---
title: "Personal agents split into two markets: productivity sells ROI, secretary sells time"
title_zh: "个人 agent 的两种生意：卖生产力，还是卖时间"
url: https://www.liamding.com/posts/2026-10-12-personal-agents-productivity-vs-secretary/
markdown_url: https://www.liamding.com/posts/2026-10-12-personal-agents-productivity-vs-secretary/index.md
html_url: https://www.liamding.com/posts/2026-10-12-personal-agents-productivity-vs-secretary/
language: en
date: 2026-10-12
lastmod: 2026-10-12
description: "Grok Bot and Muse get filed as one category. They are two businesses: one priced against the user's hourly value and sold on proven ROI, the other against the subjective value of free time, settled in transactions."
categories: ["Product Decisions"]
tags: ["AI agents", "personal agents", "product strategy", "pricing", "agent network"]
professional: true
sources:
  - title: "Anthropic engineering: How we contain Claude"
    url: "https://www.anthropic.com/engineering/how-we-contain-claude"
  - title: "Madhu Guru: consumers do not have one relationship with getting things done"
    url: "https://x.com/realmadhuguru/status/2102777931764498536"
  - title: "A16Z data via daily digest: usage broad, paid base tiny, top 1% of payers at $903/month"
    url: "https://aihot.news/items/tfuj58rvo46hvh8l2nzcbpf7n"
  - title: "Simon Willison: metered and agent-driven services should default to hard budget caps"
    url: "https://simonwillison.net/2026/Oct/3/default-hard-budget-caps/"
  - title: "California Attorney General issues investigative subpoena to OpenAI over agent cybersecurity risk"
    url: "https://aihot.news/items/iw7ix94rgvhp2jgamykh261gl"
  - title: "Nikunj Kothari: prompt to sniff the home network for devices left to automate"
    url: "https://x.com/nikunj/status/2106072546206773574"
  - title: "xAI Haggle Bot procurement agent found over $100,000 in direct savings"
    url: "https://aihot.virxact.com/items/cmtnbnfwj02vhrog1dgtzoar8"
  - title: "Nan Yu: neither Muse nor Instinct would place a call"
    url: "https://x.com/thenanyu/status/2101383106372805103"
  - title: "Meta ships Muse for Mac: personal agent acts on the computer under explicit authorisation"
    url: "https://aihot.news/items/cmu63zp9f09xsrofjmnxvrwdg"
  - title: "Peter Yang on Muse saving him $800+ a year"
    url: "https://x.com/petergyang/status/2101033599319613533"
  - title: "Peter Yang: personal agent race, and a multi-agent stack (Muse, Grok Bot, Claude, ChatGPT)"
    url: "https://x.com/petergyang/status/2101862331345154469"
  - title: "Nikunj Kothari: Codex as daily driver, Instinct and Muse as zero-config consumer agents"
    url: "https://x.com/nikunj/status/2102186665863463199"
author: "Liam Ding（丁洪）"
canonical: https://www.liamding.com/posts/2026-10-12-personal-agents-productivity-vs-secretary/
---

## The problem

"Personal agent" is one word covering two products, and the demos are the reason nobody notices. A chat box, a browser session, a phone call, a finished task — Cue and Grok Bot look like Instinct and Muse, so they get filed as one category racing to one finish line.

Filed together, they generate contradictory conclusions: consumer AI subscriptions are hitting a ceiling, and agents are the next platform shift. Both can be true at once, because one is a per-seat productivity sale and the other is a mass-market chore-removal sale. Different buyers, different pricing anchors, different moats, different failure modes — and the biggest product decisions (what to charge for, what to build first, which constraint to attack) come out differently depending on which one you are actually building.

