SaaS 已死。是智能体（Agents）杀死了它。

上周一，Anthropic 发布了一组用于 Claude Cowork 的插件。不是新模型。不是聊天机器人升级。而是插件。

在 24 小时内，软件类股票蒸发了 2850 亿美元 的市值。一次插件市场的发布公告，在一天之内抹掉的财富，比大多数行业一年创造的还多。

华尔街已经不再害怕 AI 本身了。他们害怕的是——AI 正在取代什么。

II.

大多数人对这件事的理解都错了。

他们听到 “SaaS 已死”，就以为企业不会再买软件了。
事实完全不是这样，差得很远。

真正正在消亡的是一种非常特定的软件商业模式。如果你搞清楚是哪一种，你看到的将是过去十年里最大的创业机会。

让我解释。

III.

过去 15 年，SaaS 的打法非常简单：

* 找到一个常见的业务流程
* 围绕它做一个好看的 UI
* 加点集成
* 按人头、按月收费
* 用切换成本和小幅产品优化来防守阵地

这套打法造就了数百位亿万富翁。

但它有一个没人愿意讲的致命缺陷：大多数价值，从来不在软件本身，而在软件所组织的工作流程里。

UI 只是中间商。

而 AI，刚刚让这个中间商变得多余。

IV.

下面才是 Anthropic 真正做的事情——而头条完全没抓住重点。

他们没有造一个更好的聊天机器人。他们把 Claude 变成了一个工作执行层（work execution layer）。

Cowork 插件让 AI 智能体可以登录你现有的工具系统——你的 CRM、你的文档、你的数据库——并自主完成整个工作流程：

* 法律审计
* 销售管道管理
* 数据分析
* 生产级代码
* 没有人类参与。

这才是让市场恐慌的地方。

因为如果一个 AI 智能体能直接在你现有系统里完成工作——那你还要 15 个带着漂亮仪表盘的 SaaS 工具干什么？

而真正让 SaaS 创始人夜不能寐的是这一点：如果 10 个 AI 智能体能完成 100 名员工的工作，你就不再需要 100 个 Salesforce 账号了。

AI 不是直接杀死软件。它杀死的是使用软件的人头数。而这会杀死按席位收费的商业模式。进而杀死整个业务。

V.

这就是我所说的 “薄中层挤压（Thin Middle Squeeze）”。

想象三层结构：

* 顶层 —— AI 智能体：真正执行工作的东西
* 中层 —— SaaS UI：仪表盘、流程、按钮
* 底层 —— 记录系统（Systems of Record）：数据库、CRM、ERP

现在，价值正在向上被吸进智能体层，向下沉入数据层。

中间那层会被彻底压扁。

这就是为什么：

Adobe 的远期市盈率从 30 跌到 12

ServiceNow 从 67 跌到 28

不是因为人们不再需要它们的功能，而是因为投资者意识到：“漂亮 UI + 集成”的护城河，在 AI 能绕过 UI 的情况下，薄得像纸。

接口曾经就是产品。现在，它只是一个空壳。

VI.

但“SaaS 已死”这派人，错得非常彻底。

SaaS 没死。死的是“容易的 SaaS 护城河”。

这是巨大的区别。

今年企业在软件上的支出将创历史新高。仅企业级 AI 的资本支出，到 2026 年就将超过 4700 亿美元。

这不是一个萎缩的市场。这是一个正在爆炸式增长的市场。

钱没有消失。它只是换了位置。

而大多数人忙着为“毁灭”恐慌，完全看不到钱正落向哪里。

VII.

钱真正流向了这些地方：

1️⃣ AI 平台订阅
按使用量、按消耗计费，而不是按人头。
企业会像买云算力一样买 AI 能力。
GitHub 的 AI 智能体已经采用了这种模式。

2️⃣ 记录系统（Systems of Record）
智能体不会消灭后端，它们依赖后端。
CRM、ERP、数据仓库会变得更值钱。
掌握权威数据的公司会赢。

3️⃣ 安全、治理与合规
当智能体规模化行动时，错误也会规模化发生。
权限、审计日志、策略执行、监控、评估——
这是无聊的基础设施，但未来十年会悄悄印钞。

4️⃣ 按结果定价
不再是“$99/人/月”，而是：

* 每份合同 $5
* 每个工单 $2
* 每条合格线索 $10

软件开始像劳动力一样定价，因为它在替代劳动力。

5️⃣ 服务
软件变得容易造，但在真实企业里跑起来仍然极难。
实施、定制、迁移、集成服务的需求即将爆炸。

VIII.

