSP,还没有人直接把"你的 IT 在跑"作为结果卖给公司。 Edra 在自动化 IT 流程,Serval 在自动化 IT 支持。
Every SMB outsources its IT. Patching, monitoring, user provisioning, alert triage: intelligence work running on repeat across thousands of identical environments. The existing software layer (ConnectWise, Datto) sells tools to the MSP. Nobody has yet sold "your IT runs" directly to the company as an outcome. Edra is automating IT processes. Serval is automating IT support.
供应链与采购 · $2000 亿+
大多数企业只对前 20% 的供应商认真谈判。长尾完全无人问津,因为雇人做不划算。合同泄漏占采购总支出的 2-5%。楔子是无人接手的工作——不用论证预算线、不用挤掉既有玩家,纯粹是捡钱。Magentic 在做直接采购的 AI,AskLio 做间接采购。Tacto 同时做中端市场的记录系统和 copilot。
Most enterprises negotiate seriously with only their top 20% of suppliers. The long tail gets zero attention because it's not economical to have humans do the work. Contract leakage runs 2-5% of total procurement spend. The wedge is abandoned work: no budget line to justify, no incumbent to displace, just found money. Magentic is building the AI for direct procurement, AskLio for indirect procurement. Tacto is building both the system of record and copilot for the midmarket.
招聘与人力派遣 · $2000 亿+
清单里最大的服务市场。招聘漏斗的顶部(筛选、匹配、外联)是纯智能工作,但说服候选人入职、评估文化契合是基于多年模式识别的判断。Autopilot 的楔子在高量、低判断的岗位——匹配标准化的那部分。 Juicebox、Mercor、Jack & Jill 是横跨全光谱的新晋领头羊。
The largest services market on this list. The top of the hiring funnel (screening, matching, outreach) is pure intelligence, but closing a candidate and assessing culture fit is judgement built on years of pattern recognition. The autopilot wedge exists in high-volume, low-judgement roles where matching is standardised. Juicebox, Mercor, Jack & Jill are emerging leaders building across the spectrum.
管理咨询 · $3000-4000 亿
巨大市场,但工作主要是判断。有意思的问题是:AI 能不能把咨询拆解成智能组件(数据收集、对标分析)和判断组件(战略建议),智能层自动化、判断层留给人? 最佳候选玩家:待定。
Huge market but the work is mostly judgement. The interesting question is whether AI can disaggregate consulting into intelligence components (data gathering, benchmarking) and judgement components (strategic recommendations), with the intelligence layer getting automated and the judgement layer staying human. Best candidates TBD.
收束Closing
2025 年,跑得最快的 AI 公司都是 copilot。2026 年,很多会试图变成 autopilot。他们有产品、有客户认知。但他们也面临创新者窘境:卖工作本身,意味着把自己原来的客户(那些做这个工作的人)给切掉了。这正是纯血 autopilot 的机会窗口。
In 2025, the fastest-growing AI companies were copilots. In 2026, many will try to become autopilots. They have the product and the customer knowledge. But they also face the innovator's dilemma: selling the work means cutting their own customers out of doing it. That's the opening for pure-play autopilots.
如果你在做这样的公司,欢迎联系。julien@sequoiacap.com / @julienbek
If you're building one, reach out. julien@sequoiacap.com / @julienbek