## What I concluded

**Two value propositions, two anchors.** The productivity type creates more value for the user: the agent is worth what the user's time is worth economically. The buyer is a pro consumer raising their own throughput, the test is ROI, and a subscription is the natural shape of the business. The secretary type hands the user's time back: the agent is worth what free time is subjectively worth to that person. The hour it buys back may be spent on short videos and the payment still holds. The buyer is the ordinary consumer, and what is being purchased is the removal of a chore.

That makes the pricing question different in kind. Productivity asks "what is your hour worth?" — the anchor is a wage or a budget line, so the ceiling is high and the population able to answer is small. Secretary asks "what would you pay to never do this again?" — the anchor is a chore and cheap human supply, so the ceiling is low and the population is everyone.

**The asymmetry is the whole argument.** Productivity agents monetise better per user; secretary agents have far more users. The A16Z figures reported on 7 October show how thin the paid base still is: about half of Americans have used AI, 25% daily, but as of August 2026 only 4.5% hold a personal paid subscription to ChatGPT, Gemini or Claude, and the top 1% of payers spend around $903 a month and account for 19.5% of all spend. That is the shape of a narrow, high-willingness-to-pay market. The secretary type aims at the other 95%, who will not pay $20 a month for it.

**Which is why the secretary type's natural revenue is the transaction, not the subscription.** Brian Armstrong's account of Coinbase's agentic-finance push is the reality check in both directions. Card fees start around 30 cents before any percentage, so payments below a dollar are structurally unprofitable, and 76% of the agent commerce transactions Coinbase observes are under 30 cents. His conclusion is that agents need their own accounts and stablecoin rails. Mine is narrower: a take-rate business in agent commerce has to be machine-frequency volume, not consumer checkout, and a $288 phone-call saving is a service nobody can take a share of. So the secretary type's real revenue shape is probably a subscription for trust and permission plus a take on the transactions that do clear at meaningful size — a harder business to model than either camp usually admits.

**The entry point is the prize, and nobody has won it yet.** If the last internet had apps that let people fetch services themselves, the next one has agents that complete tasks for them, and increasingly agents that coordinate with other agents to do it. That is why the winner of this market is unlikely to be the smartest agent, and more likely to be the platform that becomes the default daily entry point for hundreds of millions of people, with a large agent network behind it. Today's evidence does not show that entry point existing yet. Peter Yang runs Muse for personal, Grok Bot for work, Claude for specific jobs and ChatGPT for the rest; Nikunj Kothari keeps Codex as his daily driver, calls Instinct and Muse the best zero-config consumer agents he has used, and still finds Codex with Computer Use on a Mac unmatched. When the heaviest users run four agents, that is a stack, not an entry point. Collapsing the stack is the prize, and only the secretary type genuinely needs it.

**The receipt is real on the productivity side.** xAI gave Grok Bot access to supplier spend, contracts and usage data and built Haggle Bot, which found more than $100,000 of direct savings: 43 paid seats on one SaaS product inactive for 90 days ($14,220), $85,662 a year of unused SKUs elsewhere, and an office-supplies order negotiated from $14,629 to $6,143. Set that against the secretary type's best consumer number — Peter Yang reporting that Muse saved him more than $800 a year on cable and phone bills, including a Muse-versus-Comcast transcript worth $288 — and you have the same technology selling two different things. One prints an auditable receipt. The other performs a small favour.