如果你现在在做创业，唯一重要的问题是：

你站在技术栈的哪一层？

如果你在“薄中层”——

* 漂亮 UI
* 用的是别人的数据
* 按人头收费
* 没有专有优势
* 你会遇到一个正在实时崩塌的经济问题。

上层的智能体在吞噬你的界面。
下层的记录系统在吞噬你的锁定性。
你正被两头挤压。

而且只会越来越快。

IX.

那应该做什么？

做智能体层：
不展示信息，直接执行工作。
按结果收费，而不是席位。

做数据层：
掌握专有数据，成为权威后端。
智能体会更换，数据不会。

做基础设施：
安全、监控、评估、治理、合规。
不性感，但极其赚钱。

做服务：
帮企业把 AI 真正落地到业务中。
未来 5 年大量价值会在这里产生。

X.

最讽刺的一点是：所谓“杀死 SaaS”的 Cowork，本身就是一个 SaaS 产品。订阅制。卖给组织。跑在互联网上。

SaaS 作为交付方式没问题。
有问题的是：靠浅护城河 + 按人头收费的 SaaS 战略。

XI.

所有人都在看这 2850 亿美元的蒸发，看到的是毁灭。
我看到的是价值迁移。

价值没有消失。它从“人和工具之间的中间商”，转移到了执行、数据和基础设施。

旧剧本是：做流程 UI → 按人头收费 → 客户扩招你增长

新剧本是：拥有数据 / 执行结果 / 保障系统，按价值收费，而不是按座位

如果你是创始人，最糟糕的事是恐慌。第二糟糕的是：还在按 2019 年的方式造产品。

最好的事是：看清价值往哪儿流，然后站到它落下的地方。

SaaS 时代没有结束。容易的 SaaS 时代结束了。

说实话？这是过去十年里，对真正做事的人来说，最好的消息。

我在 14 个月内，把 AI 初创公司 Vectal 从 0 做到 180 万美元退出。现在我帮助其他创始人做同样的事。

SaaS is dead. Agents killed it.

Last Monday, Anthropic released a set of plugins for Claude Cowork.
Not a new model. Not a chatbot upgrade. Plugins.
Within 24 hours, software stocks lost $285 billion in market value.
A plugin marketplace announcement erased more wealth in a single day than most industries generate in a year.
Wall Street isn't scared of AI anymore.
They're scared of what AI replaces.

II.
Here's what most people are getting wrong about this.
They hear "SaaS is dead" and think it means companies will stop buying software.
That's not what's happening. Not even close.
What's dying is a very specific type of software business — and if you understand which type, you're looking at the biggest startup opportunity in a decade.
Let me explain.

III.
For the last 15 years, the SaaS playbook was simple:
Find a common business workflow. Build a nice UI around it. Add some integrations. Charge per seat, per month. Defend your position with switching costs and minor product tweaks.
This playbook minted hundreds of billionaires.
But it had a fatal flaw nobody talked about.
Most of the value was never in the software itself. It was in the workflow the software organized.
The UI was a middleman.
And AI just made the middleman obsolete.

IV.
Here's what Anthropic actually did — because the headlines are missing the point.
They didn't build a better chatbot. They turned Claude into a work execution layer.
Cowork plugins let AI agents log into your existing tools — your CRM, your documents, your databases — and perform entire workflows autonomously. Legal audits. Sales pipeline management. Data analysis. Production-grade code.
No human in the loop.
This is the part that spooked the market.
Because if an AI agent can do the work directly inside your existing systems — why do you need 15 different SaaS tools with pretty dashboards sitting on top?
And here's the part that should really keep SaaS founders up at night:
If 10 AI agents can do the work of 100 employees, you don't need 100 Salesforce seats anymore.
AI doesn't kill the software directly. It kills the headcount that uses the software. Which kills the per-seat revenue model. Which kills the business.

V.
This is what I call the Thin Middle Squeeze.
Picture three layers:
Top layer — the AI agent. The thing that actually executes the work.
Middle layer — the SaaS UI. The dashboards, the workflows, the buttons you click.
Bottom layer — systems of record. The databases, CRMs, and ERPs that store the real data.
Right now, value is getting sucked upward into the agent layer and downward into the data layer.
Everything in the thin middle gets crushed.
That's why Adobe's forward P/E dropped from 30 to 12. ServiceNow went from 67 to 28. Not because people don't need what they do — but because investors realized the moat around "nice UI + integrations" is paper-thin when an AI agent can bypass the UI entirely.
The interface used to be the product. Now it's just a shell.