## Constraints

- **Who can pay.** The 4.5% paid base, and the top 1% accounting for 19.5% of spend, define the productivity type's reachable market. The secretary type's base is everyone else, at a much lower price.
- **Authority and blast radius.** Before an agent can spend your money, use your accounts and carry the consequences, permissioning has to be a product. Anthropic's containment post frames risk as probability multiplied by blast radius: probability keeps falling with safeguards and training, while blast radius only grows with capability and permissions, and access a lab would have refused a year ago is now routine. Simon Willison's 3 October argument is the consumer-side version — metered, agent-driven services should default to a hard budget cap that cuts off with an error, with an explicit opt-out rather than a warning email.
- **Liability is no longer hypothetical.** California's Attorney General issued an investigative subpoena to OpenAI over AI cybersecurity risk after its agent breached Hugging Face infrastructure, and warned developers about their exposure. Whoever holds the wallet holds the case file.
- **The capability is not generally available.** Nan Yu could not place calls with either Muse or Instinct, because both said they could not do it. That is the distance between a demo and a secretary.
- **Integrations are the work, not the model.** Kothari's home-network prompt sits on top of an agent already handling Gmail, a thermostat, DoorDash and Amazon. Nobody in that stack is waiting for a better model.
- **Cheap human supply sets the ceiling.** A cleaner or an errand runner prices the secretary type's chores long before a model does.
- **Sub-dollar transactions limit the take.** With 76% of observed agent commerce under 30 cents, the take-rate story needs machine-scale volume to work at all.

## The options, and what I traded away

| Option | What it buys | What it costs | Why I picked it, or did not |
| --- | --- | --- | --- |
| Productivity type billed against measured savings | A buyer who already owns budget, ROI a champion can defend internally, revenue this quarter | Value concentrates in the top 1%; sales-heavy; renewal dies the moment the number stops | Picked for near-term revenue. Haggle Bot's shape is right, but n=1, so the measurement has to live in the product, not in a case study |
| Secretary type sold as time handed back | The largest under-served base, 95% of which does not pay for AI today, and the only route to a daily habit | Spend authority, caps and liability have to ship before trust does, and the flagship capability, calls, does not work reliably yet | Picked, conditionally. The bigger market is here, but release one is permissioning, not personality |
| One default entry point on top of an agent network | The habit, the distribution, and a take on everything routed through it | You have to collapse a four-agent stack that power users are content with, and you carry the fulfilment risk | The strategic prize, and the reason the productivity type is not the end state |
| Wait for a frontier lab to own the consumer agent | No distribution cost, and you ride the model curve | You own neither the entry point nor the network, and the secretary type's hard parts — trust, payments, liability — are exactly what labs are slowest at | Not picked. The hard parts are the moat |

## Evidence

- xAI's [Haggle Bot](https://aihot.virxact.com/items/cmtnbnfwj02vhrog1dgtzoar8): over $100,000 in direct savings, $14,220 from 43 inactive seats, $85,662 a year of unused SKUs, and $14,629 negotiated down to $6,143.
- Meta's [Muse for Mac](https://aihot.news/items/cmu63zp9f09xsrofjmnxvrwdg) acts on the computer under explicit authorisation. Peter Yang [reported](https://x.com/petergyang/status/2101033599319613533) that Muse saved him more than $800 a year, transcribed in the 19 September digest, including the Comcast negotiation worth $288. Nan Yu found that [neither Muse nor Instinct would place a call](https://x.com/thenanyu/status/2101383106372805103).
- Peter Yang's [personal agent thread](https://x.com/petergyang/status/2101862331345154469): Muse leading, the argument that an agent living inside a chat app like iMessage makes no sense, and his own four-agent stack of Muse, Grok Bot, Claude and ChatGPT.
- Nikunj Kothari on [Codex as daily driver versus Instinct and Muse as zero-config consumer agents](https://x.com/nikunj/status/2102186665863463199), and his [home-network prompt on top of an agent already running Gmail, a thermostat, DoorDash and Amazon](https://x.com/nikunj/status/2106072546206773574).
- Madhu Guru, [the roofer versus the clothes rack](https://x.com/realmadhuguru/status/2102777931764498536): browsing clothes is entertainment for some people, hiring a roofer is torment for almost everyone. That second buyer is the secretary type's customer.
- A16Z consumer numbers, reported 7 October 2026: [half have used AI, 25% daily, 4.5% paying, top 1% at $903 a month and 19.5% of spend](https://aihot.news/items/tfuj58rvo46hvh8l2nzcbpf7n).
- Brian Armstrong, transcribed in the 21 September 2026 digest: card fees start near 30 cents, and 76% of the agent commerce transactions Coinbase observes fall under 30 cents.
- Anthropic, [How we contain Claude](https://www.anthropic.com/engineering/how-we-contain-claude): probability falls, blast radius grows. Simon Willison, [hard budget caps by default](https://simonwillison.net/2026/Oct/3/default-hard-budget-caps/). California's [investigative subpoena to OpenAI](https://aihot.news/items/iw7ix94rgvhp2jgamykh261gl) over agent cybersecurity risk.

## Where this stops being true

This holds for US consumer and small-team agent products, on the model generation available in October 2026, at today's observable pricing. Six places it weakens. First, the dichotomy dissolves if the productivity agent starts absorbing chores, or the secretary agent's output reaches professional grade — the two types converge from opposite ends. Second, the productivity ROI is a single case: one company, one procurement agent, one dataset, and procurement is the friendliest possible function because the savings are already denominated in dollars, so engineering or legal work would need a different measurement. Third, the demand data is US consumer subscription data from August 2026, and a 4.5% paid share says little about markets where AI arrives bundled by default. Fourth, the claim that the entry point decides the market is not yet evidence-backed: the heaviest users run four agents, and the winning entry point is an argument, not an observation. Fifth, the transaction-share case presumes agents finishing deals at prices worth taxing; if most agent commerce settles below a cent, the take stays flat for years. Sixth, China has to be computed separately: weak consumer subscription habits, cheap human supply for errands and domestic work, and generous free model quotas make the secretary type unlikely to be a subscription business there. If calls start working generally and audit-grade containment ships by default, the constraint moves from trust to distribution, and this argument should be re-run.

## What I would do differently

- **Answer the anchoring question first, out loud.** Am I selling more valuable output, or freer time? It decides the anchor (a wage or a chore), the payer (the user or the organisation the user works inside), and where the moat goes (output quality versus entry, trust and fulfilment).
- **Build the receipt into the product, not into a case study.** Haggle Bot's shape is right, but with n=1 the product has to generate its own auditable number per customer, or renewal dies with the first quiet quarter.
- **For the secretary type, ship the cap and the permissioning as defaults.** Willison's formulation is the correct first release: cut off at a set amount with an error, require an explicit opt-out. What is being sold before anything else is permission to act.
- **Spend on fulfilment and integration, not on the model.** Every concrete secretary-type example is plumbing: calls that still do not work, Gmail, thermostats, DoorDash, Amazon. The scarce asset is the parties who can be dispatched and held accountable.
- **Treat the entry point as the asset to win, and the agent network as a fulfilment network.** "Agent network" is not a protocol diagram. It is the set of parties behind the agent that can be paid, dispatched and held responsible — and the moment a user stops maintaining four agents is the moment this starts compounding.

## 中文版

## 问题

「Personal agent」这一个词，盖住了两种不同的产品，而演示视频正是没人细看的原因。一个对话框、一次浏览器操作、一个电话、一件办完的事——Cue、Grok Bot 看上去和 Instinct、Muse 是同类，于是它们被归成一个赛道、奔着同一条终点线。

混在一起谈，就会得出互相矛盾的结论：一边说消费级 AI 订阅增长见顶，另一边说 agent 是下一个平台级机会。两者可以同时成立，因为其中一个是按席位卖的生产力生意，另一个是面向大众的"把琐事替你办掉"的生意。买家不同、计价锚不同、护城河不同、失败模式也不同——而最重要的产品决策（对什么收费、先做什么、攻哪个约束），取决于你到底在建哪一种。

## 我的结论

**两种价值主张，两个计价锚。** 生产力型帮用户创造更多价值：agent 的价值等于用户时间的经济价值。买家是给自己提效的 Pro Consumer，衡量标准是 ROI，订阅是最自然的商业模式。秘书型帮用户拿回时间：agent 的价值等于自由时间对那个人的主观价值——买回来的那一小时哪怕拿去刷短视频，这次付费依然成立。买家是普通消费者，买到的是"这件事以后不用我做了"。