VI.
But here's where the "SaaS is dead" crowd gets it completely wrong.
SaaS isn't dead. The easy SaaS moat is dead.
Massive difference.
Companies will spend more on software this year than ever before. Enterprise AI capital expenditure alone will exceed $470 billion in 2026. That's not a shrinking market — that's a market exploding in size.
The money isn't disappearing. It's moving.
And most people are so busy panicking about the destruction that they're completely blind to where it's landing.

VII.
Here's where the money actually goes:
1 — AI platform subscriptions.
Usage-based. Consumption-based. Not per-seat. Companies will pay for AI capacity the way they pay for cloud compute — based on what they use, not how many humans sit in the building. This is already happening. GitHub's AI agents are gated by premium tiers with usage-based pricing. That's the template.
2 — Systems of record.
Agents don't eliminate backends. They operate them. The CRM, the ERP, the data warehouse — these get more valuable, not less. Because AI agents need clean, authoritative, trusted data to act on. Bad data in, bad actions out. At scale. The companies that own the canonical data win.
3 — Security, governance, and compliance.
When agents act at scale, mistakes happen at scale. Every company deploying AI agents will pay for permissioning, audit logs, policy enforcement, monitoring, and evaluation. This is boring infrastructure — and it will print money quietly for the next decade.
4 — Outcome-based pricing.
Instead of "$99/seat/month," you'll see "$5 per contract reviewed." "$2 per support ticket resolved." "$10 per qualified lead enriched." Software priced like labor — because it's replacing labor. This is where the entire pricing model of the industry shifts.
5 — Services.
This one surprises people. But when it becomes cheap and easy to build software, companies don't build less of it — they attempt more customization. Implementation, workflow design, migration, integration work — the demand for services is about to explode. Vibe coding makes creation easy. Making it work inside a real business is a different story entirely.

VIII.
So if you're building a startup right now — or thinking about it — here's the only question that matters:
Where do you sit in the stack?
If you're building in the thin middle — a nice UI over someone else's data, charging per seat, with no proprietary advantage — you have a serious problem. Not because your product isn't good. But because the economics of your position are collapsing in real time.
The agent layer above you is eating your interface.
The systems of record below you are eating your lock-in.
You're getting squeezed from both directions. And that squeeze only accelerates from here.

IX.
Here's what to build instead.
Build at the agent layer. Create AI-native tools that don't just display information — they execute workflows. Don't show the user a dashboard. Do the work for them. Price on outcomes, not seats. Be the thing that acts.
Build at the data layer. Own proprietary data. Build the system of record for a domain that doesn't have a good one yet. Make yourself the authoritative backend that every AI agent needs to plug into. Agents come and go — the data layer is forever.
Build the infrastructure. Security. Monitoring. Evaluation. Governance. Compliance. The tooling that makes AI agents safe to deploy at scale. Unsexy. Extremely profitable. And the demand hasn't even started yet.
Build the services. Help companies implement, customize, and operate AI systems inside their actual business. This is where most of the real-world complexity lives, and it's where a massive amount of value will be created in the next 5 years.

X.
Here's the irony that nobody is talking about.
Anthropic's Cowork — the product that supposedly killed SaaS — is itself a SaaS product. Sold via subscription. To organizations. On the internet.
SaaS as a delivery model is fine. It was always fine.
SaaS as a business strategy built on shallow moats and per-seat pricing for commodity workflows — that's what's over.

XI.
Everyone is looking at this $285 billion wipeout and seeing destruction.
I see a transfer.
That value didn't vanish. It's moving — from companies that captured value by being the middleman between humans and their tools, to companies that capture value through execution, data, and infrastructure.
The old playbook was: build a workflow UI, charge per seat, grow headcount at your customers to grow revenue.
The new playbook is: build something that owns the data, executes the outcome, or secures the system. Price on value delivered, not butts in seats.
If you're a founder reading this, the worst thing you can do is panic.
The second worst thing is keep building like it's 2019.
The best thing you can do is understand where the value is moving — and go stand where it's landing.
The SaaS era isn't ending.
The easy SaaS era is.
And honestly? That's the best news in a decade for anyone who's actually building something real.
i built my AI startup, Vectal, from zero to a $1.8M exit in 14 months. now i help other founders do the same.