这使定价问题在性质上就不同。生产力型问的是"你一小时值多少钱"——锚定的是工资或一条预算科目，所以天花板高，但答得出这个问题的人群很小。秘书型问的是"以后都不用你做，你愿意付多少"——锚定的是琐事本身和便宜的人力供给，所以天花板低，但人群是所有人。

**这个不对称本身就是论点。** 生产力型的单位变现更好，秘书型的用户规模大得多。10 月 7 日被解读出的 A16Z 数据说明付费人群仍然很薄：美国约一半人用过 AI，25% 每天用；但截至 2026 年 8 月，只有 4.5% 持有 ChatGPT、Gemini 或 Claude 的个人付费订阅，付费者中的头部 1% 月均约 903 美元，贡献了全部消费的 19.5%。这正是"窄而付费意愿高"的形状。秘书型要打的，是剩下那 95% 不会为此每月付 20 美元的人。

**所以秘书型天然的收入在交易里，不在订阅里。** Brian Armstrong 谈 Coinbase 的 agentic finance 布局，是一份双向的现实检查：借记卡、信用卡的固定手续费约 30 美分起步，还没算百分比抽成，所以低于 1 美元的交易在结构上就不成立——而他们观察到的 agent 电商交易里，高达 76% 金额不足 30 美分。他的结论是 agent 需要自己的账户和稳定币通道。我的结论更窄：在 agent 电商上做抽成，必须是机器频率的走量生意，而不是面向消费者的结账；而一次电话砍价省下 288 美元是服务，没有人能从里面抽走一分钱。所以秘书型真实的收入形状，更可能是"为信任与授权付的订阅"加"对真正成交、金额够大的交易的抽成"——这比两个阵营通常承认的都要难建模。

**入口是奖品，但还没有人拿到。** 如果说上一代互联网是用 App 让人主动获取服务，下一代就是用 agent 替人主动完成任务，而且越来越多地由 agent 之间协作完成。这正是为什么这个市场的赢家大概率不是最聪明的 agent，而是那个成为亿万人日常生活默认入口、并在背后拥有庞大 agent network 的平台。今天的证据还看不到那个入口已经存在：Peter Yang 自己是 Muse 管个人、Grok Bot 管工作、Claude 管特定任务、ChatGPT 管其余；Nikunj Kothari 的主力仍是 Codex，他认为 Instinct 和 Muse 是他用过最好的零配置消费级 agent，同时也说 Mac 上 Codex 加 Computer Use 无人能及。最重度的用户已经在跑四个 agent——那是一个栈，不是一个入口。把这个栈收成一个入口，才是奖品，而真正需要它的只有秘书型。

**生产力型那张收据是真的。** xAI 给 Grok Bot 开放供应商支出、合同与使用数据，搭出 Haggle Bot，找出超过 10 万美元直接节省：某 SaaS 产品 43 个付费席位 90 天无活跃（省 14,220 美元）、另一个产品每年 85,662 美元未使用 SKU、一次办公用品比价把 14,629 美元的订单压到 6,143 美元。把秘书型目前最好看的消费级数字放在旁边——Peter Yang 称 Muse 帮他在有线电视与电话账单上一年省下 800 多美元，其中包括一份与 Comcast 谈判、省下 288 美元的通话记录——同一代技术，在卖两件完全不同的东西：一个打印一张可审计的收据，另一个帮忙做了件小事。

## 约束条件

- **谁付得起。** 上面那 4.5% 的付费人群、以及头部 1% 占 19.5% 消费的集中度，决定了生产力型真正可及的市场规模；秘书型面对的是其余所有人，但单价低得多。
- **授权与爆炸半径。** 在 agent 能花你的钱、用你的账号、替你承担后果之前，"权限"必须先成为一个产品功能。Anthropic 那篇 containment 文章把风险拆成概率乘爆炸半径：概率随防护与训练持续下降，爆炸半径只随能力与权限增长——一年前实验室会拒绝的权限，今天是日常。Simon Willison 在 10 月 3 日给出的消费侧版本是：按量计费的、由 agent 驱动的服务，应默认设置硬性预算上限，超限直接报错切断，想不受限的人主动勾选退出，而不是收到一封警告邮件。
- **法律责任不再是假设。** 加州总检察长就 AI 网络安全风险向 OpenAI 发出调查传票，背景是其 agent 入侵 Hugging Face 基础设施，并警告开发者注意自身风险。谁拿着钱包，谁就拿着卷宗。
- **能力还没普遍可用。** Nan Yu 让 Muse 和 Instinct 打电话，两个都说做不到。这就是演示和秘书之间的距离。
- **集成才是工作量，不是模型。** Kothari 那条嗅探家庭网络的 prompt，是架在一个已经处理 Gmail、恒温器、DoorDash、Amazon 的 agent 之上的。那一栈里没有人在等更强的模型。
- **便宜的人力供给决定了天花板。** 决定秘书型琐事定价的，是小时工和跑腿的人，而不是模型。
- **低于一美元的交易限制了抽成。** 观察到的 agent 电商里 76% 不足 30 美分，抽成这件事必须靠机器规模的走量才成立。

## 取舍选项

| 选项 | 换来什么 | 代价 | 为什么选/不选 |
| --- | --- | --- | --- |
| 按可度量节省计费的生产力型 | 本来就有预算的买家、能对内背书的 ROI、本季度的收入 | 价值集中在头部 1%；销售驱动；数字一停，续费就死 | 为近期收入选。Haggle Bot 的形状是对的，但 n=1，度量必须做进产品，而不是写进案例 |
| 以"把时间还给你"为卖点的秘书型 | 最大的未服务人群（今天 95% 不为 AI 付费），以及唯一通向日常习惯的路 | 花钱权限、上限与责任必须比信任先交付，而旗舰能力（打电话）还不可靠 | 有条件地选。更大的市场在这里，但第一个版本是授权体系，不是人格化 |
| 在 agent network 之上做单一默认入口 | 习惯、分发，以及所有被路由经过它的抽成 | 必须把重度用户已经满意的四 agent 栈收成一个，并承担履约风险 | 战略奖品，也是生产力型不是终局的理由 |
| 等前沿实验室来做消费级 agent | 没有分发成本，跟着模型曲线走 | 入口和网络都不属于你，而秘书型最难的部分——信任、支付、责任——恰好是实验室最慢的地方 | 不选。难的部分才是护城河 |

## 证据

- xAI 的 [Haggle Bot](https://aihot.virxact.com/items/cmtnbnfwj02vhrog1dgtzoar8)：直接节省超 10 万美元，43 个无活跃席位省 14,220 美元，每年 85,662 美元未使用 SKU，14,629 美元压到 6,143 美元。
- Meta 的 [Muse for Mac](https://aihot.news/items/cmu63zp9f09xsrofjmnxvrwdg) 在明确授权下直接操作电脑。Peter Yang [称](https://x.com/petergyang/status/2101033599319613533) Muse 帮他一年省下 800 多美元（据 9 月 19 日 digest 转录），其中包括与 Comcast 谈判省下的 288 美元。Nan Yu 发现 [Muse 与 Instinct 都无法打电话](https://x.com/thenanyu/status/2101383106372805103)。
- Peter Yang 的[个人 agent 之争 thread](https://x.com/petergyang/status/2101862331345154469)：看好 Muse 领跑、认为 agent 住在 iMessage 这类聊天 App 里说不通，以及他自己 Muse／Grok Bot／Claude／ChatGPT 的四 agent 栈。
- Nikunj Kothari：[Codex 是主力，Instinct 与 Muse 是零配置消费级 agent](https://x.com/nikunj/status/2102186665863463199)；他还有一条[家庭网络嗅探的 prompt，架在已经接管 Gmail、恒温器、DoorDash、Amazon 的 agent 上](https://x.com/nikunj/status/2106072546206773574)。
- Madhu Guru：[屋顶工与衣服架](https://x.com/realmadhuguru/status/2102777931764498536)——逛衣服对某些人是娱乐，请屋顶工对几乎所有人是折磨。后者才是秘书型的客户。
- A16Z 消费数据（10 月 7 日被解读）：[一半人用过 AI，25% 每天用，4.5% 付费，头部 1% 月均 903 美元、占消费 19.5%](https://aihot.news/items/tfuj58rvo46hvh8l2nzcbpf7n)。
- Brian Armstrong（据 2026 年 9 月 21 日 digest 转录）：刷卡固定手续费约 30 美分起步，而 Coinbase 观察到的 agent 电商交易里 76% 低于 30 美分。
- Anthropic：[How we contain Claude](https://www.anthropic.com/engineering/how-we-contain-claude)——概率下降、爆炸半径上升。Simon Willison：[默认硬性预算上限](https://simonwillison.net/2026/Oct/3/default-hard-budget-caps/)。加州总检察长就 agent 网络安全风险[向 OpenAI 发出调查传票](https://aihot.news/items/iw7ix94rgvhp2jgamykh261gl)。

## 这个结论的边界

这套判断适用于 2026 年 10 月、美国市场的消费者与小型团队 agent 产品，以及当下可观察到的定价。六处会失效。第一，如果生产力型开始顺手把琐事也做了，或者秘书型的产出到达专业级，这个二分就会从两端合拢、消失。第二，生产力型的 ROI 只有一个案例：一家公司、一个采购 agent、一份数据，而采购恰好是对 agent 最友好的职能，因为节省本来就以美元计价；换成研发或法务，度量方式要重新设计。第三，需求侧是 2026 年 8 月的美国消费者订阅数据，4.5% 的付费率在"AI 默认被打包进套餐"的市场里说明不了什么。第四，"入口决定市场"这一条目前没有证据支撑：最重度的用户跑的是四个 agent，那个赢家入口是论断，不是观察。第五，交易分成的前提是 agent 在完成值得抽成的交易；如果大部分 agent 电商以一美分以下结算，抽成会长期贴地。第六，中国必须单独算：C 端订阅习惯弱、跑腿与家政的人力供给便宜、大模型免费额度充足——秘书型在那里大概率不是一门订阅生意。如果打电话这类能力普遍可用、审计级的 containment 成为默认，约束就会从信任转向分发，这套判断需要重跑一遍。

## 我会怎么改

- **先把计价锚的问题说出来。** 我卖的是更值钱的产出，还是更自由的时间？它决定锚（工资还是琐事）、谁付钱（用户本人还是他所在的组织）、护城河往哪儿挖（产出质量，还是入口、信任与履约）。
- **把收据做进产品，而不是写进案例。** Haggle Bot 的形状是对的，但 n=1；产品必须能为每个客户自己生成那张可审计的数字，否则第一个安静季度就会带来续费问题。
- **秘书型先发上限和授权体系。** Willison 的说法就是正确的第一个版本：到达金额直接报错切断，不受限要显式勾选退出。在此之前，真正在卖的是"可以动我东西"的许可。
- **把钱花在履约与集成上，不花在模型上。** 现有每个具体的秘书型例子里，瓶颈都是管线：还打不通的电话、Gmail、恒温器、DoorDash、Amazon。稀缺资产是那些能被调度、也能被追责的第三方。
- **把入口当成要抢的资产，把 agent network 当成履约网络。** "Agent network" 不是一张协议图，而是 agent 背后那批能被付钱、被调度、被追责的参与方；而当用户不再维护那四个 agent 的那一刻，这件事才开始复利。